Phonological Conditions on Affixation

<strong>Phonological</strong> <strong>Conditions</strong> on Affixation 

by 

Mary Elizabeth Paster 

B.A. (Ohio State University) 2000 

M.A. (University of California, Berkeley) 2002 

A dissertation submitted in partial satisfaction of the 

requirements for the degree of 

Doctor of Philosophy 

in 

Linguistics 

in the 

GRADUATE DIVISION 

of the 

UNIVERSITY OF CALIFORNIA, BERKELEY 

Committee in charge: 

Professor Sharon Inkelas, Co-Chair 

Professor Andrew Garrett, Co-Chair 

Professor Kristin Hanson 

Spring 2006


© 2006 

by 

Mary Elizabeth Paster

Abstract 


by 

Mary Elizabeth Paster 

Doctor of Philosophy in Linguistics 

University of California, Berkeley 

Professor Sharon Inkelas and Professor Andrew Garrett, Co-Chairs 

This dissertation presents results of a cross-linguistic survey of phonologically 

conditioned suppletive allomorphy (PCSA). One hundred thirty-seven examples are 

discussed, representing 67 languages. Three major generalizations emerge from the 

survey. First, PCSA occurs at the same edge of the stem as the trigger: PCSA in prefixes 

is triggered at the left edge of the stem, while PCSA in suffixes is triggered at the right 

edge. Second, PCSA is sensitive to underlying rather than surface forms. This is 

demonstrated in several examples where phonological processes render opaque the 

conditions determining allomorph distribution. Finally, despite its characterization in 

recent literature, PCSA is not always optimizing. In numerous examples, words are no 

more phonologically well-formed than they would be if there were no allomorphy, or if 

the distribution of allomorphs were reversed. 

The generalizations arising in this survey distinguish between competing 

frameworks for modeling phonological conditions on affixation. The first is the ‘P >> M’ 

approach, where PCSA is modeled by ranking <strong>Phonological</strong> constraints over 

Morphological constraints in Optimality Theory. This model predicts that PCSA should 

1

e phonologically optimizing and that allomorphy may be sensitive to phonological 

conditions anywhere in the word. 

In the alternative approach, advocated here, PCSA is modeled by incorporating 

phonological elements of stems into the subcategorization frames of affixes. 

Subcategorization frames specify the type of stem to which affixes will attach, including 

syntactic, morphological, and (crucially) phonological features of stems. The distribution 

of suppletive allomorphs results from different requirements imposed by each allomorph 

on stems. This approach predicts that allomorphy should be sensitive to input rather than 

surface phonological elements and that PCSA should be sensitive only to elements at the 

edge of the stem where the affix attaches. These predictions are upheld in the survey. 

In the introduction and conclusion, I situate these findings in the broader context 

of the literature on the phonology-morphology interface. I also consider predictions of the 

models discussed above for other types of phonological effects in morphology, showing 

that research in those areas converges with the results presented here. 

2

Table of contents 


Chapter 1. Introduction ........................................................................................... 1 

1.1 <strong>Phonological</strong> conditions on affixation ............................................................... 3 

1.1.1 The P >> M model ....................................................................................... 6 

1.1.1.1 Description of the model ....................................................................... 7 

1.1.1.2 Predictions for phonologically conditioned suppletive allomorphy ...... 8 

1.1.2 The subcategorization model ....................................................................... 11 

1.1.2.1 Description of the model ....................................................................... 11 

1.1.2.2 Predictions for phonologically conditioned suppletive allomorphy ...... 12 

1.2 <strong>Phonological</strong>ly conditioned suppletive allomorphy .......................................... 16 

1.2.1 Previous discussions .................................................................................... 17 

1.2.2 Survey methodology .................................................................................... 22 

Chapter 2. Segmentally conditioned suppletive allomorphy ................................ 25 

2.1 Survey results ..................................................................................................... 25 

2.1.1 Allomorphy vs. morphophonology .............................................................. 27 

2.1.2 Examples ...................................................................................................... 31 

2.1.2.1 Assimilation/harmony ............................................................................ 31 

2.1.2.2 Dissimilation/disharmony ...................................................................... 39 

2.1.2.3 Syllable structure optimization .............................................................. 53 

2.1.2.4 Syllable contact constraint satisfaction .................................................. 72 

2.1.2.5 Non-optimization ................................................................................... 76 

2.1.2.6 Opaque conditioning .............................................................................. 98 

2.1.3 Summary ...................................................................................................... 103 

2.2 Analysis ............................................................................................................. 104 

2.3 Conclusion ......................................................................................................... 109 

Chapter 3. Tone/stress conditioned suppletive allomorphy .................................. 111 


3.1.1 Examples ...................................................................................................... 113 

3.1.1.1 Stress effects .......................................................................................... 113 

3.1.1.2 Stem allomorphy .................................................................................... 119 

3.1.1.3 Tone effects ........................................................................................... 126 


3.1.2 Summary ...................................................................................................... 135 

3.2 Analysis ............................................................................................................. 136 


Chapter 4. Prosodically conditioned suppletive allomorphy ................................ 140 


4.1.1 Examples ...................................................................................................... 142 

4.1.1.1 Foot parsing ........................................................................................... 144 

4.1.1.2 Minimality ............................................................................................. 159 

i


4.1.2 Summary ...................................................................................................... 172 

4.2 Historical development of prosodically conditioned suppletive allomorphy: 

The case of Pama-Nyungan ergative suffix allomorphy ................................... 174 

4.2.1 Examples ...................................................................................................... 176 

4.2.2 Previous discussions .................................................................................... 196 

4.2.3 A history of PN ergative allomorphy ........................................................... 201 

4.3 Analysis of prosodically conditioned suppletive allomorphy ........................... 205 

4.3.1 The Output Optimization approach ............................................................. 205 

4.3.2 Survey results revisited ................................................................................ 207 

4.3.3 Examples of non-optimizing syllable-counting allomorphy (SCA) ............ 208 

4.3.4 Examples of SCA possibly resulting from TETU ....................................... 211 

4.3.5 The Subcategorization approach .................................................................. 212 

4.3.6 Using Subcategorization to model SCA....................................................... 214 

4.3.7 Estonian revisited ......................................................................................... 214 

4.3.8 Summary of analysis .................................................................................... 216 


Chapter 5. Predictions of the P >> M model .......................................................... 219 

5.1 Empty morphs .................................................................................................... 221 

5.2 <strong>Phonological</strong>ly induced morphological gaps ..................................................... 223 

5.3 Reduplication ..................................................................................................... 225 

5.4 <strong>Phonological</strong>ly conditioned affix order ............................................................. 227 

5.4.1 Predictions for affix order ............................................................................ 227 

5.4.2 Cross-linguistic findings .............................................................................. 229 

5.4.3 Pulaar ........................................................................................................... 232 

5.4.3.1 Sonority-based affix order? .................................................................... 232 

5.4.3.1.1 Gombe Fula affix order .................................................................... 233 

5.4.3.1.2 A P >> M account of Gombe Fula affix order ................................. 237 

5.4.3.1.3 Fuuta Tooro Pulaar (Northeastern Senegal) ..................................... 239 

5.4.3.1.4 A scope-template account of Fuuta Tooro affix order ..................... 245 

5.4.3.1.5 A scope-based reanalysis of Gombe Fula ........................................ 248 

5.5 Infixation ............................................................................................................ 251 


Chapter 6. Conclusion .............................................................................................. 255 

6.1 Summary of predictions ..................................................................................... 256 

6.2 Summary of survey results ................................................................................ 258 

6.3 Summary of theoretical implications ................................................................. 260 

6.4 Two types of PCSA? ............ ............................................................................. 264 

6.5 M >> P vs. separate components ........................................................................ 269 

6.6 Future directions ................................................................................................ 271 

References .................................................................................................................. 274 

Appendix: Surveyed languages ............................................................................... 290 

ii

List of tables and figures 

Figure 1: Map showing location of examples of ergative allomorphy ....................... 177 

Figure 2: Map showing distribution of conditions on ergative allomorphy ............... 195 

Table 1: Examples of ergative allomorphy ................................................................. 193 

Table 2: Examples of locative allomorphy ................................................................. 194 

Table 3: Hale’s (1976b) proposed history of PN ergative/locative allomorphy ......... 199 

iii

Acknowledgements 

I have many people to thank for their various contributions to my work. My 

studies in linguistics began almost by accident when I was an undergraduate at Ohio 

State. I had changed my major several times, but was hooked by an excellent intro 

linguistics course with Beth Hume, whose enthusiasm for the subject rubbed off on me 

from the start. Dave Odden’s phonology course sealed the deal. Dave taught me how to 

do fieldwork and how to do phonology. He is one of the people I admire and respect 

most, and I am deeply appreciative to him for being a great role model. 

I would also like to thank my advisors, Sharon Inkelas and Andrew Garrett, for 

their guidance and support. I cannot thank them enough. Sharon has been a wonderful 

mentor and friend. Invariably, when I have felt at all confused or discouraged about 

something I was working on, I have always emerged from meetings with Sharon feeling 

infinitely better, with clearer thinking and better focus than before. Andrew has also been 

an excellent advisor, and in particular I am extremely grateful to him for helping me 

through a period of uncertainty early in my graduate career with encouragement and 

understanding. Andrew is someone whose standards of intellectual excellence are 

sometimes intimidating but always inspiring. 

I am also very grateful to Larry Hyman, who is always interesting to talk to and 

has influenced my work in many ways. Larry’s obvious passion for his work has been an 

inspiration at times when I have felt bogged down in my studies, and I have appreciated 

this very much. There are several other faculty members who have been especially 

helpful and influential during my time at Berkeley, including Juliette Blevins, John 

Ohala, and Ian Maddieson; in general, all of the linguistics faculty at Berkeley have been 

iv

very supportive. In addition, the linguistics staff members have been tremendously 

helpful, especially Belen Flores, to whom I am very grateful, especially for dealing with 

the hassles that have gone along with my finishing from out of town. I have really 

appreciated being part of such a friendly and intellectually engaging department. 

While at Berkeley, I have had great friends and colleagues to whom I am thankful 

for their support and for all the good times. Julie Larson and Suzanne Wilhite have 

always been fun and wonderfully supportive for me in work and life. They and the other 

members of our cohort made all of my classes pleasant and/or interesting. In particular, 

Abby Wright has been a loyal and thoughtful friend, and has continued to take an interest 

in my work even after she left the field. I am also grateful to the many other Berkeley 

students who have been my close friends and colleagues, including (but not limited to, 

and in no particular order) Alan Yu, Lisa Conathan, Jeff Good, Rosemary Beam de 

Azcona, Yuni Kim, Teresa McFarland, Lisa Bennett, and Anne Pycha. Thank you for 

inspiring and entertaining me. I am also grateful to my partner, Sean Spellman, who has 

been patient and supportive throughout the latter stages of my graduate career. 

The last portion of my dissertation writing has been conducted while I have been 

teaching at the University of Pittsburgh. I would like to thank the linguistics faculty, staff, 

and students at Pitt for making this time enjoyable and interesting for me. 

Finally, I would like to thank my parents for their support of my career choice and 

my decision to go away for graduate school. They have always given me whatever I 

needed to be successful, in the form of emotional support, encouragement, and (of 

course) money. I am truly lucky to have the benefit of their love and support, and I 

dedicate my dissertation to them. 

v

Chapter 1: Introduction 

This dissertation concerns phonological conditions on the morphological process 

of affixation, focusing on cases of phonologically conditioned suppletive allomorphy 

(PCSA). An example of PCSA is found in Dja:bugay (Patz 1991), a Pama-Nyungan 

language of Australia. As seen in the examples in (1), the genitive suffix in Dja:bugay 

has two different forms, -n and -ŋun (Patz 1991: 269). 

(1) a. guludu-n ‘dove-GEN’ b. girrgirr-ŋun ‘bush canary-GEN’ 

gurra:-n ‘dog-GEN’ gaɲal-ŋun ‘goanna-GEN’ 

djama-n ‘snake-GEN’ bibuy-ŋun ‘child-GEN’ 

When the stem ends in a vowel, as in (1)a, the -n suffix is used. When the stem ends in a 

consonant, as in (1)b, the -ŋun suffix is used. What distinguishes this as suppletive 

allomorphy is that it is not obvious how any plausible phonological process could relate 

the two allomorphs to a single underlying form, since this would involve the 

simultaneous deletion or insertion (depending on the analysis) of two segments, [ŋ] and 

[u]. We therefore conclude that there are two separate underlying forms that can be used 

to mark genitive. However, even though we are not able to write rules or constraints to 

derive the allomorphs from a single underlying form, we can nonetheless state the 

distribution of allomorphs in phonological terms, since the relevant property of the stem 

(consonant- vs. vowel-final) is phonological in nature. 

Consonants vs. vowels at stem edges are found to condition PCSA in many of the 

world’s languages (as will be seen in chapter 2), but this is by no means the only type of 

condition that is found. In this dissertation, I describe results of a cross-linguistic survey 

that attempts to discover the precise range of possible phonological conditions on PCSA, 

and I show how these results bear on the general question of how to model phonological 

1

conditions on affixation. The aims and contributions of this dissertation are twofold: to 

present a typological overview of PCSA, and to show how these typological results point 

us towards a subcategorization-based model of phonological conditions on affixation. 

The dissertation is structured as follows. In the remainder of chapter 1, I provide 

some background on the study of phonological conditions on affixation and of PCSA. I 

describe two competing models that have been advanced to account for PCSA, namely, 

the ‘P >> M’ model and a subcategorization-based model (both to be explained later in 

this chapter). In this chapter I give an overview of their respective predictions for 

phonological conditions on affixation; these predictions are discussed in more detail in 

later chapters. 

Chapters 2-4 present the results of the cross-linguistic survey of PCSA, 

generalizations about the examples and their bearing on the choice between theoretical 

models, and sample analyses of some representative examples of each type. In each 

chapter, it is argued that subcategorization provides a superior analysis that avoids some 

problems of under- and overprediction that are encountered by P >> M. The examples in 

chapter 2 involve conditioning by individual segments or their features. Those in chapter 

3 are conditioned by tone or stress. Finally, the examples in chapter 4 are conditioned by 

prosodic factors such as syllable count, mora count, or foot structure. Chapter 4 also 

features a detailed discussion of the nature and development of a set of examples found in 

the Pama-Nyungan languages of Australia; it is demonstrated that a diachronic 

perspective on PCSA is instructive in assessing the relative merits of the theoretical 

models that are contrasted in this dissertation. 

2

In chapter 5, I situate the findings of the present study within the larger context of 

phonological conditions on affixation. In that chapter, I lay out the range of predictions 

that are made by the P >> M approach for all aspects of affixation, and where possible, I 

compare these predictions with cross-linguistic survey data. As will be seen, the 

predictions generally do not match up well with the survey data. This provides another 

argument against the use of P >> M, and by extension, an argument in favor of the 

alternative, subcategorization-based approach that I advocate. 

Finally, in chapter 6, I conclude by summarizing the typological generalizations 

about PCSA and the implications of the survey results for the two competing models. I 

consider (and ultimately reject) a ‘hybrid’ approach that distinguishes two types of PCSA 

and uses both the subcategorization and P >> M models, and I conclude with some 

further remarks on strategies for modeling phonological conditions on affixation. 

1.1 <strong>Phonological</strong> conditions on affixation 

The study of the phonology-morphology interface has relatively old roots in the 

theoretical literature, particularly in the formal analysis of morphologically conditioned 

phonology (see especially Kiparsky 1982a, b and Mohanan 1986 on Lexical Phonology). 

However, only within the last decade has there been a push towards the study of 

phonological effects in morphology, and more specifically, phonological conditions on 

affixation. This dissertation is intended to contribute to this previously neglected area of 

the literature. 

Subcategorization-based models of affixation have been proposed in various 

forms over the course of many years; see, for example, Lieber 1980, Kiparsky 1982b, 

Selkirk 1982, Orgun 1996, and Yu 2003. Orgun 1996 makes explicit the idea of 

3

incorporating phonological conditions directly into subcategorization frames, though this 

is implicit in earlier versions of the subcategorization approach. Thus, there has long been 

a framework available for the analysis of phonological conditions on affixation; I will 

refer to this general framework as the subcategorization approach. The basic idea behind 

this approach is that the underlying form of affixes includes specifications as to the type 

of stems to which they will attach. This can include morphological, syntactic, and 

semantic information; crucially, it can also include phonological properties of the stem. I 

will discuss the details of this approach in §1.1.2. 

Beginning with the advent of Optimality Theory (OT; McCarthy and Prince 

1993a, b; Prince and Smolensky 1993), some researchers have pursued a different 

approach to phonological conditions on affixation. From the beginning, McCarthy and 

Prince (1993a, b) incorporated into OT the possible ranking schema ‘P >> M’, where P is 

a phonological constraint 1 that outranks M, a morphological constraint. McCarthy and 

Prince claimed that ‘[i]n all cases of prosodic morphological phenomena, prosodic 

constraints dominate morphological ones’ (1993b: 154). The details of this approach, 

which I will refer to as the ‘P >> M’ approach, will be discussed further in §1.1.1. It is 

important to note that despite the fact that McCarthy and Prince proposed ‘P >> M’ from 

1 

McCarthy and Prince (1993a) use ‘P’ in ‘P >> M’ to refer to prosodic, not 

phonological, constraints. This implies that the P constraints being referred to might be 

restricted to those concerning elements of the prosodic hierarchy and not to other 

phonological elements such as features. However, one of McCarthy and Prince’s 

constraints in a P >> M analysis does refer to a feature: the CODA-COND constraint in 

Axininca Campa (1993a: 122) mandates that coda consonants can only be nasals 

homorganic to a following stop or affricate. Furthermore, even if one wanted to limit P 

constraints in the P >> M schema to constraints referring to elements in the prosodic 

hierarchy, there would be no principled way of making this restriction. Therefore, when 

McCarthy and Prince allow for P >> M in OT, then in principle they allow any 

phonological constraint to outrank any morphological constraint. Thus, throughout the 

dissertation when I refer to ‘P >> M’, ‘P’ is meant to refer to phonological constraints. 

4

the very beginning of the OT literature, this ranking schema is not a central tenet of the 

theory. It assumes that phonological and morphological well-formedness are assessed 

simultaneously, which is separate from the more basic principle of OT phonology that 

phonological well-formedness in all dimensions is evaluated in parallel by ranked, 

violable constraints. Therefore, the arguments that I will make against the P >> M 

approach do not constitute arguments against the independent use of OT in phonology or 

morphology. 

Over the last decade, the P >> M approach has become the dominant model for 

phonological conditions on affixation, but there does not appear to have been any explicit 

comparison of P >> M with subcategorization until relatively recently. Thus, the 

widespread acceptance of P >> M has proceeded without much consideration for the 

consequences of the decision to use this model as opposed to subcategorization. One 

exception is Booij 1998. Booij argues in favor of the use of phonological constraints to 

account for some types of gaps and allomorph selection, while pointing out that 

phonological output constraints are not sufficient to account for allomorphy. Thus, in 

Booij’s approach, either P >> M or subcategorization can be used depending on the 

properties of the example to be analyzed. Booij's primary argument in favor of using 

phonological constraints to select allomorphs is that the use of subcategorization frames 

makes the selection arbitrary and cannot relate the shape of the allomorph to the 

condition under which it occurs. For example, Booij argues that a simple statement of the 

distribution of Dutch suffixes (‘-s after an unstressed syllable, -en after a stressed 

syllable’) ‘...does not explain why this particular selection principle holds. In terms of 

complexity of the grammar, it would make no difference if Dutch were just the other way 

5

around, i.e. if -s occurred after stressed syllables, and -en after unstressed ones’ (1998: 

145). In Booij’s view, P >> M should be used for cases that share this optimizing 

character with the Dutch example, while subcategorization should be used when the 

allomorphy is not optimizing. Cases of phonologically conditioned suppletive allomorphy 

are thereby split into two types depending on where in the grammar the phonological 

condition is specified (in subcategorization frames vs. in phonological constraints). 

A more recent and more explicit comparison between the P >> M approach and 

subcategorization is made by Yu 2003, a study that focuses on patterns of infixation (see 

chapter 5, §5.5). Yu (2003) surveys a broad range of languages exhibiting infixation, and 

his findings lend support to his model of Generalized Prosodic Subcategorization, which 

is contrasted with the Displacement Theory approach in OT that makes use of the P >> M 

schema. In fact, Yu (2003: 108) proposes a universal ranking M >> P in direct 

contradiction to P >> M. My own proposal is somewhat different, in that I claim there is 

no ranking relation between P and M constraints; see chapter 6 for discussion. 

In this section, I have given some historical background on the two approaches to 

be contrasted here. In §1.1.1 and §1.1.2, I give more detail regarding the implementation 

of the P >> M and subcategorization approaches, respectively. 

1.1.1 The P >> M model 

In this section, I introduce the P >> M model. This is the approach mentioned 

above in which phonological conditions on affixation are modeled by ranking 

phonological (P) constraints over morphological (M) constraints in OT. In §1.1.1.1, I 

6

present the details of this model, and in §1.1.1.2, I discuss the predictions that this model 

makes for PCSA. 

1.1.1.1 Description of the model 

As mentioned above, the term ‘P >> M’ refers to the fact that this approach 

involves ranking some P constraint over some M constraint, with the result that the 

phonology has an effect on a morphological process. This constraint ranking schema was 

first proposed by McCarthy and Prince (1993a, b). 2 Perhaps the most important evidence 

for the P >> M proposal was in the domain of infix placement (to be discussed further in 

chapter 5). A classic example of a P >> M analysis of infix placement is the Ulwa 

example from McCarthy and Prince (1993a). In Ulwa (Misumalpan, Nicaragua; Hale and 

Lacayo Blanco 1989), possessive markers occur immediately after the head foot (main 

stress) of the word, as shown below (head feet are in parentheses; examples are from 

McCarthy and Prince 1993a: 79, 109-110). 

(2) (bas)-ka ‘his/her hair’ (siwa)-ka-nak ‘his/her root’ 

(su:)-ka-lu ‘his/her dog’ (ki:)-ka ‘his/her stone’ 

(as)-ka-na ‘his/her clothes’ (sana)-ka ‘his/her deer’ 

(sapa:)-ka ‘his/her forehead’ (ana:)-ka-la:ka ‘his/her chin’ 

McCarthy and Prince (1993a: 110) propose a P constraint to account for this 

(where Ft' is the head foot), shown below. 3 

(3) ALIGN-TO-FOOT (Ulwa): Align([POSS]Af, L, Ft', R) 

2 In fact, it was originally claimed that prosodically conditioned morphology is always 

driven by P >> M. McCarthy and Prince assert (1993a: 24) that ‘[f]or morphology to be 

prosodic at all within OT, the ranking schema P >> M must be obeyed, in that at least 

some phonological constraint must dominate some constraint of the morphology’. A 

similar statement is made in McCarthy and Prince 1993b. 

3 I assume that a given constraint is a P constraint if it makes reference to a phonological 

element. This is following McCarthy and Prince: ‘…AFX-TO-FT is a P-constraint, because 

it crucially refers to a prosodic notion, the foot…’ (1993b: 114). 

7

This constraint requires possessive affixes to occur to the right of the main stress in the 

word. 

The M constraint given below (McCarthy and Prince 1993a: 111) designates the 

possessive affixes as suffixes by aligning them to the right edge of the stem. 

(4) ALIGN-IN-STEM: Align ([POSS]Af, R, Stem, R) 

This reflects the assumption that every affix is either a prefix or a suffix, according to the 

morpheme-specific Alignment constraint that determines its placement. 4 

Finally, the ranking of ALIGN-TO-FOOT (the P constraint) over ALIGN-IN-STEM 

(the M constraint) yields the infixation pattern observed in Ulwa, as seen in (5)a 

(McCarthy and Prince 1993a: 112) and (5)b. 

(5) a. /siwanak, ka/ ALIGN-TO-FOOT ALIGN-IN-STEM siwa,ka,nak 

a. (siwa)nak-ka *! ‘his/her root’ 

b. (siwa)-ka-nak * 

/sapa:, ka/ ALIGN-TO-FOOT ALIGN-IN-STEM sapa:-ka 

a. (sapa:)-ka ‘his/her forehead’ 

b. sa-ka-pa: *! * 

This has been a brief introduction to one use of the P >> M model. In the following 

section, I discuss predictions that are made by the P >> M model that refer specifically to 

PCSA. 

1.1.1.2 Predictions for phonologically conditioned suppletive allomorphy 

The P >> M approach makes four major predictions for PCSA, to be discussed 

here. The overall results of the survey with respect to these predictions are summarized in 

4 See Yu (2003) for an argument against this assumption, based on the existence of ‘true 

infixes’, which always surface as infixes, and never as prefixes or suffixes. 

8

chapter 6, §6.3. Beyond these, more detailed predictions are made for specific types of 

PCSA; these are discussed in chapters 2-4, each of which deals with PCSA conditioned 

by a different type of phonological element. 

The first major prediction of P >> M for PCSA is that this phenomenon results 

from preexisting well-formedness constraints. This means that usually, the constraints 

used for allomorph selection in an analysis of PCSA should be motivated elsewhere. This 

may mean that the relevant P constraint is an active constraint in the language with 

effects manifested in areas of the grammar other than the allomorphy being analyzed. 

Alternatively, PCSA could result from the Emergence of the Unmarked (TETU; 

McCarthy and Prince 1994). This means that the relevant P constraint would not have to 

be motivated in the particular language being analyzed, but it should be a constraint that 

is well-motivated in other languages, so that it may be considered a universal constraint. 

A second prediction of P >> M is that PCSA should be sensitive to phonological 

elements in surface forms, not in underlying forms. This prediction follows from the fact 

that constraints in OT generally refer to surface well-formedness. It would be 

contradicted by examples where allomorphy is clearly sensitive to underlying 

phonological elements, and where the conditions on PCSA are rendered opaque by a 

phonological process. 5 

A third prediction is that, in P >> M, phonological conditions on PCSA are 

expected to come from either the ‘inside’ or the ‘outside’ (referring to elements closer to 

or farther from the root, respectively). This means, for example, that phonological 

properties of one affix can affect the distribution of allomorphs of an affix closer to the 

5 In fact, as will be seen in chapters 2-4, there do exist examples of precisely this type, 

contradicting this prediction of P >> M. 

9

oot; it also means that affixes can condition stem allomorphy. This prediction is due to 

the fact that standard P >> M analyses do not take morphological constituency into 

account; inputs consist of unordered morphemes including the root and affixes, and the P 

and M constraints establish both the order of affixes and the choice of allomorphs. 6 

A final prediction of P >> M for PCSA is that conditions on allomorph selection 

can be located anywhere in the word. In principle, there is nothing in the P >> M model 

that would prevent, e.g., an OCP constraint from driving allomorphy in a prefix that is 

sensitive to the stem-final segment, or allomorphy in a suffix that is sensitive to the stem- 

initial segment. No feature of the model itself predicts that allomorph distribution must be 

conditioned at the edge of affixation; any such effect must be explained by adjacency 

requirements specified in the definition of each of the relevant P constraints. 

The predictions of P >> M for PCSA, discussed above, are summarized below. 

(6) a. PCSA is ‘optimizing’ and analyzable using preexisting P constraints 

b. PCSA is sensitive to phonological elements in surface forms, not underlying 

forms 

c. <strong>Phonological</strong> conditioning between stem and affix can be bidirectional 

d. <strong>Conditions</strong> on allomorph selection can be located anywhere in the word 

These predictions should be borne in mind when one considers the examples in chapters 

2-4; these predictions (as well as more specific predictions made for each type of PCSA) 

are stated in each of those chapters for ease of comparison between the attested affects 

and those predicted by P >> M. In the following section, I discuss a competing model, the 

6 

It should be noted that this prediction would not be made if P >> M were couched in a 

version of OT, such as Stratal OT (SOT; Kiparsky 2000), that respects morphological 

constituency. This is because if words are built from the inside-out in ‘layers’ as in SOT 

(and in the pre-OT predecessor to SOT, Lexical Phonology), then an outer affix will 

never be able to condition allomorphy in an inner affix, since the inner affix will already 

have been selected and attached to the stem prior to the attachment of the outer affix. 

10

subcategorization model; as in this section, I begin by laying out the model, then discuss 

the predictions that it makes for PCSA. 

1.1.2 The subcategorization model 

This section introduces the subcategorization approach, in which phonological 

conditions on affixation are modeled by incorporating phonological aspects of stems 

directly into the selectional requirements of affixes. In §1.1.2.1, I present the details of 

this model, and in §1.1.2.2, I discuss the predictions of subcategorization for PCSA. 

1.1.2.1 Description of the model 

In the subcategorization approach, as mentioned above, suppletive allomorphy 

results when two or more different affixes with the same meaning have different 

subcategorizational requirements, which are selectional requirements imposed by affixes 

on stems. In a morphological subcategorization approach (Lieber 1980, Kiparsky 1982b, 

Selkirk 1982, Orgun 1996, Yu 2003, in press), affixation satisfies missing elements that 

are required as specified in the lexical entry for each morpheme. An example of a 

subcategorization account will be given below for an example from Tahitian (Polynesian, 

French Polynesia; Lazard and Peltzer 2000), which will be discussed again in chapter 2. 

In Tahitian, the causative/factitive is marked by ha’a when the root begins with a 

labial ((7)a), and fa’a elsewhere ((7)b) (Lazard and Peltzer 2000: 224-225). 

(7) a. fiu ‘se lasser’ ha’a-fiu ‘ennuyer, s’ennuyer’ 

mana’o ‘penser’ ha’a-mana’o ‘se rappeler’ 

veve ‘pauvre’ ha’a-veve ‘appauvrir’ 

b. ’amu ‘manger’ fa’a-’amu ‘faire manger, nourrir’ 

rave ‘faire’ fa’a-rave ‘faire faire’ 

tai’o ‘lire’ fa’a-tai’o ‘faire lire’ 

11

In a subcategorization account, the distribution of ha’a- and fa’a- would result not from 

the ranking of a P constraint over an M constraint, but from the different 

subcategorizational requirements of the two forms, as shown in constructions 1 and 2 

below. 

(8) Tahitian causative construction #1 Tahitian causative construction #2 

causative verb causative verb 

SYNSEM causative SYNSEM causative 

PHON ϕ(ha’a, ) PHON ϕ(fa’a, ) 

causative prefix verb stem causative prefix verb stem 

PHON ha’a PHON PHON fa’a PHON 

The same constructions can be schematized alternatively using bracket notation, shown 

below. 

(9) Tahitian causative construction #1 Tahitian causative construction #2 

[ha’a- [#C[labial]...]verb stem]causative verb [fa’a- [ ]verb stem]causative verb 

Because the phonological requirements of the ha’a form are more specific than those of 

the fa’a form, ha’a is selected when the stem is labial-initial; fa’a is selected in the 

‘elsewhere case’. 

Thus, the distribution of affixes in the subcategorization approach is determined 

by subcategorizational requirements. These are properties of each affix, and they 

determine the types of stems to which each affix will be allowed to attach. 

1.1.2.2 Predictions for phonologically conditioned suppletive allomorphy 

The subcategorization approach makes a number of predictions for PCSA, several 

of which are different from those of the P >> M approach, allowing us to distinguish the 

12

two models in terms of how well they match up with the survey results to be presented in 

chapters 2-4. In chapter 6 (§6.3), I contrast the two models in terms of the accuracy of 

their respective predictions. As in the discussion of the predictions of P >> M in §1.1.1.2, 

the predictions to be discussed here are general. In each of chapters 2-4, more specific 

predictions for particular types of PCSA will be discussed where relevant. 

One prediction made by the subcategorization approach is that PCSA is not 

necessarily phonologically optimizing, and that there should exist cases in which no 

plausible phonological well-formedness constraint can be written to account for the 

pattern of allomorphy. This is because in the subcategorization approach, phonological 

constraints play no role in determining the distribution of allomorphs. The phonological 

conditions on affixation are simply stated in the lexical entry for each affix, so the 

relationship between the condition on allomorphy and the shape of the affix is arbitrary. 

It should be stressed that this does not mean that PCSA never has an optimizing 

character. To the contrary, it has already been established (prior to the present survey) 

that many examples of PCSA look as if they maximize the well-formedness of words. 

This is not necessarily problematic for the subcategorization approach; since PCSA 

probably originates in many cases from the loss of a regular rule, it should not be 

surprising that many cases appear to be phonologically optimizing. In this approach it 

would be surprising, and problematic, if there were no examples of arbitrary, non- 

optimizing examples of PCSA. This is the type of example that would be problematic for 

P >> M, since PCSA is driven in that approach by phonological well-formedness. 

13

Therefore, the existence or non-existence of non-optimizing allomorphy is an empirical 

question that bears directly on the choice between these competing models. 7 

A second prediction is that PCSA should be sensitive to phonological elements in 

input forms, rather than surface forms. This is because morphology is assumed to apply 

before phonology, at least at each layer of morphology. Unlike the P >> M approach, the 

subcategorization approach is not based on surface well-formedness constraints. 

Therefore, the phonological requirements imposed by affixes on stems should involve 

input phonological properties, not properties that are derived by the application of later 

phonological processes. We therefore predict that there should be some languages in 

which the phonological conditions on allomorph distribution are rendered opaque by the 

operation of a regular phonological process. This is another prediction that distinguishes 

subcategorization from the P >> M approach, since as discussed above, P >> M predicts 

that PCSA should be sensitive to surface phonological properties. 

A third prediction of subcategorization for PCSA is that phonological 

conditioning can come only from the ‘inside’. This means that the distribution of 

allomorphs of a particular affix can only be conditioned by phonological properties of the 

‘stem of affixation,’ i.e. those parts of the stem that are already present at the stage of the 

derivation in which the affix in question will be attached. Linear order among affixes on 

the same side of the stem is usually assumed to correspond to the order in which affixes 

7 

In chapters 2-4, I will present several cases of apparently non-optimizing PCSA, which 

provide an argument in favor of subcategorization and against P >> M. The problem in 

dealing with such examples is to determine whether a particular pattern of allomorphy is 

optimizing or not, since it is impossible to anticipate every phonological well-formedness 

constraint that could possibly be proposed for any language. Therefore, the existence of a 

single instance of apparently non-optimizing PCSA would not constitute a knockout 

argument against P >> M in favor of subcategorization. Instead, one should consider the 

cumulative weight of all of the apparently non-optimizing examples to be discussed. 

14

are attached; for example, in a word with multiple suffixes, it is assumed that the leftmost 

suffix was attached first, and the rightmost suffix was attached last. Of course, the 

possibility of infixes intervening between other affixes prevents this generalization from 

being true one hundred percent of the time. In the usual case, however, there is a 

correspondence between surface linear order and the order of affixation. Therefore, we 

predict that the phonological properties of a particular affix cannot condition allomorphy 

in an affix that occurs closer to the root. This is yet another prediction that is different 

from the P >> M model, since the P >> M model allows for both ‘inside-out’ and 

‘outside-in’ conditioning. The existence of ‘outside-in’ conditioning would therefore 

provide evidence in favor of the P >> M model and against subcategorization (or, at least, 

against the particular implementation of it assumed here, in which words are built from 

the inside-out). 

A fourth and final prediction of subcategorization for PCSA is made when we 

incorporate a general principle of grammar that has been proposed to account for other 

phenomena but, as will be seen, is quite useful in an analysis of PCSA as well. This is the 

Generalized Determinant Focus Adjacency Condition (GDFAC; Inkelas 1990: 201, 

building on Poser’s (1985) Determinant Focus Adjacency Condition). The GDFAC is 

given below. 

(10) Generalized Determinant Focus Adjacency Condition: Each phonologically 

constrained element must be adjacent to each constraining element. 

If the GDFAC is incorporated into the subcategorization approach, we predict that affix 

allomorphs should occur immediately next to the phonological elements of stems that 

condition their distribution. For instance, prefix allomorphs should be conditioned by 

elements at the left edge of the stem, while suffix allomorphs should be conditioned by 

15

elements at the right edge. This prediction, as with the previous three predictions, 

constitutes a difference between the P >> M and subcategorization approaches. 8 

In this section, we have discussed four predictions of the subcategorization 

approach for PCSA, summarized below. 

(11) a. PCSA is not always phonologically optimizing 

b. PCSA is sensitive to phonological elements in underlying/input forms, not surface 

forms 

c. <strong>Phonological</strong> conditions on PCSA can come only from the ‘inside’ 

d. Affix allomorphs occur adjacent to the phonological elements of stems that 

condition their distribution 

Each of the predictions has been contrasted with predictions of the P >> M approach. We 

are now equipped to consider the results of the cross-linguistic survey of PCSA, since we 

have set up some expectations for the types of examples that should be found under each 

of the two competing models. In the following section, I present some background for the 

survey of PCSA, beginning with a discussion of previous discussions and surveys of 

PCSA (§1.2.1), and concluding the chapter in §1.2.2 with a discussion of the 

methodology for the new survey whose results will be reported in this dissertation. 

1.2 <strong>Phonological</strong>ly conditioned suppletive allomorphy 

In this section, I focus on PCSA, first discussing at length some previous 

treatments of the phenomenon (§1.2.1). I then explain (§1.2.2) the setup and 

methodology for the survey of PCSA, whose results will be presented in chapters 2-4. 

8 

It should be acknowledged that the GDFAC is not a property of the subcategorization 

model in itself but is rather an assumption that is made about the nature of 

subcategorization. However, it should also be pointed out that while the GDFAC is easily 

incorporated into the subcategorization approach, it cannot be readily implemented in P 

>> M since that model does not really involve any ‘phonologically constrained elements’. 

16

1.2.1 Previous discussions 

The first major and explicit discussion of PCSA is in Carstairs 1988. The primary 

concern of Carstairs 1988 is to use examples of PCSA to gain insight into how (if at all) 

to distinguish between inflection and derivation. Thus, the point of the article is not to 

document PCSA for its own sake, though Carstairs presents several examples 9 of PCSA 

(1988: 70-72) in order to document the existence of the phenomenon, which was not well 

known at the time. Carstairs discusses the notion that derivational morphology never 

exhibits suppletion because the concept of suppletion relies on the existence of a 

paradigm, and derivational morphology does not involve paradigms (1988: 74). Carstairs 

argues against this idea because derivational morphology does have a paradigmatic 

aspect (van Marle 1985, 1986) and because some of the examples that Carstairs cites do 

involve suppletion in derivational morphology. Furthermore, derivational morphs can 

have phonological restrictions (e.g., English -en can only attach to adjectives ending in a 

certain restricted set of obstruents), so it is unclear why two derivational morphs with the 

same meaning cannot to enter into suppletive relationships conditioned by phonology. 

Carstairs makes the generalization (1988: 75) that most of the examples in his set 

involve inflection rather than derivation 10 . The explanation provided for this fact invokes 

the principle of ‘inflectional parsimony’, which states that for any combination of 

morphosyntactic features that can mark members of a particular word class, each word 

9 

Fifteen examples are discussed, from ten languages. 

10 

Of course, the concept of ‘most’ used here is not very meaningful given the small 

sample size (fifteen examples). However, this finding does appear to hold up in the larger 

survey to be described in this dissertation. It is difficult to obtain a count of cases of 

inflection vs. derivation since many affixes do not clearly fall into one category or 

another, but among those that do, there are more cases of inflection than derivation. 

17

will have exactly one inflectional realization. According to Carstairs (1988: 76), there are 

three ways in which a language can resolve a potential violation of parsimony: first, all 

but one of the possible realizations could drop out of use; second, all of the realizations 

could be distributed arbitrarily into conjugations or declensions; third, all of the 

realizations could be distributed according to some independent principle (whether 

semantic, syntactic, morphological, or phonological). Thus, the development of 

phonologically conditioned suppletion is just one of many ways in which a language can 

adhere to the parsimony principle. 

Carstairs goes on to argue that ‘...a principle of parsimonious coverage does 

appear to exercise an influence over not only inflectional morphology and syntactic 

structure but also certain areas of lexical organisation involving even monomorphemic 

items. If this is so, it would be surprising if the principle could not also affect 

morphologically complex lexical items, including derived words’ (1988: 79). An example 

of the former type of effect in English is pointed out, where animal species have one item 

in each of the categories adult male, adult female, and young. According to Carstairs, it is 

this type of derivation, which he describes as ‘meaning-driven’ 11 , in which parsimonious 

coverage is apparent. Another example is male vs. female titles of ranks of the British 

peerage (duke and duchess, etc.). An example of English derivation that is not meaning- 

driven is deverbalization using -ion, -al, -ment, -ance, or stress shift. In this type of 

derivation, the exact meaning of the noun derived from the verb is not predictable (e.g., 

remit vs. remission), and we do not find parsimonious coverage: some verbs have 

11 Carstairs defines ‘meaning-driven’ morphology as having ‘semantic coherence, 

describable in terms of semantic features or components,’ whereas ‘expression-driven’ 

morphology exhibits ‘semantic vagueness or unpredictability’ (1988: 88). 

18

multiple possible nominal forms (e.g., commission, commital, commitment). In Carstairs’ 

view, the fact that only meaning-driven derivation obeys the principle of parsimonious 

coverage accounts for the fact that cases of phonologically conditioned suppletion are 

more often inflectional than derivational. The question is then posed whether we might 

find some example of phonologically conditioned suppletion in a type of meaning-driven 

derivation, and Carstairs does present one such case from Dutch, where the distribution 

of suffixes that create ‘neutral deverbal personal names’ (e.g., terms meaning ‘drummer,’ 

‘camper,’ etc.) is conditioned by the stem-final segment(s) (1988: 84). Carstairs 

concludes by suggesting that the concepts of ‘meaning-driven’ vs. ‘expression-driven’ 

morphology may turn out to be more useful than inflection vs. derivation. Thus, PCSA is 

used to make an argument against the distinction between inflection and derivation. 

A second important paper dealing with PCSA is Carstairs 1990. This brief article 

outlines some implications of PCSA for phonological and morphological theory. One 

primary consequence for phonology, of course, is that ‘the existence of a phonologically 

conditioned alternation does not by itself prove the existence of some synchronic 

phonological process giving rise to it’ (Carstairs 1990: 19). 

Carstairs makes some generalizations about what types of phonological factors 

can condition allomorphy. These seem to be based on the set of eleven examples that 

Carstairs provides (1990: 18). One generalization is that there are no cases where a 

phonological characteristic of a word or morph conditions allomorph selection for a 

separate phonological word. A second generalization is that, as predicted by Lexical 

Phonology, conditioning is generally ‘outward’, i.e., involving a root or inner affix 

conditioning the distribution of allomorphs of an outer affix. There are some apparent 

19

exceptions to this, but Carstairs points out that in each case of apparent 'inward' 

conditioning, the alternants are either phonetically uninterpretable in isolation, or else 

they can be analyzed as ‘empty morphs’. 12 Thus, the direction of conditioning in PCSA 

follows from a layered model of word-building such as Lexical Phonology. 

The first major paper dedicated to PCSA in the OT era was Mester 1994. 13 Mester 

provides an analysis of Latin stress, arguing in favor of bimoraic minimality and 

maximality in foot structure. Mester’s analysis likely constitutes the first use of 

phonological constraints to determine the distribution of suppletive allomorphs, aside 

from McCarthy and Prince 1993a, b. Mester uses this approach to account for three 

instances of allomorphy in Latin, all of which appear to be configured in such a way as to 

avoid 'trapping' configurations in which a syllable is left unfooted due to strict bimoraic 

foot minimality and maximality. Further details of Mester’s Latin examples will be 

discussed in chapter 4. 

Another important contribution to the literature on PCSA was made by Kager 

1996. Kager describes a phenomenon of ‘syllable-counting allomorphy’ in Estonian, 

arguing that this type of allomorphy is a form of output optimization. Kager argues that 

syllable-counting allomorphy reduces to foot-based allomorphy, and that the result is to 

maximize foot parsing. As will be discussed in chapter 4, the problem with this view is 

12 

As will be seen in chapters 2-4, both of the observations made by Carstairs (1990) are 

upheld in a larger survey of PCSA. First, PCSA always occurs within the (phonological) 

word (always between an affix and the stem of attachment, or else between a clitic and its 

host); no cases of between-word conditioning are found. Second, conditioning does 

appear to be ‘inside-out’, not ‘outside-in’, and the very few cases of apparent ‘outside-in’ 

conditioning yield to alternative analyses. 

13 

In fact, Mester’s analysis is couched in the earlier, similar theory of ‘Harmony- 

Theoretic Phonology’ rather than OT, which was only beginning to be developed at the 

time of publication. However, the Harmonic approach was also constraint-based, so the 

basic insights of Mester’s analysis could easily translate into OT. 

20

the assumption that all syllable-counting allomorphy is optimizing and based on foot 

structure. This ignores the many cases (to be discussed in §4.1.1.3) where syllable- 

counting allomorphy is not optimizing and may not reduce to considerations of footing. 

A final important study of PCSA is Carstairs-McCarthy 1998. This article 

provides a useful way of looking at PCSA which unifies it with the phenomenon of 

phonologically conditioned morphological gaps. In Carstairs-McCarthy’s view, both 

phenomena result from phonological restrictions on the distribution of particular affixes. 

Different languages react to these gaps in one or more of three ways: (a) unsystematic 

filling of the gaps, (b) systematic morphological filling of the gaps, and (c) systematic 

syntactic filling of the gaps via periphrasis. Carstairs-McCarthy points out an interesting 

asymmetry: in the domain of inflection, strategies (b) and (c) are used, while in the 

domain of derivation, strategy (a) is used. 14 Carstairs-McCarthy relates this (1998: 148) 

to ‘the importance of the paradigmatic dimension in inflection’, though as discussed 

above, Carstairs (1988) points out that paradigms may play a role in certain types of 

derivation as well. 

As can be seen, the theoretical treatment of PCSA has been relatively sparse, 

since the five papers discussed above constitute the important recent literature on the 

topic to deal with PCSA cross-linguistically. 15 This is despite the fact that, as pointed out 

repeatedly in this literature, PCSA has enormous potential to shed light on the nature of 

14 

Only seven examples are considered; in the larger survey of PCSA to be discussed in 

chapters 2-4, this generalization is contradicted by the existence of many examples of 

PCSA in derivational affixes (though, as pointed out earlier in this section in the 

discussion of Carstairs 1988, the generalization does seem to hold as a tendency in the 

data). 

15 

There are several other papers that discuss PCSA to varying degrees of depth; see, e.g., 

Mascaró 1996, Tranel 1996a,b, Vaux 2003, Bonet 2004, Bonet et al in press, and Bye to 

appear. 

21

phonology, morphology, and especially the phonology-morphology interface. The focus 

of most of these earlier works has been on using PCSA to answer broader questions about 

morphology (Carstairs 1988), providing constraint-based analyses for it (Mester 1994, 

Kager 1996), or explaining why PCSA occurs (Carstairs-McCarthy 1998). Only one 

paper, Carstairs 1990, attempts to make serious typological generalizations about the 

phenomenon, and that paper relies on only eleven examples of PCSA. It should therefore 

be clear why a larger cross-linguistic survey of PCSA is desirable. I introduce such a 

survey in the following section, discussing the methodology before presenting the results 

of the survey in chapters 2-4. 

1.2.2 Survey methodology 

In order to determine the precise range of phonological conditions that can 

determine suppletive allomorph distribution, and the specific ways in which this 

conditioning can be manifested, a cross-linguistic survey of PCSA was undertaken. The 

results of the survey will be reported in chapters 2-4, but it is important here to discuss 

the methodology used in the construction of the survey. 

The primary problem in constructing a large cross-linguistic survey is that one can 

never survey enough languages. Comprehensive grammars have been written for only a 

small fraction of the world’s living languages, and this completely ignores extinct 

languages. It is impossible to know what type and amount of bias this introduces into 

cross-linguistic surveys. This is especially problematic when making negative 

generalizations of the type ‘No language has property X’. 

22

Fortunately, most of the important generalizations that I draw based on the 

present survey are positive generalizations. For example, as will be discussed, I have 

found that PCSA can be conditioned by feet, syllables, moras, stress, tone, segments, or 

features. This type of generalization is much easier to make than a negative 

generalization, since any phenomenon for which there are multiple examples from 

different language families (as is the case for the generalization stated above) is clearly an 

‘attested’ phenomenon, regardless of the sample size. 

In some cases, I do make negative generalizations. One such generalization is that 

based on the survey data, it appears that there is no language in which PCSA in an affix is 

triggered at the opposite edge of the stem; for example, there is no language where prefix 

allomorphy is triggered by the stem-final segment. Of course, this generalization can be 

falsified if another researcher uncovers such an example in future research. In that case, a 

good scientific way to proceed would be to discern whether the apparent counterexample 

can be interpreted so that it does not directly contradict the theory that was devised in 

order to explain the generalization; if this proves impossible, then the theory should be 

abandoned. In the mean time, my confidence that the generalization does reflect some 

true fact about human languages (rather than a sampling error) can be increased by my 

own deliberate efforts in constructing the survey, which I discuss below. 

The best way to avoid making spurious negative generalizations is to make the 

survey as large and broad as possible. I have attempted to do this in a variety of ways. 

First, I have scoured several types of sources in my search for examples. About 600 

sources were consulted in total, though of course not all of these yielded examples of 

PCSA. Most of the examples to be presented were found by searching through 

23

descriptive and teaching grammars on the shelves at the UC Berkeley library. I also 

became aware of some examples by querying several specialists in phonology and 

morphology. This was complemented by the addition of many examples already 

discussed in theoretical literature (mainly in journal articles) and in linguistics textbooks. 

Whenever an example was found in one of these secondary sources, I have attempted to 

locate the primary source to confirm the examples and the characterization of patterns. 

The minimal standard for inclusion of a particular example in the survey was that the 

shape and distribution of allomorphs must be clear, and examples must be given. In 

general, I have used the transcriptional or orthographic conventions employed in each 

source, except where noted. In some instances where examples are neither provided in 

the primary description nor accessible elsewhere, I have described these cases in 

footnotes. In general, I have tried to make the survey as large as possible. 

Though I did not deliberately seek out examples from every language family in 

creating the database, I did attempt to balance the survey. Throughout the course of the 

survey research, I continued to take stock of the examples, and whenever it appeared that 

a major language family was not well-represented in the survey, I made a particular effort 

to obtain grammars of languages in that family. 16 Thus, the survey was not intentionally 

balanced in a systematic way, but the result of the methodology described here seems to 

have resulted in a relatively balanced set of examples. The reader may assess this for 

him- or herself by consulting the Appendix, which lists the languages and families from 

which I obtained the data to be discussed in the following three chapters. 

16 

Except where noted, all language family classifications are from the Ethnologue 

(Gordon 2005). 

24

Chapter 2: Segmentally conditioned suppletive allomorphy 

In this chapter I present examples and an analysis of suppletive allomorphy 

conditioned by consonants and vowels and their features. The ‘P >> M’ ranking schema 

discussed in chapter 1, where phonological constraints are ranked ahead of morphological 

constraints, predicts that any segment or feature anywhere in the word should in principle 

be able to condition the distribution of suppletive allomorphs of affixes or stems for 

reasons of output optimization. The extent to which this prediction is or is not borne out 

is discussed in §2.1, where I present examples of segmentally conditioned allomorphy 

revealed by a cross-linguistic survey. In §2.2, I give an analysis of representative 

examples of this type in terms of subcategorization, demonstrating that such an analysis 

is feasible and superior to a P >> M analysis. The chapter is concluded in §2.3 with a 

summary of the findings and the analysis. 

2.1 Survey results 

In this section, I give examples of suppletive allomorphy conditioned by segments 

and their features. These examples were collected as part of a cross-linguistic survey of 

phonologically conditioned suppletive allomorphy (PCSA), described in chapter 1. 

Grammars of over 600 languages were surveyed, yielding 137 examples of PCSA in 67 

languages. Seventy-two of these cases (from 32 different languages) involve conditioning 

by consonants or vowels or their features and are discussed in this chapter. The examples 

in this chapter are organized according to the effect that the pattern of allomorph 

distribution has on the word, as follows. 

25

§2.1.2.1 discusses cases where allomorph distribution has an assimilatory or 

harmonizing effect; that is, where the degree of featural agreement between the stem and 

affix is higher than if there were no allomorphy or if the allomorphs had a different 

distribution. The examples in §2.1.2.2 involve patterns that result in dissimilation or 

antiharmony, where the degree of featural agreement between the stem and affix is lower 

than if there were no allomorphy or if the allomorphs had a different distribution. In 

§2.1.2.3, I discuss examples where the distribution of suppletive allomorphs appears to 

improve the overall syllable structure of the word. A related set of examples is discussed 

in §2.1.2.4, which describes examples where allomorph distribution appears to relate to 

syllable contact (rather than the shape of individual syllables). In §2.1.2.5, I discuss cases 

where the allomorphy has an effect that is non-optimizing. Finally, in §2.1.2.6, I present 

examples where phonological processes render opaque the conditions on PCSA. As I 

discuss, these are especially problematic for the P >> M model. 

The organization of this section according to the ‘outcome’ of allomorph 

distribution may seem to presuppose the truth of the claim inherent in the P >> M 

approach, that phonologically conditioned suppletive allomorphy is phonologically 

optimizing. However, as will be discussed in the summary in §2.1.3, this is far from a 

foregone conclusion. As will be discussed, the examples in §2.1.2.5 where the allomorph 

distribution is phonologically non-optimizing are problematic for the claim that PCSA 

results in output phonological optimization, as are the examples in §2.1.2.6 where PCSA 

is shown to be sensitive to phonological features in the input rather than the output. The 

theoretical consequences of these examples are discussed in §§2.2 and 2.3. 

26

Some generalizations that will emerge in the presentation of examples is as 

follows. First, the phonological element that conditions PCSA is always adjacent to the 

location of attachment of the affix in question. Prefix distribution is conditioned by 

segments and features at the left edge of the stem, while suffix distribution is conditioned 

by segments and features at the right edge. Second, a wide range of consonant and vowel 

features, as well as the C/V distinction itself, can condition the distribution of suppletive 

allomorphs. Third, a substantial number of examples of PCSA appear to be non- 

optimizing, a finding that is not predicted by P >> M and that is, in fact, problematic for 

that model. Finally, a related finding is that PCSA seems to be sensitive to phonological 

elements in input forms, rather than outputs. 

2.1.1 Allomorphy vs. morphophonology 

An issue that arises in the presentation of examples to follow in this chapter (and 

also in chapters 3 and 4 where I discuss examples with other types of phonological 

conditioning) is the difficulty of determining whether a particular example of 

phonologically conditioned allomorphy involves suppletion. Kiparsky (1996) discusses 

this problem and proposes the following criteria to distinguish between instances of 

morphophonology (phonological rule/constraint-driven allomorphy) and (suppletive) 

allomorphy (1996: 17): 

(1) Morphophonology Allomorphy 

a. general (not item-specific) item-specific 

b. involve a single segment may involve more than one segment 

c. observe phonological locality conditions obey morphological locality conditions 

d. follow all morpholexical processes ordered prior to morphophonemic rules 

Some of these criteria are more useful than others, and as will become evident in the 

presentation of examples, there is a significant gray area left by these criteria, such that 

27

some examples are difficult to classify. Criterion (a), involving the generality of the 

pattern, is one of the more useful criteria. If a phonological rule/constraint proposed to 

account for a pattern of allomorphy in a particular morpheme can also account for one or 

more other patterns of allomorphy in the same language, this suggests that the 

allomorphy is best analyzed as resulting from the application of phonological 

rules/constraints to a single underlying form (i.e., that the allomorphy is not suppletive). 

If, on the other hand, the rule/constraint that would need to be posited to account for a 

pattern of allomorphy would only be manifested in that particular morpheme, then the 

allomorphy is more likely to be suppletive. However, in some cases it can be argued that 

a particular affix or group of affixes is associated with a ‘co-phonology’ (Inkelas, Orgun, 

and Zoll 1997, Inkelas 1998) that might result in particular rules or constraint rankings 

that do not apply throughout the entire language. This is especially useful if a 

construction that includes a particular affix (or one of a group of affixes) seems always to 

have a particular phonological property (as with, for example, stress-shifting affixes). 

A related factor that bears on whether to analyze a particular pattern as suppletion 

or item-specific rule application is the phonetic naturalness of the proposed rule. Suppose 

that a particular morpheme presents the only instance in the language of the phonological 

configuration that would trigger the rule. In this case, the rule would be both ‘general’ (in 

the sense that it applies everywhere in which the phonological environment for its 

application is met) and ‘item-specific’ (since it would only apply to one morpheme). 

Therefore, in this hypothetical situation, criterion (a) is of no help in determining whether 

this is a case of rule-driven allomorphy or suppletion. 1 A secondary consideration in such 

1 

This situation is exemplified by an example from Spanish to be discussed in §2.1.2.5. 

28

a situation is the plausibility of the proposed phonological rule. If the rule is item-specific 

but is also phonetically natural and attested in a number of languages, this is an argument 

in favor of rule-derived rather than suppletive allomorphy. On the other hand, if the 

proposed rule would be phonetically unnatural and not attested in other languages, this 

suggests that the pattern involves suppletion. This criterion does, of course, involve some 

subjective judgment as to what constitutes ‘phonetic naturalness,’ but this notion in 

combination with the frequency of such rules in other languages of the world should 

provide some idea of the overall plausibility of the rule. 

Criterion (b), which refers to the number of segments involved in the alternation, 

is also relatively useful but not unproblematic. It is quite clear-cut when each of the 

allomorphs has several segments and none of them are the same (or even similar) from 

one allomorph to the other; in such a case, we would almost certainly analyze the pattern 

as suppletive. This is not a common situation, however. More common is a situation 

where one or two segments are different in the different allomorphs. If only one segment 

differs, the pattern may be rule-derived, though in this case we would have to evaluate 

the proposed rule based on the discussion above to decide whether its generality and 

plausibility merit positing a phonological rule for the language. If two or more segments 

differ, the pattern is more likely suppletive unless all the segments can be altered (or 

deleted or inserted) via a single phonological rule. If multiple rules are required, is each 

of the rules independently motivated in the language or do we have to posit multiple, 

separate, item-specific phonological rules just to describe the pattern of allomorphy for 

one morpheme (in which case it is probably suppletive allomorphy)? 

29

Criterion (c), involving phonological vs. morphological locality, is somewhat less 

useful than (a) and (b) since the question of whether suppletive allomorphy can be 

conditioned at a distance is an empirical question being investigated here. We have not 

yet established whether allomorph distribution can be determined by any element in the 

word including those that occur ‘outside’ the morph in question (as predicted by P >> M) 

or only by an element in the stem to which the morph attaches. Both of these situations 

might be said to observe ‘morphological locality’. We also have not established that 

PCSA does not obey phonological locality. In fact, as will be seen in the examples to be 

presented here, PCSA does appear to obey phonological locality, making criterion (c) 

irrelevant in deciding between rule-derived and suppletive allomorphy. 

Similarly, criterion (d) involving ordering is of little use, though in this case due 

to theoretical considerations. In OT, particularly in a version of OT allowing P >> M, the 

concept of allomorphy being ordered after all morpholexical rules or before all 

morphophonemic rules has no status. Thus, if we want to talk about examples of 

allomorphy in terms of OT in order to compare possible analyses, we cannot rely on 

criterion (d) to determine how to classify the examples. 

In summary, though some of Kiparsky’s (1996) criteria for identifying suppletive 

vs. non-suppletive allomorphy are useful, others are not, and the set of criteria as a whole 

leaves a substantial gray area, so that subjective judgment will sometimes decide the 

categorization of a particular case. The existence of this gray area is not surprising, since 

many examples of suppletive allomorphy probably result from historical processes of rule 

telescoping and/or the restriction of productive phonological rules to particular 

morphological contexts. Thus, they may have many of the properties of regular 

30

phonological processes but also lack some of those properties. Fortunately, the distinction 

between suppletive and non-suppletive allomorphy is not crucial to the argument made in 

this dissertation. Since the P >> M mechanism in effect extends a phonological model to 

handle some morphological processes, we expect that the mechanism should handle the 

purely phonological (i.e., non-suppletive) examples well and that the problematic cases 

(if any) should be the suppletive ones. This means that if anything, if we admit any cases 

of rule-derived allomorphy into our discussion of suppletive allomorphy, we will err on 

the side of including too many examples that favor the P >> M approach. To offset any 

such bias, in making generalizations about PCSA I will focus on the clearest cases of 

suppletive allomorphy and not rely too heavily on examples that fall into the gray area. 

2.1.2 Examples 

2.1.2.1 Assimilation/harmony 

Given that the P >> M mechanism accounts for suppletive allomorphy using the 

same phonological constraints that drive purely phonological processes, this model 

predicts that every phonological constraint can show effects in morphology. That is, 

constraints that commonly drive phonological processes in the world’s languages should 

also condition suppletive allomorph selection. Assimilation is a very common type of 

phonological process with roots in coarticulation, and the P >> M model therefore leads 

us to expect assimilation to condition suppletive allomorphy. Several constraints have 

been proposed to drive harmony and other assimilation of various types, including 

AGREE, SPREAD, and ALIGN (see Pulleyblank 2002 for a summary of approaches to 

31

harmony). Under the P >> M approach, any one of these commonly used constraints (at 

least one of which must be highly ranked in any language exhibiting assimilation) could 

outrank a morphological constraint; this should give rise to PCSA driven by assimilation. 

In this section, I discuss five languages where allomorphy could be claimed to result in 

assimilation of consonant or vowel features. 

One example of PCSA resulting in harmonization of vowel features is found in 

Kwamera (Central-Eastern Oceanic, Vanuatu; Lindstrom and Lynch 1994). The 

perfective prefix in Kwamera has two allomorphs that are distributed based on the initial 

vowel/segment of the stem (see below), and their distribution results for the most part in 

words that are harmonic in terms of vowel height. The Kwamera vowel system is as 

follows (Lindstrom and Lynch 1994: 3): 

(2) Front Central Back 

High i u 

Mid e ɨ o 

Low a 

The perfective prefix allomorphs are distributed as follows (Lindstrom and Lynch 1994: 

12): /ɨn-/ occurs before verbs beginning with /a/, /ɨ/, and /o/, as well as most verbs 

beginning with /e/. As seen above, these are the non-high vowels of Kwamera. The /uv-/ 

allomorph occurs before verbs beginning with consonants, /i/, /u/, and a few verbs 

beginning with /e/ (and in these cases, /e/ surfaces as [a]). Examples are shown below 

with page numbers from Lindstrom and Lynch 1994. 

(3) a. r-p-ua ia-p-ɨn-ata ia-p-ɨn-osi 

3sg-COND-state 1exc-COND-PERF-see 1exc-COND-PERF-hit 

‘If I had seen it, I would have hit it’ (10) 

32

ik-ɨn-ata iakunóuihi óuihi nah ua 

2-PERF-see child small PREVREF or 

‘Did you see that small child?’ (21) 

b. in r-uv-kusi kafete 

he/she 3sg-PERF-weave mat 

‘She wove a mat’ (10) 

iak-uv-regi kwanage ira 

1exc-PERF-hear story LOC:3sg 

‘I heard his/her story (the one told about him/her)’ (23) 

Thus, the /ɨn-/ allomorph is used when the following vowel is [-high] ((3)a), and the /uv-/ 

allomorph is used elsewhere ((3)b), including before [+high] vowels (though the /uv-/ 

prefix is only shown before consonant-initial roots among Lindstrom and Lynch’s 

examples). The result of the allomorphy is therefore to harmonize the prefix vowel with 

the first vowel of the stem in terms of vowel height (recall that /ɨ/ is a non-high vowel in 

Kwamera). Note, however, that this cannot be attributed to a rule of vowel harmony, 

since the two prefix allomorphs also differ in their consonants (n vs. v). The difference in 

consonants is phonologically neutral, having no obvious effect on the well-formedness of 

the word. 

The Kwamera example demonstrates, first, that vowel height is a vowel feature 

that apparently can condition PCSA resulting in an assimilatory pattern. Second, note that 

the affix in question is a prefix and that the stem vowel that conditions the allomorphy is 

the leftmost vowel of the stem. As will be seen throughout the presentation of examples 

in this chapter, this is a common property of examples of PCSA. Prefix allomorphy is 

conditioned by elements at the left edge of the stem, while suffix allomorphy is 

conditioned by elements at the right edge of the stem. 

33

Another example of PCSA resulting in vowel harmony is found in Hungarian 

(Kenesei, Vago, and Fenyvesi 1997 [KVF], Rounds 2001). The 3sg suffix in definite 

present tense forms has two variants, /-i/ and /-ja/, whose distribution is conditioned by 

the quality of the stem vowel. After a back vowel, the form -ja is used ((4)a); after a front 

vowel, -i is used ((4)b) (Rounds 2001: 28). Examples below are from Rounds (2001: 29) 

except where noted. 

(4) a. ad-ja ‘he gives’ ró-ja ‘he carves’ (KVF 290) 

olvas-sa ‘he reads’ játsz-sza ‘he plays’ 

b. visz-i ‘he carries’ (KVF 290) főz-i ‘he cooks’ 

kér-i ‘he asks for’ lö-vi ‘he shoots’ (KVF 291) 

There are two minor complications to the description of the pattern. The first is that, as 

seen above, when -ja follows a stem-final sibilant, /j/ assimilates to the sibilant (Rounds 

2001: 27). The second is that, as also seen above, when -i follows a vowel-final stem, [v] 

is inserted before the suffix vowel (KVF 291). These can be attributed to purely 

phonological rules/constraints that apply after one of the two suppletive forms of the 

suffix has been selected. The Hungarian example shows that in addition to height, 

backness can also condition PCSA. 2 

2 

This example is somewhat problematic for the GDFAC, as Hungarian is one of only 

two languages in the survey to exhibit allomorphy conditioned by an element that is not 

at the immediate edge of the stem; a set of similar examples comes from Bari, to be 

discussed below. In both cases, allomorphy is triggered by an edgemost vowel that is not 

the edgemost segment of the stem (i.e., a consonant intervenes between the triggering 

vowel and the stem edge). To account for these examples in the subcategorization model, 

one might modify the GDFAC to state that the conditioning element must be the 

edgemost element on its C/V tier. This would mean that the element either must be the 

initial/final vowel or the initial/final consonant of the stem, but not necessarily the 

initial/final segment. I avoid making this move because only two languages motivate it 

and no language exhibits a type of allomorphy in which the initial/final consonant 

triggers PCSA across a vowel. Therefore, a better approach to Hungarian and Bari may 

be to appeal to something like extrametricality for the intervening consonants, rather than 

34

Other types of assimilation including consonant harmony across a vowel are also 

common but are not represented in this survey among examples of PCSA. There are two 

examples of what may have originated in lenition that could be related to C-to-V feature 

assimilation. One of these is found in Yidiɲ (Pama-Nyungan, Australia; Dixon 1977). In 

Yidiɲ, the locative/instrumental/allative allomorphs are distributed based on whether the 

stem is consonant-final or vowel final (Dixon 1980: 296). Their distribution is as follows: 

-la occurs after vowel-final stems, 3 as in (5)a, while -da (with assimilation of the initial 

stop to the place of a preceding nasal and predictable lengthening of the last vowel of the 

stem) occurs after consonant-final stems, as in ((5)b). Examples are shown below 

(examples are from Dixon 1977: 128-129 except where different page numbers are noted 

next to an example). 

(5) a. gabuḑu-la ‘white clay-LOC’ ḑimuru-la ‘house-LOC’ (518) 

b. muḑa:m-ba ‘mother-LOC’ warḑa:n-da ‘boat-LOC’ 

muyga:l-da ‘hole, trap-LOC’ maŋgumbar-da ‘grub sp.-LOC’ 

These allomorphs are not related through a general rule of the language, so the 

allomorphy is likely suppletive. One possible way of looking at the pattern is as lenition 

of the suffix-initial consonant between vowels. Perhaps in the surface form, the stem- 

modifying the GDFAC, the strong version of which does hold for all other examples in 

the survey. 

3 

According to Dixon (1977: 128), a more precise statement of the distribution is that 

vowel-final stems with an odd number of syllables take -la, while vowel-final stems with 

an even syllable count have their final vowel lengthened instead. However, Dixon points 

out (1977: 128) that these two allomorphs could relate to a single underlying form /-la/ 

since, based on the usual phonotactics of the language, -la would reduce to -:l for this 

stem type, and one could therefore write a rule deleting [l] in this affix (though, as Dixon 

mentions, this would be an ad hoc rule). The possibility of a rule-based analysis of the 

post-vocalic allomorphs must be the reason why Dixon (1980: 296) characterizes these 

allomorphs as having a single underlying form. 

35

final vowel and suffix-initial consonant share the [+continuant] feature in a configuration 

just like one that would result from feature spreading in a lenition rule in phonology. 

Another example that may involve lenition is found in Korean (Lee 1989), where 

the conjunctive suffix takes the form -wa or -kwa. The -wa form is used following a 

vowel-final stem ((6)a), while -kwa is used following a consonant-final stem ((6)b) 

(Lapointe 1999). Examples are shown below (examples are from Lee 1989: 113 except 

where noted). 

(6) a. ai-wa ‘child and X’ (Lapointe 1999: 271) 

sɛ-wa ‘bird and X’ 

b. pap-kwa ‘rice and X’ (Lapointe 1999: 271) 

san-kwa ‘mountain and X’ 

gaŋ-kwa ‘river and X’ 

mul-kwa ‘water and X’ 

This example is noteworthy because it seems to decrease well-formedness in terms of 

syllable structure. As pointed out by Bye (to appear), we might have expected the 

opposite distribution of allomorphs since this would have avoided onset consonant 

clusters. As will be seen in §2.1.2.3, Korean does exhibit PCSA that appears to optimize 

syllables by avoiding consonant clusters, so it is interesting that the example seen here 

has the opposite effect. As with the Yidiɲ example above, in this example it appears that 

the allomorphy relates to lenition, since the ‘weaker’ consonant occurs in intervocalic 

position. If we assume that the initial w of -wa ends up sharing the [+continuant] feature 

with the stem-final vowel, this could be seen as an example of PCSA involving 

assimilation of a consonant to a vowel feature. 

Another possible example of PCSA resulting in consonants assimilating to vowels 

is the allomorphy exhibited by some suffixes in the Sibe variety of Manchu (Tungusic, 

36

China; Li 1996). According to Li (1996: 201), there are five suffixes in Sibe whose initial 

consonant surfaces as either a velar or a uvular. If a low vowel occurs in the stem, the 

uvular-initial allomorph is used; otherwise, the velar-initial allomorph is used. The 

suffixes involved in the pattern are the adjectival diminutive (with the variants -kɨn, -kun, 

-qɨn, -qun), the comparative (-kɨndi, -kundi, -qɨndi, -qundi), the self-perceived immediate 

past (-xɨ, -xu, -χɨ, -χu), the non-self-perceived past (-xɤi, -xui, -χɤi, -χui), and the self- 

perceived remote past (-xɨŋ, -xuŋ, -χɨŋ, -χuŋ). Examples of the self-perceived immediate 

past are shown below (Li 1996: 202). 

(7) tükɛ-χu ‘to watch’ gɨnɨ-xɨ ‘to go’ 

bɔdu-χu ‘to consider’ türü-xu ‘to rent’ 

lavdu-χu ‘to become more’ utu-xu ‘to dress’ 

ömi-χɨ ‘to drink’ tɨsu-xu ‘to satisfy’ 

This example is also discussed by Bye (2005), who has two arguments for why 

these examples should be considered cases of PCSA. The first is that the pattern violates 

strict locality, since a high front vowel can intervene between the triggering low vowel 

and the initial consonant of the stem, suggesting that this is not a regular phonological 

rule. The second reason is that not all velar-initial suffixes undergo the alternation: for 

example, the imperative suffix surfaces as -kin even when the stem has a low vowel. 

Thus, Bye (2005) analyzes the suffixes as each having two underlying forms in a 

suppletive relationship. However, the problems for a purely phonological analysis that 

were pointed out by Bye seem to be less problematic than the loss of a generalization in 

the suppletive analysis. If the allomorphy is suppletive, we cannot explain why five 

suffixes exhibit the same pattern. If, on the other hand, we assume that the allomorphy is 

37

not suppletive, we have only to explain the transparency of high vowels and the failure of 

the imperative to participate. 

The transparency of high vowels can be achieved via underspecification. As one 

might have noticed in looking at the examples above, there is no mid vowel category in 

Sibe, so the vowels /ɛ/, /ɔ/, and /ö/ pattern with /a/ in triggering the use of the uvular- 

initial suffix allomorphs. Therefore, we need only one feature to describe vowel height in 

Sibe (see also Dresher and Zhang in press), and this means that if we specify the low 

vowels for height (either with [+low] or [-high]), then high vowels do not need any 

underlying height features in order for the contrast to be maintained. We could say, for 

example, that the initial consonant in each of the five participating suffixes is a [-coronal, 

-labial] suffix that is unspecified for [±high]. If the stem has a [-high] segment, then this 

feature will spread to the suffix-initial consonant, resulting in a uvular consonant. 

Otherwise, [+high] is filled in by default on both the unspecified vowels and the suffix- 

initial consonant, resulting in high vowels and a velar consonant, respectively. The 

underspecification analysis also allows us to account for the behavior of the imperative 

suffix. All that must be said is that unlike the five suffixes whose initial consonant is 

unpecified for [±high] and therefore alternates between velar and uvular, the initial 

consonant of the imperative suffix is prespecified as [+high], and it therefore invariably 

surfaces as a velar. Because five suffixes participate in this pattern and because we can 

formulate a straightforward phonological analysis using underspecification, I conclude 

that this is probably not an example of PCSA. 4 

4 

I have included the example here for completeness since Bye (2005) categorizes it as an 

example of PCSA. 

38

To summarize, we have seen here that PCSA can be conditioned by consonant or 

vowel features, resulting in assimilatory patterns. In each case, the stem segment bearing 

the feature conditioining the allomorphy occurs at the edge of the stem where the affix 

attaches. 

2.1.2.2 Dissimilation/disharmony 

Though perhaps less common than assimilation, dissimilation/disharmony is 

common enough cross-linguistically that we expect to find some cases of PCSA resulting 

in dissimilatory or disharmonic patterns. In phonology, dissimilation is often accounted 

for in OT analyses using the Obligatory Contour Principle (OCP; Leben 1973, McCarthy 

1986). One version of the OCP as an OT constraint is given below (Pulleyblank 1996: 

330). 

(8) OBLIGATORY CONTOUR PRINCIPLE: A sequence of identical elements within a tier 

is prohibited. 

The P >> M model predicts that in some languages, this constraint should drive PCSA, 

resulting in dissimilatory effects. We do in fact find some examples of this type, which 

will be discussed below. 

According to Hansson (2001: 165-166), consonant place dissimilation, even long- 

distance dissimilation, is a relatively common phenomenon. The P >> M model therefore 

predicts that we should find cases of PCSA involving consonant place dissimilation. As 

will be seen, this survey did reveal some such cases. 

One example of PCSA involving consonant place dissimilation is found in 

Tahitian (Polynesian, French Polynesia; Tryon 1970, Lazard and Peltzer 2000). The 

causative/factitive in Tahitian is marked by one of two prefix forms, fa’a- or ha’a-. The 

39

ha’a- form occurs when the root begins with a labial (f, m, p, or v), and the fa’a- form 

occurs elsewhere (Lazard and Peltzer 2000: 224), with a small number of exceptional 

roots that can take either form. The distribution is shown below (Lazard and Peltzer 

2000: 225). 5 

(9) fiu ‘se lasser’ ha’a-fiu ‘ennuyer, s’ennuyer’ 

mana’o ‘penser’ ha’a-mana’o ‘se rappeler’ 

veve ‘pauvre’ ha’a-veve ‘appauvrir’ 

’amu ‘manger’ fa’a-’amu ‘faire manger, nourrir’ 

rave ‘faire’ fa’a-rave ‘faire faire’ 

tai’o ‘lire’ fa’a-tai’o ‘faire lire’ 

The allomorphy appears to be suppletive since a rule of place dissimilation 

needed to relate the allomorphs to a single underlying form would not be a general rule of 

the language. There do not seem to be any affixes with /f/ that we could compare with the 

causative, but we do find many surface counterexamples to /f/ dissimilation in (apparent) 

reduplicated forms and within roots (10), some of which are seen below (Tryon 1970: 

148-160). 

(10) fefe ‘twisted’ fe:fe: ‘a boil’ 

fifi ‘to be in difficulties’ faufa’a ‘gain, profit, worth’ 

Thus, there is no general ban against sequences of labial consonants (with intervening 

vowels and with or without intervening glottal stop). There is, however, some other 

evidence for a relationship between /f/ and /h/. According to Tryon (1970: 2), there are 

5 

It is worth pointing out that there is a large number of roots (43, to be exact) that can 

take either form of the prefix (Tryon 1970: 42), and there are several that take a form not 

predicted by Lazard and Peltzer’s generalization. For example, Tryon (1970: 149-150) 

lists six f-initial roots and two m-initial roots that take fa’a-. However, all of the roots 

listed as taking ha’a- are, as predicted by Lazard and Peltzer, labial-initial. Perhaps the 

discrepancy reflects a dialectal difference. An interesting alternative is that the pattern 

may have been in the process of becoming regularized in 1970, and perhaps the 

difference simply reflects the result of this process over the 30 years that passed between 

the two studies. 

40

several words where we find ‘f and h as variants’. The examples that are cited are shown 

below. 

(11) pufa ~ puha ‘copra’ 

u:fi ~ u:hi ‘yam’ 

tufa’a ~ tuha’a ‘share’ 

Still, since there is no indication of a synchronic rule of f → h in the language, I assume 

this to be an example of suppletive allomorphy. This example is noteworthy because the 

consonants that are, in effect, dissimilated by the suppletive allomorphy are not strictly 

adjacent or even adjacent on the CV tier, since vowels and a glottal stop intervene 

between the two consonants involved in the pattern. 6 But note that this is not problematic 

for the GDFAC, since the ‘constraining’ (conditioning) element (a stem-initial labial) is 

at the left edge of the stem and is therefore immediately adjacent to the ‘constrained’ 

element (the prefix). 

An example involving fricative dissimilation is found in Hungarian (Kenesei, 

Vago, and Fenyvesi 1997, Rounds 2001). In present tense indefinite verbs, the 2sg is 

usually marked by [-s] (Kenesei, Vago, and Fenyvesi 1997: 289-290; note that [s] 

corresponds to Hungarian orthographic , as in the examples below). When the stem 

ends in CC or in a long vowel + t, a ‘linking vowel’ is used between the root and the 

suffix (Rounds 2001: 26). Examples using this allomorph are seen in (12)a. However, 

when the stem ends in a sibilant ((12)b), the 2sg is marked by -El (where E is a mid 

6 

The [’] in these forms is a full glottal stop (Tryon 1970: 4-5), rather than a glottalization 

feature on the vowel as is the case in other language families. For any sequence of two 

vowels (whether they are identical or not), [’] can occur before the first vowel, between 

the vowels, in both locations, or neither (Tryon 1970: 5). According to Tryon (1970:5), 

the sequence V’V is disyllabic, even if the two vowels are identical. 

41

vowel that undergoes backness and rounding harmony). 7 Examples are from Abondolo 

(1988: 102), except where noted. 

(12) a. mond-a-sz ‘you say’ 

vág-sz ‘you cut’ 

vár-sz ‘you wait’ 

nyom-sz ‘you press’ 

rak-sz ‘you place’ 

b. vonz-ol ‘you attract’ 

edz-el ‘you train’ 

hajhász-ol ‘you seek’ 

főz-öl ‘you cook’ (Rounds 2001: 27) 

I am treating the pattern seen here as suppletive for two reasons. First, the allomorphs 

differ in more than one segment since the -El form has an initial vowel that the -sz form 

lacks. Also, there is no general process of sibilant dissimilation in the language. In fact, 

adjacent sibilants occur regularly and are subject to a rule of sibilant assimilation in 

which the first sibilant assimilates in place of articulation to the second (Kenesei, Vago, 

and Fenyvesi 1997: 444-446). This case demonstrates that [+sibilant] is a feature that can 

condition PCSA. 

In Caddo (Caddoan, Oklahoma; Melnar 2004), the simple future suffix exhibits 

allomorphy that results in dissimilation of sequences of glottal stops. When the stem ends 

in glottal stop, the future suffix appears as -waɁ. Otherwise, the allomorph -Ɂ.aɁ. is used. 

Examples are provided below (Melnar 2004: 65; Melnar takes the examples from the 

mostly unpublished field notes of Wallace Chafe). 

7 

It is not clear whether this form is used with all sibilant-final stems. Rounds (2001: 26) 

states that it is used when the stem ends in s, sz, z, or dz, while Kenesei, Vago, and 

Fenyvesi (1997: 290) state that it is used when the stem ends in s, sz, or z and do not 

mention dz (which they do treat as a single affricate segment rather than as a sequence of 

/d/+/z/ (1997: 383)). It may simply be that there there is a lack of stems ending in 

sibilants other than s, sz, and z that would allow us to distinguish whether or not the 

allomorphy extends to all sibilants. Since there do not appear to be any counterexamples 

given, I will assume that -El occurs after sibilants. 

42

(13) basisɁ.aɁ dikattumbakáiɁwaɁ 

/ba=sis-ɁaɁ/ /dikat#nu-n-baká=hiɁ-waɁ/ 

boil-Future what.Inter#3Dat-App-have.to.say-Future 

‘it will boil’ ‘What will he have to say (to someone)?’ 

These allomorphs differ in only one segment, and the use of a separate allomorph for 

glottal stop-final words appears to be functionally motivated since a series of two glottal 

stops might be difficult to produce. However, the rule relating these allomorphs would 

have to be item-specific, since elsewhere in the language, a sequence of two glottal stops 

simplifies to a single glottal stop, rather than dissimilating to [wɁ] (Melnar 2004: 203). 

Thus, in Caddo it seems that the dispreferred sequence /ɁɁ/ is repaired differently 

depending on whether we look at the behavior of the future suffix (which exhibits 

dissimilation) or the general phonological pattern in the language (which fuses /ɁɁ/ into 

[Ɂ]). If both patterns were to be handled via a phonological constraint such as *ɁɁ, as 

would be done under the P >> M approach, we would not have a way of accounting for 

the two different repair strategies without a specific constraint referring to the future 

suffix. 

In addition to consonant dissimilation, this survey also revealed some unusual 

instances of PCSA involving vowel dissimilation/disharmony in Bari (Eastern Nilotic, 

Sudan; Spagnolo 1933). 8 Vowel disharmony in itself seems to be relatively uncommon, 

especially in comparison to vowel harmony. 9 Even more unusual in Bari is that several of 

8 

Hall and Yokwe’s (1981) discussion of ATR harmony includes examples of some of the 

affixes discussed by Spagnolo (1933) and accounts for some of the patterns seen here. 

Where relevant, Hall and Yokwe’s (1981) analysis is referred to below, though Spagnolo 

gives some disharmonic examples that Hall and Yokwe (1981) do not mention or account 

for in their analysis. 

9 

See Krämer 1998 for a model of harmony/disharmony that accounts for the relative 

rarity of vowel disharmony. 

43

these instances of PCSA result in vowel disharmony in terms of height and possibly ATR 

even in a language that appears to exhibit vowel harmony. The vowel inventory of Bari is 

as follows (Spagnolo 1933: 8), with one modification. 

(14) i u 

ɪ ʊ 

e o 

ɛ ɔ 

ə 

a 

The vowel given as [ə] above is described by Spagnolo as a ‘central vowel, between i and 

u... the sound is not lip-rounded’ (1933: 4). I characterize it above as being low, which is 

inconsistent with Spagnolo’s description of this vowel as being ‘between i and u’, but 

consistent with the way in which this sound behaves in the patterns to be described 

below. Spagnolo uses the symbol [ӧ] for this sound, suggesting Spagnolo himself did not 

intend it to be considered a high vowel; regardless of its actual phonetic realization, this 

vowel patterns as if it were the [+ATR] counterpart to /a/, as will be shown below. 10 In 

the examples to be discussed, I have replaced Spagnolo’s [ӧ] with [ə]. The feature values 

that I am assuming for the Bari vowels are as follows. 

(15) i u ɪ ʊ e ə o ɛ ɔ a 

high + + - + - - - - - - 

low - - - - - + - - - + 

back - + - + - + + - + + 

ATR + + - - + + + - - - 

10 This is also the analysis of Hall and Yokwe, who describe this vowel in Bari as ‘a 

somewhat raised and markedly centralized low vowel… slightly lower and slightly more 

back than the vowel in American English but’ (1981: 55). This vowel is also treated as 

the [+ATR] counterpart to /a/ in Kukú (a closely related language spoken in Uganda and 

Sudan). However, it is transcribed as [ɨ] in Kukú, and Cohen (2000: 6) describes it as ‘a 

central-like vowel of high-mid height’. 

44

The causative/reciprocative prefix exhibits height disharmony as well as ATR 

harmony. Both categories are marked by the prefix tV-, where V is a [+back, -low] vowel 

whose other features are determined by the vowel of the root, though not in an 

immediately obvious way, as shown in the examples below (Spagnolo 1933: 157). 

(16) nək ‘to suck’ tu-nək ‘to give suck’ 

ŋa ‘to open’ tʊ-ŋa ‘to open with force’ 

rik ‘to drive away’ to-rik ‘to pursue’ 

gwut ‘to beat’ to-gwut ‘to beat each other’ 

rem ‘to stab’ to-rem ‘to cause to stab; to stab each other’ 

mor ‘to insult’ to-mor ‘to abuse violently; to insult each other’ 

mɛt ‘to see’ tɔ-mɛt ‘to look at each other; to cause to see’ 

kɔr ‘to divide’ tɔ-kɔr ‘to settle a dispute’ 

yɪŋ ‘to hear’ tɔ-yɪŋ ‘to listen attentively; to attract attention’ 

kʊr ‘to till’ tɔ-kʊr ‘to help each other with the tilling’ 

This pattern can be schematized as follows. 11 

(17) Stem vowel Prefix vowel 

ə u 

a ʊ 

i, u, e, o o 

ɪ, ʊ, ɛ, ɔ ɔ 

The generalization can be stated as follows: stems with [-low] vowels take a suffix vowel 

that is [-high]; this vowel then undergoes ATR harmony 12 with the stem vowel to be 

realized as either [o] or [ɔ]. Stems with other vowels (i.e., the [+low] vowels /ə/ and /a/) 

take a suffix vowel that is [+high], which also undergoes ATR harmony with the stem 

vowel, surfacing as [u] or [ʊ]. This pattern is interesting because (assuming the feature 

specifications that I have laid out for the Bari vowel system) this is not straightforward 

dissimilation involving a single feature, but rather a type of indirect dissimilation that 

11 The exact same pattern is exhibited in Kukú for the same prefix (Cohen 2000: 17). 

12 A general process of ATR harmony has been documented in Bari by Hall and Yokwe 

(1981). 

45

involves the interaction of two related features, [±high] and [±low]. One observation that 

can be made about the emphatic suffix pattern is that whatever value the stem vowel has 

for [±low], the prefix vowel will have the same value for [±high]. This is reminiscent of 

the so-called ‘α-rule’, which has since been argued to be too powerful a notational device 

since it predicts unattested rules (see, e.g., McCawley 1979). Thus, the pattern of PCSA 

seen here exemplifies a pattern of phonologically conditioned morphology that is not 

supposed to exist in pure phonology. This is problematic for the P >> M model, which, as 

has been discussed above, predicts that the same effects that are manifested in phonology 

should also be manifested in PCSA since the same phonological constraints are supposed 

to be responsible for both. 

Several suffixes also show patterns of allomorphy determined by stem vowel 

quality in such a way as to produce vowel disharmony. The emphatic verb suffix 

(Spagnolo 1933: 133-134) has the form -jV, where V is always a [+back] vowel that 

patterns as in the examples below (Spagnolo 1933: 134). 

(18) gir-jə ‘to wipe a plate, etc.’ kur-jə ‘to borrow’ 

kər-ju ‘to spoil, contaminate’ kar-ju ‘to spoil, ruin’ 

kor-ju ‘to bore beads for threading’ ker-ju ‘to steer’ 

kɛr-ja ‘to notch’ kər-ja ‘to divide, separate’ 

dɪr-ja ‘to look at’ tʊr-ja ‘to pour in’ 

The stem and suffix vowels correspond as follows. 

(19) Stem vowel Suffix vowel 

i, u ə 

ə, e, o, a u 

ɪ, ʊ, ɛ, ɔ a 

46

Stems with [+high, +ATR] vowels have a [-high, +low, +back, +ATR] suffix vowel ([ə]). 

Stems with a [-low, -ATR] vowel have the suffix vowel [a], which is [-high, +low, 

+back, -ATR]. All other stems take [u], which is [+high, -low, +back, +ATR]. In these 

examples, words are not fully harmonic with respect to the feature [±ATR] (under our 

assumed feature system), since [a] is [-ATR] but corresponds as a stem vowel with the 

[+ATR] suffix vowel [u]. 13 A possible explanation for this is that ATR harmony (which I 

have assumed to be a phonological rule rather than part of a pattern of suppletive 

allomorphy) is stem-controlled and right-to-left. This would account for why the 

causative/reciprocal prefix undergoes ATR harmony (as seen in the previous set of 

examples), but suffixes and stems do not. In this case, ATR harmony could be part of the 

suppletive pattern. Thus, PCSA in the emphatic verb suffix results in both harmony and 

disharmony: ATR harmony, but height disharmony in some cases (i.e., in some instances 

a [+high] stem vowel is paired with a [-high] suffix vowel, and a [-low] stem vowel is 

paired with a [+low] suffix vowel. The pattern of harmony/disharmony is unlike a 

phonological rule of dissimilation because two of the suffix allomorphs have the features 

[-high, +low], so the dissimilation is only apparent when we compare the disharmonic 

examples with what they would have looked like if they took the ‘elsewhere’ allomorph 

instead. If stems with /i/ and /u/ took the elsewhere (/-ju/) allomorph, then these words 

would be harmonic in terms of both [±high] and [±low]. If stems with /ɪ, ʊ, ɛ, ɔ/ took the 

/-ju/ allomorph, the resulting words would be harmonic in terms of [±low], though words 

13 

Hall and Yokwe (1981: 58) note that there are some regular exceptions to vowel 

harmony, including the fact that a [+high, +ATR] vowel in a suffix will ‘fail to interact 

with vowel harmony’ in certain morphemes. The behavior of the suffix vowel u here is 

subsumed under this generalization, though Hall and Yokwe do not offer an explanation 

for it. 

47

with stem vowels /ɛ, ɔ/ would be disharmonic in terms of [±high]. Thus, the PCSA seen 

here results in a pattern of disharmony that nonetheless does not work in quite the same 

way as a pattern that is derived via the application of a phonological rule or constraint to 

a single underlying form. This is somewhat problematic for the P >> M model, which 

predicts that PCSA ought to behave just like other, purely phonological processes driven 

by P constraints. 

A second suffix in Bari exhibiting allomorphy is the emphatic imperative, which 

Spagnolo characterizes as having two allomorphs, /-é’/ and /-í’/ (1933: 136). Examples 

are shown below. 

(20) gir-jé’ ‘wipe a plate’ ’bidd-yé’ ‘beat’ 

kur-jé’ ‘borrow’ jund-yé’ ‘vindicate’ 

rəb-bí’ ‘wrapt’ [sic] ped-dí’ ‘tether’ 

kɛg-gí’ ‘hit target’ gɪ’-yí’ ‘strip off’ 

kam-bí’ ‘paddle’ son-dí’ ‘send’ 

jɔŋ-gí’ ‘take’ rʊ’-yí’ ‘press’ 

The actual form of the suffix appears to be -CV because it appears (based on the available 

data) that the suffix has an initial consonant whose features come from the root-final 

consonant. Leaving this issue aside, we can characterize the phonological conditioning of 

the vowel alternation as follows. 

(21) Stem vowel Suffix 

i, u /-é’/ 

all others /-í’/ 

The vowel of the suffix is a [-low, +ATR] whose distribution is as follows: the /-é’/ form 

(whose vowel has the features [-high, -low, +ATR]) occurs after [+high, +ATR] stem 

vowels; /-í’/ ([+high, -low, +ATR]) occurs elsewhere (i.e., it occurs with stems whose 

48

vowel is either [-high] or [-ATR] or both). 14 The result is height disharmony for most 

stem types, and [±ATR] disagreement for some stems. Once again, however, the pattern 

is different from pure phonological disharmony because we can only understand it as 

disharmonic given the incompatibility of the two features involved. One observation that 

we can make is that if the stem vowel is not [+high, +ATR], then the suffix vowel has to 

disagree with it in at least one of the features [±ATR] or [±high]. This is not a typical 

phonological rule type, again posing a potential problem for the P >> M model. It is 

reminiscent of the anti-identity effects described by, e.g., Alderete 2001, where an output 

segment must differ minimally from its correspondent in the input or in a 

morphologically related form, but these constraints generally refer to a single feature 

(e.g., ¬IDENT(voice) discussed by Alderete 2001) and would therefore have to be 

formulated in a different way to account for the case being discussed here. 

Another suffix exhibiting phonologically conditioned allomorphy in Bari is the 

passive suffix, which has two variants, /-wɛ’/ and /-we’/. Unlike some of the Bari 

allomorphy described above, this suffix would appear to involve ATR dissimilation, 

since Spagnolo states (1933: 144) that the distribution of passive suffix variants is as 

follows. 

(22) Stem vowel Suffix 

i, u, ə, e, o /-wɛ’/ 

a, ɛ, ɪ, ɔ, ʊ /-we’/ 

This could be characterized as [ATR] dissimilation, since [+ATR] stem vowels 

correspond with the [-ATR] suffix vowel variant, while [-ATR] stem vowels correspond 

14 

This is another example that falls under the category of exceptions to ATR harmony 

given by Hall and Yokwe (1981: 58), since the i here is a [+high, +ATR] vowel in a 

suffix. 

49

with the [+ATR] suffix vowel. However, Spagnolo’s characterization may have been in 

error, since the two examples that are presented contradict the generalization, and in a 

way that makes sense based on the ATR harmony exhibited elsewhere, since both forms 

are harmonic with respect to ATR. The examples are given below (Spagnolo 1933: 144). 

(23) yuŋ-we’ ‘to be born’ ’bɔk-wɛ’ ‘to be dug out’ 

One cannot say for sure whether the initial characterization of the allomorphy was stated 

incorrectly since only two examples are given, but this seems likely since the pattern is 

otherwise highly unusual. 

A second type of passive in Bari is marked by one of three suffix allomorphs, all 

of which have the form -V (where the vowel is [+back, -high] and seems to exhibit height 

disharmony with the stem vowel. Representative examples are shown below (Spagnolo 

1933: 107-108). 

(24) kət-ə ‘to be cleared away’ lip-ə ‘to be nagged’ 

tur-ə ‘to be pursued’ ter-ə ‘to be steered’ 

kor-o ‘to be bored for threading’ gwalak-a ‘to be smashed’ 

tɛr-a ‘to be satisfied’ lɪp-a ‘to be skimmed off’ 

kɔr-a ‘to be divided’ tʊr-a ‘to be poured in’ 

kɪn-ʊ ‘to be closed’ kʊr-ʊ ‘to be cultivated’ 

The allomorphs are distributed as follows. 


i, u, ə ə 

e, o o 

a, ɛ, ɪ, ɔ, ʊ a (sometimes ʊ after ɪ or ʊ) 

The generalization (apart from the roots that sometimes take [ʊ], which I assume are 

lexical exceptions) is statable as follows: stems with [-high, -low, +ATR] vowels have 

the suffix vowel [o]. These words end up being harmonic with respect to [±low]. Other 

50

stems take a [+low] suffix vowel, whose value for [±ATR] matches that of the stem 

vowel. Thus, words using this passive marker end up being harmonic in terms of ATR, 

and those with stem vowels /e, o, ə/ are also harmonic in terms of height, but those with 

stem vowels /i, u, ɪ, ʊ, ɛ, ɔ/ end up being disharmonic with respect to [±low]. Thus, the 

allomorphy in this passive suffix does not have a uniform effect on the harmony or 

disharmony of vowels in a word: for some stems, it is harmonizing, while for others, it 

causes disharmony. 15 

A final example from Bari is the imperative active, which is marked by one of 

three suffix allomorphs (Spagnolo 1933: 111): /-í/, /-é/, or /-ɛ́/. Examples are shown 

below. 

(26) ’bət-í ‘skin’ lip-é ‘nag’ 

tur-é ‘pursue’ ter-é ‘steer’ 

kor-é ‘bore and thread’ kam-ɛ́ ‘paddle’ 

tɛr-ɛ́ ‘satiate’ lɪp-ɛ́ ‘skim off the best’ 

kɔr-ɛ́ ‘divide’ kʊr-ɛ́ ‘till’ 

The allomorphs are distributed as follows. 


ə i 

i, u, e, o e 

a, ɛ, ɪ, ɔ, ʊ ɛ 

The generalization for the distribution of the active imperative suffix allomorphs is as 

follows. Stems with the [+low, +ATR] vowel take a [+high, -low, -back, +ATR] suffix 

vowel. All other stems take a [-high, -low, -back] suffix vowel, which agrees with the 

15 Hall and Yokwe (1981: 58) account for this pattern by positing a rule of Mid-Vowel 

Assimilation, in which a [+low, +back, +ATR] vowel becomes [-low, -high] following a 

[-low, -high, +ATR] vowel. Given this and a rule of ATR harmony, all of the allomorphs 

of this passive suffix could be reduced to /-a/. 

51

stem vowel in ATR. The suffix allomorphs both have a vowel that is [-low, -back], which 

might suggest a single underlying suffix form with a [-low, -back] vowel that undergoes 

ATR harmony. However, the specification for [±high] is not derivable from the stem 

vowel via any normal phonological rule. Thus, it seems we have to stipulate that stems 

with the vowel /ə/ take the /i/ suffix vowel. This is reminiscent of the pattern seen above 

in the causative/reciprocal prefix where the affix vowel has the same value for [±high] as 

the stem vowel has for [±low]. However, in this case, the pattern holds only for stems 

with the /ə/ vowel. 

To summarize the Bari examples presented above, in each case, the affix vowel 

quality is conditioned by the root vowel, but not in any consistent way. In several cases, 

the effect of the allomorphy is dissimilatory, but not always in the same dimension. The 

distribution of affix vowels almost seems random, yet the phonological conditions on the 

distribution are straightforwardly statable. Since in each case, the allomorphs are 

phonetically similar (all -V, all -CV, etc.), it might seem reasonable to posit single 

underlying forms for each set of alternants and to relate the allomorphs through item- 

specific rules based on the generalizations given for each example. However, this would 

violate Kiparsky’s generalization that non-suppletive allomorphy results from rules that 

are not item-specific, since no one set of rules could account for the behavior of all of the 

affixes, and in fact, in the case of ATR, some of the affixes exhibit harmony while others 

sometimes disagree with the stem in ATR (and the passive may even exhibit disharmony, 

though this could have been a descriptive error). This fact would be difficult to handle 

under a P >> M account where all of the allomorphy (including pure phonological 

allomorphy, in this case including ATR harmony) is driven by a single set of P 

52

constraints interranked with M constraints. We could get a P >> M analysis to work if we 

assumed separate co-phonologies (with different constraint rankings) for each of the 

affixes described above, but the standard version of OT incorporating the P >> M schema 

does not make use of the co-phonology concept. Furthermore, if we did posit individual 

co-phonologies for each affix, some would have patterns that are contradictory to each 

other, and this could not be modeled via a simple difference in the ranking of a small 

number of constraints as is usually the case in analyses making use of co-phonologies. 

Thus, even if one is willing to posit affix-specific co-phonologies, they would probably 

be incompatible with each other. It seems therefore that these patterns of allomorphy in 

Bari are suppletive, and furthermore that they would be somewhat difficult to account for 

using the P >> M approach. 

In this section, we have seen several cases of PCSA resulting in dissimilatory 

patterns. The Bari vowel disharmony examples discussed here are of particular interest 

(and worthy of future study). Such a case is unexpected under P >> M since there are 

several different patterns that contradict each other in terms of harmony vs. disagreement 

or disharmony, making it difficult to account for the overall pattern in terms of a single 

set of phonological constraints, as would be done in the P >> M approach. 

2.1.2.3 Syllable structure optimization 

Many phonological rules/constraints seem to serve the purpose of optimizing 

syllable structure. In fact, it was ‘conspiracies’ such as the ‘law of open syllables’ in 

Slavic (see, e.g., Martinet 1952, Mareš 1965) that provided part of the impetus for 

Optimality Theory. This is because rule-based approaches have no way of formally 

53

encoding the fact that in many languages, multiple different rules often work together in 

producing a particular kind of outcome, especially in the domain of syllable structure. In 

most languages, the ‘optimal’ syllable has an onset and has no coda, or a simple coda. 

Candidates with optimal syllable structures are selected by constraints such as ONSET, 

NOCODA, and COMPLEXCODA, which feature prominently in the phonological systems of 

many languages. We therefore expect under the P >> M model that we should find 

examples of PCSA resulting in optimal syllables. It does appear that there are several 

such cases (although, as will be seen in §2.1.2.5, there seem to be just as many cases in 

which PCSA referring to the C/V distinction fails to optimize syllables). 

An example of PCSA that optimizes syllable structure by avoiding complex codas 

is found in Russian (Timberlake 2004). According to Timberlake (2004), the Russian 

reflexive marker exhibits both morphologically and phonologically conditioned 

suppletive allomorphy. The reflexive suffix has two allomorphs, [s j a] and [s j ]. The [s j a] 

variant is always used in active participle forms. In other forms, the [s j a] variant occurs 

after consonants, while the [s j ] variant occurs after vowels (Timberlake 2004: 345). This 

pattern has the effect of eliminating a potential final coda cluster that would be created by 

affixing [s j ] to a consonant-final stem. The examples below are from Wade (2002: 137) 

(the transliterations are mine). 

(28) a kupaju-s j ‘I bathe myself’ on kupajet-s j a ‘he bathes himself’ 

kupajt j e-s j 

‘bathe yourselves!’ kupaj-s j a ‘bathe yourself!’ 

ona kupala-s j 

‘she bathed herself’ on kupal-s j a ‘he bathed himself’ 

Though the allomorphs are phonetically similar, I treat this as suppletive allomorphy 

since the pattern does not result from the application of any general rule of the language 

to a single underlying form for the suffix. In this example, the allomorphy optimizes 

54

syllables since if -s j were used with consonant-final stems, the result would be word-final 

CC clusters. Therefore, under P >> M, this pattern could be accounted for using 

*COMPLEXCODA. 

Turkish (Lewis 1967) exhibits an alternation in the causative that is conditioned 

by the stem-final segment and by syllable count, the result of which may be said to 

optimize syllable structure. The causative is marked by -t with polysyllabic stems ending 

in vowels, /r/, or /l/; the suffix -DIr (where D is a coronal stop having surface alternants 

[t] and [d], and I is a high vowel) is used with all other stems (except for some specific 

monosyllabic stems that take a different, lexically determined allomorph). Some 

examples are shown below (Haig to appear: 6). 

(29) bekle-t- ‘wait-CAUS’ öl-dür- ‘die-CAUS’ 

bayil-t- ‘faint-CAUS’ ye-dir- ‘eat-CAUS’ 

getir-t- ‘bring-CAUS’ çalis-tir- ‘work-CAUS’ 

There are two ways in which this pattern may be said to be optimizing. The first is in 

terms of syllable structure, since the avoidance of -t after obstruent-final stems prevents 

causative stems from ending in a sequence of two low-sonority consonants. The second is 

that, as pointed out by Haig (to appear: 6), the distribution of allomorphs is such that in 

words with multiple causatives, there is never a sequence of two phonologically identical 

causative suffixes. Some examples involving multiple causatives are shown below. 

(30) öl-dür-t- ‘cause to kill, have killed’ (Haig to appear: 5) 

die-CAUS-CAUS- 

ye-dir-t- ‘cause to feed’ (Németh 1962: 88) 

eat-CAUS-CAUS- 

kapa-t-tɪr- ‘cause to make closed’ (Underhill 1976: 353) 

close-CAUS-CAUS- 

55

The distribution of allomorphs prevents two identical causative suffixes from occurring 

next to each other. Causative stems formed with the -DIr suffix will always be 

polysyllabic and /r/-final and will therefore take the -t allomorph if another causative is to 

be added. Stems formed with -t end in an obstruent and will therefore always take the 

suffix -DIr if a second causative is used. Though Haig resists formulating specific 

constraints against repeated morphs, he does acknowledge (to appear: 10) that the 

Turkish causative allomorphy ‘appears to conspire to avoid’ adjacent suffixes that are 

identical in both their form and their meaning. Thus, it could be the case that the pattern 

seen here is phonologically optimizing, not in terms of syllable structure, but in terms of 

the avoidance of morpheme repetition. Note, however, that this cannot be what drives the 

allomorphy, since a constraint against morpheme repetition would account for the 

distribution of allomorphs only in words with double causatives. 

An example of PCSA in a clitic that appears to optimize syllable structure is 

found in Moroccan Arabic (Harrell 1962; also discussed by Mascaró 1996). The 3sg 

masculine object/possessor clitic takes the form =h after a vowel-final stem and =u after 

a consonant-final stem (Harrell 1962: 135). Some examples are shown below (Harrell 

1962: 136). 

(31) xt ˁ a=h ‘his error’ mɛa=h ‘with him’ 

šafu=h ‘they saw him’ 

ktab=u ‘his book’ menn=u ‘from him’ 

šaf=u ‘he saw him’ 

Mascaró uses the constraints ONSET and NOCODA to select the allomorphs, since the 

distribution of the allomorphs maximizes onsets and minimizes coda consonants. This is 

our first example of PCSA involving a clitic rather than an affix (though since clitics 

56

ehave like affixes for phonological purposes, we expect them to exhibit PCSA just as 

affixes do). 

The a ~ an alternation in the English indefinite article has been characterized as 

suppletive (Zwicky 1986) since there is no longer any general rule deleting /n/ before 

another consonant or epenthesizing [n] between vowels. Any such rule would have to be 

limited to applying only to the indefinite article. If we accept that the alternation does 

involve suppletion, we may view it as optimizing syllable structure because it provides an 

onset to V-initial words while avoiding a coda consonant. This then is another example of 

a clitic (in this case, a proclitic) participating in PCSA. 

Another example that seems to maximize alternating C-V sequences is found in 

Tzeltal (Mayan, Mexico; Slocum 1948, Kaufman 1971). The 2sg is marked by aw- before 

V-initial stems and by a- before C-initial stems in this language. Slocum (1948: 80) 

appears to assume that V-initial stems begin with underlying glottal stop, which would 

mean that the distribution is determined by the presence of glottal stop. However, it 

seems possible that V-initial stems do not have underlying initial glottal stops, and 

Slocum does not provide evidence for them. Some examples are given below (Slocum 

1948: 80). 

(32) Ɂinam ‘wife’ aw-inam ‘your wife’ 

(a)h-kanan ‘idol’ aw-ah-kanan ‘your idol’ 

lumal ‘land’ a-lumal ‘your land’ 

mamlal ‘husband’ a-mamlal ‘your husband’ 

If the stems taking aw- do not have initial glottal stops, then we can say that the pattern 

maximizes onsets, since the use of a- with a vowel-initial stem would yield two onsetless 

syllables. On the other hand, if these stems do have initial glottal stops, then the 

motivation is not clear; for the moment, I will assume that these stems do not have initial 

57

glottal stops. According to Kaufman (1971: 11), [w] is allowed to occur in the 

environment V_C, so the observed alternation does not result from a general constraint 

against wC in the language. Assuming there are true vowel-initial stems, then this 

example is parallel to the English a/an example in that it maximizes alternating C-V 

sequences and avoids both codas and vowel hiatus. 

Another example of PCSA possibly motivated by syllable structure considerations 

in Tzeltal is found in the 1sg marker. In Tzeltal, the 1sg is marked by k- with V-initial 

stems and h- with C-initial stems (Slocum 1948: 80). 

(33) (a)h-Ɂat’el ‘workman’ k-ah-Ɂat’el ‘my workman’ 

Ɂakan ‘leg’ k-akan ‘my leg’ 

k’ab ‘hand’ h-k’ab ‘my hand’ 

talel ‘character’ h-talel ‘my character’ 

The characterization of this pattern is complicated by the fact that, as mentioned earlier, 

Slocum (1948: 80) apparently assumes that V-initial stems are underlyingly glottal stop- 

initial and undergo glottal stop deletion. No evidence is given for this position. If it is true 

that these stems are glottal stop-initial, then the generalization here is that /k-/ occurs 

before glottal stop and /h-/ occurs before other consonants (or ‘elsewhere’). Either way, 

this appears to be a case of suppletive allomorphy because any rule of h → k before 

vowels (or before glottal stop, depending on the analysis) would be specific to the 1sg. 

The pattern could be driven by syllable structure considerations since hC is a 

pronounceable onset and could therefore be straightforwardly syllabified, while kC in 

many instances would result in an onset sequence that is difficult to produce and might 

therefore need to be broken up via epenthesis. This is avoided because of the distribution 

of allomorphs. 

58

A third example of PCSA in Tzeltal resulting in syllable structure optimization is 

in the 3rd person marker, which surfaces as y- before V-initial stems and s- before C- 

initial stems (Slocum 1948: 80). 

(34) Ɂahwal ‘ruler’ y-ahwal ‘his ruler’ 

Ɂat’el ‘work’ y-at’el ‘his work’ 

mul ‘sin’ s-mul ‘his sin’ 

k’op ‘language’ s-k’op ‘his language’ 

As with the 1sg prefix described above, the allomorphs of the 3rd person prefix could 

only be related to a single underlying form via an item-specific rule, so based on 

Kiparsky’s (1996) criteria, this should be considered suppletive allomorphy. As with the 

1sg prefix, the distribution seen here optimizes syllables since it produces pronounceable 

onset clusters when the stem is consonant-initial. If y- occurred in this position, we would 

have onsets with falling sonority, which are universally dispreferred. Thus, once again, 

the pattern seen here can be said to optimize syllables. 

Modern Western Armenian (Andonian 1999) exhibits another example of PCSA 

that appears to optimize syllable structure, in this case by maximizing onsets and 

minimizing coda clusters. The definite article surfaces as -n after a vowel-final stem, and 

-ə after a consonant-final stem. Examples are shown below (Andonian 1999: 18, except 

where noted; transliterations based on Andonian 1999: 8-9). 

(35) lezu-n ‘tongue’ 

kini-n ‘wine’ 

gadu-n ‘the cat’ (Vaux 1998: 252) 

atorr-ə ‘the chair’ 

kirk-ə ‘the book’ 

hat-ə ‘the piece’ (Vaux 1998: 252) 

The distribution of allomorphs results in an alternating C-V pattern that seems to be 

universally preferred. However, it should be noted that although this pattern does have 

59

the effect of providing onsets and preventing coda clusters, it creates codas when the 

stem is vowel-final; codas are said to be universally dispreferred, and therefore although 

the pattern is optimizing in some respects, this comes at a price. Thus, this example, 

though it can be said to optimize syllables, also foreshadows some examples (such as in 

Haitian Creole) to be discussed in §2.1.2.5 in which PCSA patterns referring to the C/V 

distinction actually make syllable structure less optimal, not more. 

In Warrgamay (Pama-Nyungan, Australia; Dixon 1980: 266) as well as some 

other Pama-Nyungan languages (including Yidiɲ and Biri) 16 , ergative is marked by the 

suffix /-ŋgu/ after a vowel-final stem, or /-du/ after a consonant-final stem, as shown 

below. 

(36) ŋulmburu-ŋgu ‘woman-ERG’ wurrbi-ŋgu ‘big-ERG’ 

wurrbi-bajun-du ‘very big-ERG’ 

In this example, the optimizing effect is that CCC sequences are avoided. However, there 

is no reason why the vowel-final stems could not also take the -du allomorph. If -du were 

the only surface form of the ergative suffix, this would also satisfy UNIFORMEXPONENCE 

or whatever other constraint is used to ensure that in the usual case, each morphological 

category is marked by a single affix. Thus, we must have some way of making clear that 

the -du allomorph occurs only in the special case and that -ŋgu is the ‘elsewhere’ 

allomorph. This problem is addressed by Wolf and McCarthy (to appear), who analyze a 

similar example in Dyirbal by stipulating a ‘priority relationship’ between the two 

allomorphs such that the higher priority suffix must be ‘tried first’, and if this fails to 

produce a well-formed output, then the other allomorph is used. Of course, this notion of 

16 

These and other examples of ergative allomorphy in Pama-Nyungan are discussed in 

chapter 4, §4.2. 

60

the specific vs. elsewhere allomorphs must be accounted for regardless of the framework 

that is used to analyze the pattern. However, this seems less problematic in a 

subcategorization-based approach than in an approach where OT constraints select the 

allomorphs, since the OT analysis advanced by Wolf and McCarthy seems to require the 

grammar to ‘try’ allomorphs in a sequential order that is antithetical to the parallel nature 

of OT. 17 Pama-Nyungan ergative allomorphy, including the Dyirbal example mentioned 

here, will be discussed further in chapter 4 since many related cases involve conditioning 

by syllable count rather than (or in addition to) having to do with syllable structure as is 

the case here. 

An interesting example of syllable-optimizing allomorphy is found in Midob 

(Nubian, Sudan; Werner 1993), where the verbal extension denoting ‘affirmation’ takes 

the form -nò- before a consonant-initial suffix and -nòn- before a vowel-initial suffix 

(Werner 1993: 50). The two available examples are reproduced below (Werner 1993: 49; 

interlinear glosses mine). 

(37) tii-nò-hèm tii-nòn-ùwà 

drink-affirm-1sg.perfect drink-affirm-1sg.continuous.indicative 

‘I drank (completely/really)’ ‘I really drink completely’ 

This example is striking because one would assume that the order of affixation is the 

same as the surface order of the morphemes. If true, this would mean that the grammar 

has to ‘look ahead’ when selecting the form of the affirmative affix since its distribution 

is based on whether the initial segment of the following morpheme is a consonant or 

17 It is possible that Wolf and McCarthy’s characterization of this sequential ordering of 

tasks is only metaphorical and that the ‘priority relationship’ could be encoded in a way 

that does not require sequential ordering; Wolf and McCarthy do not spell out the specific 

mechanism by which the priority relationship is manifested. 

61

vowel. However, the example does not necessarily involve suppletive allomorphy; the 

fact that the two allomorphs are very similar is one indication that it does not. A possible 

analysis not involving suppletion is to say that the -nò form is based on /-nòn/ and 

undergoes deletion of /n/ before another consonant. This analysis would not require any 

‘lookahead’ on the part of the grammar. However, the rule deleting /n/ would have to be 

specific to this construction, because nC sequences are permitted elsewhere in the 

language, as in the following examples provided by Werner: èdéekìndóohèm ‘I sold’ 

(1993: 23), kándír ‘on top’ (1993: 23), kúndúl ‘name of well’ (1993: 32), àndèn ‘that’ 

(1993: 38), tìdánwà ‘I am a Midob’ (1993: 57), and àrdànkírídnûm ‘it surrounded’ (1993: 

81). 

An alternative, non-suppletive analysis of this Midob example is in terms of a 

‘ghost’ or ‘latent’ consonant, as in some examples discussed by Zoll (1996). 18 Following 

the analysis assumed by Zoll for, e.g., latent consonants in French (1996: 32-33), we 

could assume that the affirmative suffix in Midob is /-nò(n)/, where the final /n/ is a latent 

consonant that only surfaces when followed by a vowel. We could formulate a rule or 

constraints resulting in the insertion of a root node in the environment V_V, and the 

features of the latent /n/ could associate to this root node, so that the suffix would end up 

having the shape [-nòn]. In environments in which a root node was not inserted (namely, 

before a consonant-initial ending), the features of the latent /n/ would simply be stray- 

erased, resulting in [-nò]. 

On either analysis, the example does not necessarily involve suppletive 

allomorphy. This is important because, as mentioned above, this would have been our 

18 Thanks to Sharon Inkelas for pointing out this alternative analysis. 

62

first example that seemed to be sensitive to properties of an affix ‘outside’ the affix in 

question, a situation that otherwise seems not to be attested. This type of situation is 

predicted by OT using P >> M, since in its standard form, this model has not included 

any sensitivity to morphological constituency, and therefore it predicts that PCSA can be 

sensitive to phonological elements anywhere in the word. The Midob example could 

therefore have confirmed a prediction of P >> M that otherwise has yet to be attested in 

the literature. However, because there are alternative analyses available (discussed above) 

that do not require lookahead, we cannot rely on this example as confirmation of the 

prediction. What is needed is a similar example in which the allomorphs are so 

phonetically distinct as to eliminate the possibility of an analysis in terms of latent 

segments. 19 

19 

Another similar example is found in Kashaya (Pomoan, northern California; Oswalt 

1960, Buckley 1994), where the negative suffix has two phonologically selected 

allomorphs, -t h and -t h i (Buckley 1994: 334). According to Buckley, the -t h allomorph 

occurs before a vowel, while the -t h i allomorph occurs before a consonant, and the [i] in 

the preconsonantal allomorph cannot be treated as epenthetic (Buckley 1994: 334). This 

pattern is of interest because, as in the Midob example, the selection of the negative in 

Kashaya seems to require the grammar to ‘look ahead’ to see what suffix will be added 

next and whether it begins with a vowel or a consonant. The examples cited are moh-t h ‘it 

isn’t running across’ and moh-t h i-q h ‘it didn’t run out’ (Oswalt 1960: 234). Based on these 

forms, the generalization would have to refer to underlying segments rather than surface 

segments. Both of these forms in are claimed to contain the factitive marker, which 

appears to consist of a mora (since it lengthens a preceding vowel) and a tonal contour 

which ‘...consists of a pause of variable duration coupled with a slight lengthening of a 

preceding vowel, sonorant, or [mora]... and a fall in tone and fade-out or diminution in 

force on the preceding syllable’ (Oswalt 1960: 30). It apparently does not affect 

obstruents, which is why it is omitted in the examples cited here. However, Oswalt seems 

to assume that the mora contributed by the factitive counts as a vowel for the purposes of 

the negative allomorphy. It is not clear whether there may be examples where the -t h 

allomorph occurs before a vowel that surfaces, or whether there may be some different 

generalization (for example, based solely on the examples provided, the generalization 

could be that [i] is epenthesized between -t h and a following consonant). Even if the 

generalization is correct, we could analyze the [i] in terms of a latent /i/ (along the lines 

of the latent /n/ proposed for Midob above), since here again the allomorphs are very 

63

The durative suffix in Kashaya (Pomoan, northern California; Oswalt 1960, 

Buckley 1994) exhibits another pattern of allomorphy that seems to optimize syllable 

structure, though in this case there is no issue of ‘lookahead’. According to Buckley, the 

allomorphs are distributed as follows (1994: 328). ‘After a vowel-final stem, two 

allomorphs occur: -cin’ if the visible stem is monosyllabic, otherwise -men’. When the 

stem ends in an /n’/ which is part of a suffix (rather than the root), the Durative is -icen'. 

After any consonant except the alveopalatals /c, c’/, including the case of /n’/ at the end 

of a root, the Durative is -an’.’ The uses of the durative suffix that are not covered in the 

description given above involve very complex conditions that are partly morphologically 

determined and will not be discussed here. Some examples are shown below (Oswalt 

1960: 212-213). 20 

(38) sime-sime-ciˑd-u ‘to keep on sprinkling’ buwi-ciˑd-u ‘to keep on stringing’ 

duk’ilci-meˑd-u ‘to keep pointing once’ mohqa-méˑd-u ‘to drive’ 

mom-áˑd-u ‘to keep running across’ duhlud-áˑd-u ‘to keep picking one’ 

The shape of the allomorphs relates to the shape of the stems since the vowel-final stems 

take one of two -CVC allomorphs, while most consonant-final stems take a -VC 

allomorph. This avoids both consonant clusters and vowel hiatus, and can therefore be 

said to have an overall optimizing effect on the syllable structure of the word. 

Also in Kashaya, the plural movement suffix (which denotes plural people or 

objects moving) has two variants whose distribution could be said, at least in part, to 

similar, suggesting that they can reduce to a single underlying form and do not have a 

suppletive relationship. 

20 

Note that what Buckley refers to as -cin’ is transcribed as [ciˑd], Buckley’s -men’ is 

transcribed as [meˑd], and Buckley’s -an is transcribed as [aˑd]. 

64

optimize syllable structure. The allomorphs are -h-, which is infixed before the final 

consonant of a C-final root, and -ht, which is suffixed after V-final roots (Buckley 1994: 

323). Some examples are provided below (Oswalt 1960: 154, 178). 

(39) mo-ht-an’- ‘they run along (caused by some outside force)’ 

mo-ht-ac’ ‘they run along (of their own will)’ 

ké,h,l-ala-w ‘several to peer down’ 

du-k’í,h,s-im-’ ‘to scratch across several’ 

The pattern seen here always results in stems ending in hC. This is suggestive of a pattern 

of ‘imbrication’ 21 since all surface forms that are marked with some particular 

morphological category have a common shape that may be arrived at by fusing some part 

of the stem with some part of the affix. When necessary, a full version of the affix 

attaches in order to achieve the necessary surface shape, but in some cases the stem 

already contains some part of the output that would otherwise have been supplied by the 

affix, and it is in these instances where the imbrication or fusion occurs. In the case of the 

Kashaya plural movement suffix, we could say that forms marked for plural movement 

need to end in hC, and that if the stem ends in a vowel, the required hC is supplied 

straightforwardly by the -ht suffix. However, if the stem already ends in C, then the t of 

the suffix fuses with the final consonant of the stem, giving the appearance of having 

infixed -h-. This would account for the phonetic similarity between the suffix allomorph 

and the apparent infixing allomorph. 

Under an analysis involving two separate allomorphs, -ht and -h-, this could still 

be viewed an example of PCSA that functions in order to fulfill a particular surface 

requirement. It also optimizes words in a more general way, in that it avoids a possible 

21 

This is a term used by Bantuists to describe patterns where an affix fuses with part of a 

stem; see Bastin 1983, Hyman 1995, and Hyman and Inkelas 1997. 

65

CCC cluster that would occur and need to undergo epenthesis if consonant-final stems 

took the -ht form. Thus, it optimizes syllable structure, and in particular it allows the 

grammar to avoid having to fix the syllable structure using the general strategy of 

epenthesis that is otherwise applied in order to break up consonant clusters. Under the 

analysis with two suppletive allomorphs, this is an example of an infix alternating with a 

suffix (which is an effect that is predicted by P >> M, though this is the first such 

example that we have seen so far). 22 Recall, however, that under an imbrication analysis 

we would not necessarily assume that there are multiple underlying forms that mark 

plural movement in Kashaya. 

Another example where suppletive allomorphs differ in their position in the word 

is found in Biak (West New Guinean, New Guinea; discussed by Booij 2005: 174-175). 

In this language, the 2sg is marked by -w- or wa-, and the 3sg is marked by -y- or i-. The 

choice of whether a stem will take the prefix or the infix variant is lexically determined, 

but there is a clear phonological generalization involved as well, which is that all stems 

22 

Another Californian language that may have exhibited PCSA was Chimariko (Northern 

Hokan, Northwestern California; Dixon 1910). According to Conathan (2002), based on 

the unpublished field notes of George Grekoff, a set of pronominal markers had different 

forms and positions in the word that may have been phonologically conditioned, such that 

vowel-initial stems took prefixes while consonant-initial stems took suffixes. There are 

two classes of verbs (Conathan’s set I and set II) that take different sets of affixes, and 

there may have been additional factors involved, but Grekoff’s analysis was purely 

phonological aside from the distinction between sets I and II (Conathan 2002: 24). 

Examples from Grekoff’s notes appear to support the generalization (if one assumes that 

some roots assumed by Dixon (1910) to be vowel-initial actually have underlying initial 

glottal stops). These include forms (Conathan 2002: 20) with vowel-initial roots /y-ama/ 

‘I eat’ (set I) and /čhu-iman-damu-t/ ‘I fell down’ (set II), and forms with consonantinitial 

roots /kow-Ɂi/ ‘I holler’ (set I) and /čheleɁ-či-t/ ‘I am black’ (set II). However, this 

is not a clear-cut example of PCSA since other researchers (e.g., Dixon 1910) proposed 

other analyses involving semantic rather than phonological conditioning. 

66

with an initial consonant cluster take the prefix variants for both affixes. Some examples 

are shown below (Booij 2005: 175). 

(40) stem 2sg 3sg gloss 

a. ra r,w,a r,y,a to go 

b. rov wa-rov i-rov to fly 

c. kvok wa-kvok i-kvok to stand up 

snai wa-snai i-snai to be clear 

smai wa-smai i-smai to have 

A comparison of examples (40)a and b shows that the allomorphy is partly lexically 

conditioned, since these stems have the same onset but take different allomorphs of the 

2sg and 3sg affixes. The examples in (40)c, on the other hand, are representative of all 

stems in the language having an initial consonant cluster. Stems of this type invariably 

take the prefix allomorphs. This can be seen as an example of syllable structure 

optimization because if these stems took the infixing allomorphs, the result would be 

initial CCC clusters, which are avoided by using the prefix allomorphs of the 2sg and 

3sg. 

In Korean, as discussed by Odden (1993), some suffixes exhibit allomorphy based 

on whether the stem-final segment is a consonant or a vowel. The accusative is marked 

by -rɨl after a vowel and -ɨl after a consonant; the topic marker takes the form -nɨn after a 

vowel and -ɨn after a consonant, and the nominative is marked by -i after a consonant and 

-ka after a vowel (Odden 1993: 133). Some examples are shown below (this example is 

reproduced directly from Odden 1993: 133). 

(41) citation: param pori 

nominative: param-i pori-ka 

accusative: param-ɨl pori-rɨl 

topic: param-ɨn pori-nɨn 

‘wind’ ‘barley’ 

67

Odden advocates writing (presumably item-specific) rules deleting the initial consonants 

of the CVC forms of the accusative and topic suffixes to generate the postconsonantal 

allomorphs. The nominative alternation looks more like suppletive allomorphy, but 

Odden posits a rule of Nominative Destructuring (1993: 134), which derives the -i 

allomorph from /-ka/ by deleting all of the segmental content of /-ka/, resulting in a mora 

not associated to any segmental features. The mora then acquires some features via 

default rules, and ends up surfacing as [i]. While it seems reasonable in some cases to 

posit item-specific rules rather than suppletive allomorphs, this is probably best limited to 

cases where the allomorphs are very similar in their surface phonological shape (not the 

case with -i vs. -ka) and where the rule relating the allomorphs is a natural rule that is 

also attested in other languages (also not true of the proposed Nominative Destructuring 

rule). Although we cannot rule out the possibility that the allomorphs do have a single 

underlying form and that Nominative Destructuring is responsible for the allomorphy, I 

will assume for cases like this that there are two underlying forms, not only for the 

reasons given above but also because it seems likely that a language learner will posit 

multiple underlying forms if the allomorphs are not similar in shape. Thus, I conclude 

that the nominative -i ~ -ka should count as a case of suppletive allomorphy conditioned 

by C vs. V-final stems and resulting in well-formed syllables. The accusative and topic 

allomorphy examples, on the other hand, are not necessarily suppletive since both could 

be derived via a rule deleting a consonant after another consonant (which, even if not a 

general rule of the language, would generalize at least to these two suffixes). 

Also in Korean (Martin 1954: 28, 37-39) there two main verb classes, determined 

by whether the root is consonant- or vowel-final. Although Martin characterizes these as 

68

lexical classes, they are nonetheless unified by a phonological criterion. There are several 

verb suffixes that have separate forms for the two classes of roots, including the 

honorific, formal, prospective, processive, declarative, imperative, modifier, adversative 

and extended adversative, sequential and extended sequential, substantive, conjunctive 

and extended conjunctive, contingent, assumptive, intentive, purportive, frustrated 

intentive, prospective assertive, prospective attentive, intentive assertive, and cajolative 

suffixes. Interestingly, the difference between the two forms of each suffix is not the 

same in every case. For example, some suffixes (such as the authoritative indicative 

assertive -so ~ -o) have an initial s with vowel-final roots corresponding to Ø with 

consonant-final roots; others (such as the conjunctive -mye ~ -umye) have initial u with 

consonant-final roots corresponding to Ø with vowel-final roots; others (such as the 

formal indicative assertive -sumnita ~ -mnita) have both s and u with consonant-final 

roots corresponding to Ø with vowel-final roots; while yet others (such as the plain 

processive assertive -nunta ~ -nta) have n and u after consonant-final roots corresponding 

to Ø with vowel-final roots. Because the different suffixes have different 

correspondences between allomorphs that occur with consonant vs. vowel-final roots, any 

rules written to derive the allomorphy would have to be specific to certain affixes or sets 

of affixes. We could potentially approach these examples using a ‘latent segment’ 

analysis such as the ones proposed earlier for Midob and Kashaya. Thus, each of these 

affixes may not necessarily exhibit PCSA. 

Kager (1996) discusses a case of allomorphy in Dja:bugay (Pama-Nyungan, 

Australia; Patz 1991) that seems to optimize syllable structure. In Dja:bugay, the genitive 

suffix has an -n allomorph that occurs with vowel-final stems, and a -ŋun allomorph that 

69

follows consonant-final stems. Some examples are shown below (examples are from Patz 

1991: 269 except where a different page number is given in parentheses). 

(42) guludu-n ‘dove-GEN’ girrgirr-ŋun ‘bush canary-GEN’ 

gurra:-n ‘dog-GEN’ gaɲal-ŋun ‘goanna-GEN’ 

djama-n ‘snake-GEN’ bibuy-ŋun ‘child-GEN’ 

ɲurra-n ‘2sg-GEN’ (272) 

According to Kager (1996: 155), Dja:bugay does not allow coda clusters, which could 

account for the fact that consonant-final stems take -ŋun instead of -n. However, as Kager 

acknowledges (1996: 156), this accounts for only part of the distribution, since there 

appears to be no reason why vowel-final stems cannot also take -ŋun. This means that we 

must have some way of encoding the notion of the specific allomorph and the elsewhere 

allomorph. A similar problem was encountered in the Warrgamay example in §2.1.2.3. 

In Dakota (Siouan, northern United States; Shaw 1980), the 1du/pl marker is 

marked by either ų- or ųk- depending on whether the stem begins with a consonant or a 

vowel, respectively. Examples of the preconsonantal allomorph are shown below (Shaw 

1980: 77). 23 

(43) ų-híyu ‘we (dual) start to come’ 

ų-híyaya ‘we (dual) go past’ 

ų-xá=pi ‘we (pl.) bury’ 

ų-xą́=pi ‘we (pl.) do, work’ 

ų-k’a ‘we dig it’ (Shaw 1980: 193) 24 

23 

Shaw includes the word-initial glottal stops that are inserted by rule; I omit these here. 

24 

This example is from the Teton dialect rather than the Santee dialect, which is the 

primary source of Shaw’s data. 

70

As seen below, vowel-initial stems take the ųk- prefix (Shaw 1980: 71; data are from the 

Teton dialect). 

(44) ųk-ípi ‘we go’ 

ųk-úpi ‘we are coming’ 

ųk-ų́spepi ‘we know how’ 

However, there is further evidence from the behavior of Ɂ-initial stems suggesting that the 

pattern of allomorphy seen here may not be suppletive. Ɂ-initial stems, rather than 

patterning with other consonant-initial stems in taking the ų- allomorph (as would be 

predicted under the ONSET-driven approach), take the ųk- allomorph, as shown below 

(Shaw 1980: 71). The glottal stop surfaces as glottalization of the /k/ of the prefix. 

(45) ųk-’į́yąka ‘we run’ 

ųk-’į́pi ‘we wear it’ 

ųk-’ų́pi ‘we use it’ 

ųk-’ópi ‘we shoot it’ 

In these examples, the roots must have underlying initial glottal stops, because they differ 

minimally from the true vowel-initial stems in glottalizing the /k/ of the prefix. There are 

near-minimal pairs in the data above that illustrate this point; e.g., ųk-úpi ‘we are coming’ 

vs. ųk-’ų́pi ‘we use it’. 

We can account for the behavior of the glottal stop-initial stems if we assume that 

the 1du/pl marker has a single underlying form, /ųk-/. The /k/ is deleted before another 

consonant, with the exception of /Ɂ/ (this is Shaw’s analysis (1980:77)). The 

exceptionality of /Ɂ/ can be handled in one of at least two different ways. The first would 

be to assume that the rule deleting /k/ is triggered by a consonant with a consonant-place 

71

node. This would prevent /Ɂ/ from triggering the rule if we assume that /Ɂ/ has no place 

features. Alternatively, we could capture the pattern using rule ordering or its equivalent 

translation into OT constraints, the idea being that the fusion of the glottal stop with the 

/k/ of the prefix prevents the deletion of /k/, either because fusion applies ‘first’ or 

because it is preferenced by the constraint ranking. 25 

We have seen a number of examples here in which allomorphy is conditioned by 

the C/V distinction and results in optimal syllable structure (in terms of eliminating codas 

or consonant clusters, or providing syllable onsets). In every case, it is the stem that 

conditions allomorphy in an affix, and the stem segment that conditions the allomorphy is 

immediately adjacent to the affix in question. 

2.1.2.4 Syllable contact constraint satisfaction 

Another phonological factor that we might expect to condition PCSA is syllable 

contact. In many languages (see Gouskova 2004 for examples and discussion), codas are 

required to have higher sonority than a following onset, or at least not to exceed the 

following onset in sonority by more than a specified amount. The Syllable Contact Law 

(SCL) was formulated for this purpose (see Davis 1998, Rose 2000b for recent 

25 

McCarthy and Prince (1993a) discuss this affix in Dakota, but focus on a set of stems 

including ali ‘climb’ and manų ‘steal’ that are said to be lexically specified as taking 

infixed rather than prefixed person marking (Shaw 1980 does not discuss these stems). 

McCarthy and Prince cite two examples, ųk-ali ‘we (du.) climb’ and ma-ų-nų ‘we (du.) 

steal’ (1993a: 114) in support of the claim that the use of prefixed ųk- in ųk-ali is 

determined by ranking the phonological constraint ONSET over the morphological 

constraint ALIGN-ROOT. Though it is implied that these constraints are also responsible 

for the selection of allomorphs, the selection follows naturally from the generalization 

that vowel-initial stems take ųk- while consonant-initial stems take ų-, and it is really the 

placement of the affix (prefix vs. infix) that is relevant in McCarthy and Prince’s 

discussion. 

72

implementations). Most recently, Gouskova (2004) has proposed a hierarchy of relational 

constraints to account for the range of syllable contact effects found in a number of 

languages. Regardless of the specific implementation, many analyses make use of 

phonological constraints to account for syllable contact phenomena, and therefore the P 

>> M model predicts that we should find cases of PCSA driven by these constraints. Two 

possible examples of this were found in the present survey, both in Martuthunira (Pama- 

Nyungan, Australia; Dench 1995), though as will be discussed, these may not be true 

cases of suppletive allomorphy. 

The first example is found in the accusative and genitive suffixes in Martuthunira, 

which exhibit allomorphy based on the root-final segment when marking common nouns 

(Dench 1995: 63). Following a nasal, the accusative and the genitive take the form -ku 

((46)a). Following a lateral or rhotic, both suffixes take the form -yu ((46)b) (the only 

consonants allowed at the end of a word are nasals, laterals, and rhotics (Dench 1995: 

30); this apparently holds for roots as well). With vowel-final stems, the accusative 

lengthens the final vowel, while the genitive is marked by -wu ((46)c) (examples are from 

Dench 1995: 91 except where page numbers are given). 

(46) a. tharlwan-ku-wuyu 

tame-ACC-SIDE 

‘on the tame side’ (98) 

b. tharratal-yu 

bird(sp.)-GEN 

‘tharratal’s’ 

c. nhartu-u nganaju-wu-lu yaan-tu 

something-ACC 1SG.OBL-GEN-EFF spouse-EFF 

‘(about) something’ (98) ‘...by my wife’ 

The alternation involving consonant feature assimilation is the -ku ~ -yu alternation 

conditioned by nasal vs. liquid-final roots, respectively. According to Dench, ‘a number 

73

of morphemes’ exhibit lenition of k to y after a stem-final lateral or rhotic (1995: 38), but 

the only morpheme that is cited as exhibiting the alternation aside from the accusative 

and genitive is the ‘body-noise’ verbal derivational suffix, which takes the form -karri 

after nasals and -yarri after laterals and rhotics (1995: 38). It is not clear what form the 

body-noise suffix takes after vowels since the suffix is not fully productive. 

Another, related example is the ‘belonging’ suffix in Martuthunira, which takes 

the form -ngura when marking proper nouns, pronouns, and demonstratives; otherwise, 

the suffix -kura is used after nasals ((47)a), and -wura after a vowel, rhotic, or lateral 

((47)b) (Dench 1995: 64). Some examples are given below (Dench 1995: 91). 

(47) a. ngurnu-ngura parnparn-kura 

that.OBL-BELONG budgerigar-BELONG 

‘...belonging to that budgerigar...’ 

b. yirna-tharra-wura-a kanparr-wura 

this.OBL-DU-BELONG-ACC spider-BELONG 

‘...the ones belonging to these two’ ‘...of the spider’s...’ 

The relevant alternation here is the k ~ w alternation between nasals vs. rhotics and 

laterals. This parallels the situation for the accusative, genitive, and body-noise suffixes 

described above, except that in this case k alternates with w instead of y. The belonging 

suffix is apparently the only suffix to exhibit this particular alternation. 

Considering only the allomorphs that occur after consonants here, both of these 

examples from Martuthunira can be analyzed in terms of syllable contact. In both cases, a 

possible generalization is that sonority is not allowed to increase by three or more points 

on the syllable contact scale as formulated by Gouskova (2004: 211). Gouskova (2004: 

208) uses the sonority scale laid out by Jespersen (1904: 191): glides > rhotics > laterals 

> nasals > voiced fricatives > voiced stops > voiceless fricatives > voiceless stops. 

74

Applying this scale to the accusative/genitive suffix, we see that the sequence n.k is 

permitted, and r.y is permitted, while n.y does not occur. On Gouskova’s scale, these 

sequences would correspond to a change in sonority of -4, +1, and +3, respectively. If the 

grammar does not allow an increase of three points between the coda and following 

onset, then we can account for why n.y is not allowed and why the accusative/genitive 

surfaces as -ku instead of -yu after nasals. The same principle can be applied to the 

belonging suffix allomorphs -kura and -wura. The sequence n.w would result in a change 

of +3, which is not allowed, and therefore the -kura allomorph is used after nasals. We 

could use the relational constraints used by Gouskova (2004) to account for this 

phenomenon in a P >> M analysis. 

However, a P >> M analysis may not be necessary. It is unclear whether glide ~ k 

alternations involve suppletive allomorphy. If we assumed that the glide-initial allomorph 

were the underlying form in each case, then we could write a single phonological rule or 

set of constraints that could change a glide into [k] after a nasal, either because of the 

syllable contact principles discussed here, or perhaps because the feature [-continuant] 

spreads from the nasal to the glide. This would account for all of the forms of the 

belonging suffix and also the postconsonantal allomorphs of the accusative/genitive 

suffixes, leaving only the postvocalic forms of the accusative/genitive to account for. We 

may need to posit separate postvocalic allomorphs /-V/ and /-wu/ for the accusative and 

genitive suffixes, respectively. These would probably be considered suppletive 

allomorphs, and note that their distribution does not appear to be optimizing; this 

anticipates some examples to be discussed in the following section in which entire 

patterns of suppletive allomorphy are non-optimizing or even detrimental in terms of 

75

well-formedness. 

In this section, we have seen two examples from Martuthunira in which 

allomorphy seems to result in optimized syllable contact configurations. As discussed, 

though, these cases may not necessarily involve suppletive allomorphy, and therefore we 

cannot be sure that we have a case of PCSA driven by syllable contact considerations. 

2.1.2.5 Non-optimization 

Though the examples in the preceding sections have all been cases where the 

distribution of allomorphs can be construed as phonologically optimizing or at least could 

arguably be accounted for using phonological constraints that have already been 

proposed in the literature, this is by no means always the case (see, e.g., Paster 2005a on 

non-optimization in ‘syllable-counting allomorphy’, a type of PCSA to be discussed in 

chapter 4; see also Bye to appear for several examples and discussion of non-optimizing 

PCSA). In many instances, PCSA is neutral or even detrimental with respect to the 

overall well-formedness of the word. Such ‘non-optimizing’ examples are discussed in 

this section. An important point should made here, which is that I do not claim that there 

are no phonological constraints that have ever been proposed or will be proposed that 

could possibly account for the allomorphy exhibited in these examples. Presumably there 

is some P >> M account that could be proposed for every example if we allow the use of 

stipulated, language-specific P constraints. However, as will become clear in the 

presentation of examples, characterizing these examples in terms of optimization would 

be to mischaracterize them. They do not appear to relate to any universal well- 

76

formedness generalizations, nor does any example appear to follow from the general 

phonotactics of the language. 

Under P >> M, we do not expect non-optimizing PCSA to be very common since 

their phonological counterparts, ‘crazy rules’ (Bach and Harms 1972), though attested, 

are not particularly common and do not result from universal phonological constraints. 26 

In this section, we will see many examples of PCSA that are non-optimizing. In an OT 

account using P >> M, each of these cases would require a language-specific, stipulated 

constraint to create the pattern of allomorphy. This is also true of crazy rules, but the 

PCSA examples are more problematic because it seems that non-optimizing examples of 

PCSA are more frequent relative to the total number of examples of PCSA than crazy 

rules are to the total number of attested phonological rules. 

There are 12 examples to be discussed in this section. The first set of examples 

includes cases where PCSA is conditioned by the C/V distinction but the pattern 

nonetheless does not seem to involve syllable structure optimization. These examples can 

be contrasted with those presented in §2.1.2.3, where some instances of C/V conditioned 

allomorphy seemed to optimize syllables by avoiding consonant clusters and/or vowel 

hiatus. The existence of the examples to be discussed here demonstrates that PCSA can 

refer directly and arbitrarily to the C/V distinction, and that this does not always reduce to 

syllable structure considerations. As will be seen, in every case, the relevant consonant or 

vowel of the stem is at the edge of stem where the affix attaches. 

26 

In fact, if we assume a version of OT in which constraints are universal and cannot be 

language-specific, then under P >> M we would never expect to find non-optimizing 

PCSA (though a version of OT with universal constraints would have difficulty handling 

crazy rules as well, so presumably whatever ‘patch’ were applied to the model to account 

for crazy rules could also account for non-optimizing PCSA). 

77

One example of apparently non-optimizing allomorphy conditioned by the C/V 

distinction is found in Turkish (Lewis 1967). The Turkish passive is marked by -n 

following a stem ending in a vowel ((48)a) or in /l/ ((48)b), and by -l elsewhere ((48)c). 

Examples are shown below (Underhill 1976: 332). 

(48) a. ara-n- ‘be sought’ 

de-n- ‘be said’ 

oku-n- ‘be read 

b. çal-ɪn- ‘be struck’ 

bil-in- ‘be known’ 

c. kullan-ɪl- ‘be used’ 

yor-ul- ‘be tired’ 

kaybed-il- ‘be lost’ 

We can treat the alternation between -C and -VC forms of each allomorph as epenthetic, 

so we are left with two underlying forms, -n and -l, whose distribution is determined 

primarily by the C/V distinction with no apparent optimizing effect on syllable structure 

or any other aspect of well-formedness. This is an example in which C/V conditioned 

PCSA cannot be driven by syllable structure constraints. The overall pattern of 

allomorphy in the passive suffix is not entirely arbitrary, however, since the use of -n 

with stems ending in /l/ is dissimilatory. Turkish does not exhibit sonorant dissimilation 

elsewhere in the system (Kornfilt 1997: 510-511), so this aspect of the allomorphy could 

be suppletive as well, though the dissimilation could also be said to result from an item- 

specific dissimilation rule. Regardless, the pattern of allomorphy involving the distinction 

between consonant- and vowel-final stems does not appear to be optimizing in any way. 

An interesting observation about this pattern is that, as with the Turkish causative 

suffixes described in §2.1.2.3, the allomorphy prevents sequences of identical affixes, as 

discussed by Haig (to appear). In Turkish words with a ‘double passive’, the two passive 

allomorphs will never have the same shape, since the attachment of one variant to form 

78

the first passive automatically conditions the use of the other variant to form the double 

passive. When the first passive is marked by -n, then the second passive is marked by -l, 

and vice versa. Examples are from Underhill (1976: 332), except where noted. 

(49) de-n-il- ‘say (double passive)’ 

iste-n-il- ‘want (double passive)’ 

döv-ül-ün- ‘hit (double passive)’ (Haig to appear: 5) 

It is possible that avoidance of repeated morphemes had something to do with the 

historical origins of the modern allomorphy pattern, but just as in the causative example 

discussed earlier, this cannot be responsible for the overall pattern. Using a constraint 

against repeated morphemes would account only for the distribution of allomorphs in 

words with double causatives and would not explain the pattern as a whole. Thus, we still 

need either an item-specific rule of dissimilation or a subcategorizational requirement to 

account for the use of -n with l-final stems. 

An unusual case of non-optimizing C/V-sensitive PCSA is found in Mafa 

(Chadic, Cameroon). Le Bleis and Barreteau (1987) report that in the verbal suffix 

indicating ‘le directionnel de rapprochement’ occurs as -ká when preceding a word that 

begins with a consonant, and as -káɗá elsewhere (Le Bleis and Barreteau 1987: 108-109; 

English translations mine). 

(50) á mbálə-ká kəda 

il-INACC chasser-RAPPR chien 

‘il court après le chien vers nous’ 

[‘he runs after the dog towards us’] 

m ɓálə-ká yim (á duwzlak) 

(no interlinear glosses provided) 

‘il a puisé de l’eau (l’a versée dans la jarre) et l’a rapportée’ 

[‘he drew some water (poured it in the jar) and brought it back’] 

79

n tə́v-káɗa aa gírzhe 

il-ACC monter-RAPPR sur rocher 

‘il est monté sur le rocher (qui se trouve entre l’endroit d’où il vient et celui où se 

trouve le locateur)’ 

[‘he climbed onto the rock (which is located between the location from which he 

came and the location of the speaker)’] 

kalədə-káɗá 

tomber.CAUS-RAPPR 

‘jette-le vers moi!’ 

[‘throw it towards me!’] 

This example is particularly interesting in that the allomorph distribution seems to be 

determined by a property of a separate word from the one to which the affix in question 

belongs. No other examples having this property were revealed by the survey. Even in the 

case of English a/an discussed in §2.1.2.3, the morpheme that is affected is a clitic and 

therefore part of the same phonological word as what follows despite being a separate 

morphosyntactic word. We do not have evidence for the directional morpheme in Mafa 

being a clitic; it is not described as having other properties that we could use to argue that 

it is a clitic rather than an affix (for example, it seems always to occur immediately after 

the verb stem rather than allowing other words to intervene). However, perhaps with 

more data we could argue that this morpheme is, in fact, a clitic. In any case, since it is 

the only apparent example exhibiting conditioning by a separate phonological word, we 

do not have enough evidence to conclude that PCSA in an affix can refer to a property of 

a separate phonological word. 

Aside from the problem of the word-external conditioning, another point to be 

made regarding the Mafa example is that there is no apparent functional motivation for 

the distribution of the allomorphs. Both allomorphs are vowel-final and should syllabify 

equally well with the following word regardless of whether it is consonant- or vowel- 

80

initial. Therefore, this is another case of PCSA whose effect is neutral rather than 

optimizing. 

Another case of PCSA conditioned by consonants vs. vowels is found in 

Winnebago (Siouan, Wisconsin; Lipkind 1945). Here, the declarative is marked by the 

suffix /-nə̃/ after vowel-final roots and by /-səńə̃/ after consonant-final roots, as shown in 

the examples below (Lipkind 1945: 33). No examples of the postconsonantal allomorph 

are given. 

(51) wa-jəra sepsə́-nə̃ nãná tira-je-nə̃ 

boat-DEF black-DEC tree-DEF move-standing.position-DEC 

‘the boat was black’ ‘the tree is growing’ 

wasí-nə̃-gi nã-wãnã-nə̃ 

dance-COND-SUBORD sing.1sg-COND-DEC 

‘if he danced I would sing’ 

Despite the fact that we can recognize the post-vocalic allomorph in the shape of the 

post-consonantal allomorph, this is a clear case of suppletion because if these allomorphs 

reduced to a single underlying form, we would have to write a highly unnatural and 

unusual rule inserting or deleting [sə́]. It is possible that /sə́/ could be a separate 

morpheme, perhaps an empty morpheme that is added to consonant-final roots to form 

stems. However, even if this is true, there is no apparent motivation for its presence in 

that environment. Regardless of whether /sə́/ is a separate morpheme or just a part of the 

/-səńə̃/ allomorph, its distribution is still arbitrary and non-optimizing since it does not 

affect syllable structure or any other apparent factor in the well-formedness of words in 

Winnebago. The condition on allomorphy (the C/V distinction) does not seem to relate to 

the differences between the allomorphs (the most salient of which is their syllable count). 

81

Another case in which PCSA is conditioned by the C/V distinction but does not 

result in syllable structure optimization is found in Jivaro (Jivaroan, Ecuador; de Maria 

1918). In Jivaro, the negative is marked by one of two suffix allomorphs, as follows (de 

Maria 1918: 9). Among nouns, adjectives, and adverbs, all stems ending in a consonant 

take the negative suffix -cha. Also, ‘some’ verbs take -cha (this is not addressed further, 

so apparently verbs have lexically specified negative forms). Among vowel-final stems, 

there is a distinction between monosyllabic and polysyllabic stems. All polysyllabic 

vowel-final stems take -chu. Among monosyllabic stems, those ending in /e/, /i/, or /a/ 

take -cha. All other stems are said to take -chu (though no examples of monosyllabic 

stems not ending in /e/, /i/, or /a/ are given). The distribution is shown in the examples 

below (de Maria 1918: 9). 

(52) Apár-cha ‘no padre’ 

Puéngar-cha ‘no bueno’ 

Aho-chu ‘no él’ 

Tuná-chu ‘no tonto’ 

Ví-cha-itijae ‘no soy yo’ 

Though this is a complicated pattern, the element of PCSA here seems to be that 

consonant-final stems take the -cha allomorph and vowel-final stems take the -chu 

allomorph (though, as to be discussed, the vowel of this suffix undergoes a vowel 

harmony rule that complicates the picture). Looking only at the suppletive distribution of 

allomorphs based on consonant- vs. vowel-final stems, this is an example of non- 

optimizing allomorphy because there is no apparent way in which the distribution 

improves the overall well-formedness of words containing either type of stem. What is 

interesting about this case, as in the Winnebago example described above, is that 

82

although the allomorphs are distributed based on whether the stem is consonant- or 

vowel-final, the pattern ends up having no effect on syllable structure. 

The distribution of -chu vs. -cha on vowel-final stems can be derived by rule from 

underlying /-chu/. One thing to notice about the distribution of these allomorphs is that 

-cha occurs when the stem has a front vowel (assuming that /a/ is a front vowel) and 

when the suffix can form a single disyllabic foot with the stem (i.e., when the stem is 

monosyllabic). This suggests that there could be a foot-bounded fronting harmony 

process responsible for the change of /u/ to [a] in the suffix. When the stem has two 

syllables, even if the stem-final syllable has a front vowel, the suffix vowel surfaces as 

[u] because the stem forms its own foot to the exclusion of the suffix, and therefore the 

suffix vowel /u/ is outside the domain of fronting harmony. The one troubling fact about 

this proposed harmony rule is that we would expect it to change /u/ to its high, front 

counterpart [i] rather than the low vowel [a]. However, given that the process appears to 

be so regular and that it accounts for a pattern of allomorphy that is otherwise mystifying, 

we should accept this one complication to the rule. There is further reason to assume that 

this is a regular rule of the language, which is that it also applies to the genitive suffix, to 

be described below. 

The genitive suffix in Jivaro exhibits allomorphy that is very similar to that of the 

negative suffix, except that there are four surface allomorphs. The generalization is as 

follows (de Maria 1918: 6). All stems ending in a consonant take the -na suffix. All stems 

ending in /e/ take -ña. Among stems ending in /i/, polysyllabic stems take -ñu, while 

monosyllabic stems take -ña. All other stems (that is, stems ending in any vowel except i 

83

or e) take -nu. 27 Examples are shown below (de Maria 1918: 6). 28 

(53) Yátzúm Yatzúm-na 

Gé Gé-ña 

Nucurí Nucurí-ñu ‘your mother’s’ 

Vi Vi-ña ‘my’ 

Apa Apá-nu ‘father’s’ 

Aho Aho-nu ‘his’ 

Núcu Nucú-nu 

The number of underlying forms can be reduced if we assume that the n ~ ñ alternation 

results from palatalization following i or e. 29 This leaves us with two suppletive 

allomorphs, distributed as follows: stems ending in a consonant take -na. Stems ending in 

/e/ take -na. Among stems ending in /i/, those with one syllable take -na. All other stems 

take -nu. We can simplify this, as we did above for the negative, as follows. Consonant- 

final stems take -na, and vowel-final stems take -nu. The same vowel fronting harmony 

described above applies to the suffix vowel when the stem vowel is front and when the 

suffix syllable can form a foot with the stem, resulting in -na. 

There are two minor problems with this. The first is that, unlike for the negative, 

we have examples of the genitive suffix combining with trisyllabic stems. Under the 

analysis assuming a foot-based harmony domain, we would expect trisyllabic forms to 

behave like monosyllabic stems, since the first two syllables of the stem can form a 

disyllabic foot, leaving the stem-final syllable to form a foot along with the suffix, 

27 

In the very closely related language/dialect Achuar, the genitive allomorphs are -nau 

and -nu (Fast Mowitz et al 1996: 31). Though the distribution of these allomorphs is not 

discussed, the two examples given (áints-nau ‘de un hombre’ and nuwá-nu ‘de una 

mujer’) are consistent with the distribution in Jivaro if we assume that Achuar -nau 

corresponds to Jivaro -na. 

28 

Examples are not glossed; the glosses here are from Fast Mowitz et al (1996) and Jintia 

et al (2000). 

29 

Fast Mowitz et al (1996: 15) document a rule in Achuar in which some consonants are 

palatalized following /i/. 

84

thereby creating a domain in which the harmony rule can apply. However, as seen in the 

example Nucurí-ñu above, this is not what happens. Instead, trisyllabic stems behave like 

disyllabic stems in not triggering the vowel harmony rule. A possible explanation for this 

is that perhaps Jivaro allows trisyllabic feet in cases where this would allow the right 

edge of a stem to be aligned with the right edge of a foot. Thus, the example Nucurí-ñu 

would be footed as (Nucurí)Ft-ñu, thus explaining the failure of the suffix vowel to 

undergo harmony. The second problem is that, unlike the in negative example described 

above, the genitive suffix vowel is claimed not to surface as [a] when the stem vowel is 

/a/. The only genitive example provided that has a stem-final /a/ happens also to have a 

disyllabic stem, so the suffix vowel would not be expected to undergo harmony in this 

example anyway. With no further examples, I will assume that monosyllabic stems with 

final /a/ do take the -na allomorph, since otherwise the application of the harmony rule to 

the genitive suffix exactly parallels its application to the negative suffix. 

The PCSA seen in both the genitive and the negative suffixes in Jivaro has been 

argued to reduce to allomorphy conditioned by whether the stem ends in a consonant or a 

vowel. As in the Winnebago example discussed above, this example is of interest because 

unlike some other examples involving conditioning of this type, the pattern seen here 

does not optimize syllable structure. Therefore, the distribution based on the C/V 

distinction is not simply an artifact of syllable optimization. If the allomorphy optimized 

syllables, we might have been able to state the generalization in terms of syllable types 

and structures such as ‘onset,’ ‘nucleus,’ and ‘coda’ instead of C and V. Since this is not 

the case, we must conclude that the mechanism responsible for PCSA is able to refer 

directly to C and V. 

85

Another example of C/V conditioned allomorphy is found in Haitian Creole (Hall 

1953, Klein 2003). In this language, there is a determiner whose form varies between -a 

and -la in a pattern that is the exact opposite of what would be expected if the pattern 

optimized CV syllable structure: -a occurs after vowel-final stems while -la occurs after 

consonant-final stems. Some examples are shown below (Hall 1953: 32). 

(54) panié-a ‘the basket’ pitit-la ‘the child’ 

trou-a ‘the hole’ ãj-la ‘the angel’ 

figi-a ‘the face’ kay-la ‘the house’ 

chẽ-ã ‘the dog’ madãm-lã ‘the lady’ 

In this example, if regular phonological well-formedness constraints were responsible for 

the allomorphy, we would have expected the opposite distribution, since this would avoid 

coda consonants and vowel hiatus. In this case, then, the allomorphy is not only non- 

optimizing, but it is ‘perverse’ in the sense that it makes words less phonologically 

optimal with respect to syllable structure (see also Bye to appear for discussion of this 

example). 30 

We could analyze this as non-suppletive allomorphy resulting from deletion of /l/ 

between vowels. However, this would not be a general rule of the language, and it is also 

30 

Bonet et al (in press) propose an analysis of this example in terms of P constraints on 

syllable contact and alignment of stems with syllable boundaries. Such an analysis 

requires proposing a PRIORITY constraint stipulating that the /a/ allomorph takes 

precedence over the /la/ allomorph. The analysis requires many constraints and is 

arguably much more complex than a subcategorization account, but Bonet et al reject the 

subcategorization approach to this example in footnote 11 on the grounds that ‘… just 

stating the distribution misses a phonological generalization; the choice is not random but 

systematic, and the systematicity is directly related to phonological information.’ If 

‘systematic’ here is meant to mean ‘phonologically natural’, then it is not clear that the 

statement is true at all, since the constraint-based analysis crucially depends on a 

stipulated priority relationship between allomorphs. If, instead, ‘systematic’ merely 

means ‘able to be captured via reference to a consistent phonological generalization,’ 

then the subcategorization approach is just as well-equipped to capture the generalization 

as is the account proposed by Bonet et al. 

86

an unnatural rule since it results in VV sequences and since V_V is an ideal environment 

for a consonant to be perceived and produced. Thus, even if this were analyzed as a 

‘crazy rule’ rather than PCSA, we would still have difficulty accounting for it using 

regular, natural phonological constraints. 

Another potential alternative analysis is a ‘ghost consonant’ analysis as discussed 

for the Midob example in §2.1.2.3 (in fact, analyses of Haitian involving ghost 

consonants (or ‘floating segments’) are proposed by Cadely 2002 and Nikiema 1999). 

The determiner could be represented as /-(l)a/ with a ‘latent’ /l/ consisting of consonant 

features not attached to any root node (following the representation advocated by Zoll 

1996: 48) that would only surface when a root node is inserted for its features to associate 

to. One problem for such an analysis is that we would have to explain why a root node is 

inserted in the context C_V but not in the context V_V; the pattern would still be non- 

optimizing even if it were not analyzed as suppletive. Another problem for a ‘ghost 

consonant’ analysis of this phenomenon is that when the stem has a final [+ATR] vowel, 

a glide is inserted between the stem and suffix (Klein 2003), as seen below. 

(55) pape[j]-a ‘the paper’ 

bato[w]-a ‘the boat’ 

lapli[j]-a ‘the father’ 

tu[w]-a ‘the hole’ 

In an OT analysis involving a latent suffix-initial /l/, we would have to explain why a 

different consonant gets inserted in this context. Since the choice of the inserted glide is 

predictable from the roundness of the stem-final vowel, we would not want to supply the 

inserted glide with underlying consonant features. However, if the glide insertion rule 

simply inserted a root node, then we might expect the root node to take on the features of 

the latent /l/ rather than taking on features from the preceding vowel, since this would 

87

avoid the violation of MAX that would be incurred when the latent features of /l/ failed to 

surface. 

Two points should be made regarding glide insertion in the Haitian Creole 

examples. First, there is the question of whether this example can be claimed to be ‘non- 

optimizing’, given that these examples with the -a suffix do end up having well-formed 

CV syllables on the surface. Since glide insertion applies only to stems ending in a 

[+ATR] vowel, the distribution of allomorphy still does create VV sequences in words 

where the stem ends in a [-ATR] vowel. Therefore, the result of the allomorphy is that 

some words are indeed less well-formed, with respect to the constraint ONSET, than they 

would be with the opposite distribution of allomorphs. A second point is that in words 

which undergo glide insertion, the condition that determines allomorphy (C- vs. V-final 

stem) is rendered opaque, since the suffix in these words does end up being preceded by a 

consonant rather than a vowel. This suggests that glide insertion operates on the output of 

suffixation, rather than applying at the same time as the suffix allomorph is selected. In 

§2.1.2.6, I discuss more examples in which conditions on allomorph distribution are 

rendered opaque by phonological processes. 

Some examples of non-optimizing PCSA involve conditioning by features of a 

consonant or vowel in the stem. As with the C/V-conditioned cases described above, in 

these examples the triggering segment of the stem is always at the same edge at which the 

affix attaches. In these examples, unlike in the assimilatory and dissimilatory examples 

presented in §2.1.2.1 and §2.1.2.2, the features that condition allomorphy do not seem to 

be related to features of the suppletive allomorphs. This makes it difficult to account for 

these examples in terms of any well-known phonological constraints. 

88

Harris (1979: 288-289) discusses an interesting case of allomorphy in Spanish 

that is conditioned by vowel quality. According to Harris, the imperfective aspect is 

marked by -a following /i/, and as -ba elsewhere. Examples are provided below (Harris 

1979: 288). 

(56) viví-a ‘lived’ pasá-ba ‘passed’ 

caí-a ‘fell’ tomá-ba ‘took’ 

Harris (1979: 289) proposes a rule of b → Ø after /i/ and a morpheme boundary, but since 

the rule would only apply to the imperfective suffix and since this rule is apparently not 

phonetically motivated, it seems more reasonable to posit two separate underlying forms, 

/-a/ and /-ba/. This does not seem to be a common rule type cross-linguistically, and there 

is no immediately apparent phonetic motivation for deleting /b/ after a high vowel. 

A second example in which PCSA is triggered by stem-final vowel features is 

found in Yucunany Mixtepec Mixtec (Otomanguean, Mexico; Paster and Beam de 

Azcona 2005). This language exhibits another pattern of allomorphy that does not seem 

to maximize phonological well-formedness (though as to be discussed, depending on 

one’s perspective, the pattern could be said to be optimizing in a different way). In this 

language, the distribution of allomorphs of the third person singular familiar (subject and 

possessor) is determined by the final vowel of the root. The allomorphs, -à and -ì, occur 

in place of the final vowel of the verb or noun phrase and are selected based on the 

quality of the final vowel. When the final vowel is [i], the third person singular familiar is 

marked using the à allomorph; with all other final vowels, the -ì allomorph is used. All 

native roots end in vowels, but there are some consonant-final Spanish loanwords that 

take -ñaà; this is allomorph is probably better viewed as being specific to loanwords 

89

ather than conditioned by the root-final segment. Examples of the -à allomorph are given 

below (Paster and Beam de Azcona 2005: 74; underlining indicates nasalization). 

(57) sì’i ‘leg’ sì’aà ‘his leg’ 

kachìí ‘cotton’ kachìáà ‘his cotton’ 

tzí’ì yù ‘I am dying’ tzí’à ‘she is dying’ 

Occurrences of the -ì allomorph are shown below (Paster and Beam de Azcona 2005: 74). 

(58) sàmá ‘clothing’ sàmíì ‘his clothing’ 

vàá’a ‘bad’ vàá’ì ‘it is bad’ 

tá’a ‘relative’ tá’ì ‘his relative’ 

nda’á ‘hand’ nda’íì ‘her hand’ 

ma tzá’nu ‘grandmother’ ma tzá’nì ‘her grandmother’ 

kù’ù ‘woman’s sister’ kù’ì ‘her sister’ 

This pattern does not make more well-formed words in any obvious way. It does, 

however, have the effect of preventing homophony with the plain form, since plain stems 

ending in L-toned [i] would be homophonous with their 3sg familiar forms if the -à 

allomorph were not available. In addition, the use of -à prevents homophony with the 1sg 

form of underlyingly H-M and M-M roots ending in [i], which would otherwise take the 

same form in the 3sg familiar. Both allomorphs consist of a single vowel that takes the 

place of the root-final vowel, but the pattern is not derivable via a regular phonological 

rule of the language. I conclude that the allomorphy is suppletive, and though it 

apparently does not optimize the phonological well-formedness of individual words, one 

might argue that it maximizes distinctness among morphologically related words. If one 

is willing to posit phonological constraints enforcing non-homophony within paradigms, 

then perhaps this example could be analyzed along these lines using the P >> M model. 

90

The survey also revealed some examples in which allomorphy is triggered by 

features of stem-peripheral consonants. For example, the locative suffix in Martuthunira 

(Dench 1995) exhibits non-optimizing allomorphy based on both the root-final segment 

and on mora count. Bimoraic stems with a final vowel take -ngka, stems with three or 

more morae with a final vowel take -la, stems with final n take -ta, stems with final rn 

(apico-postalveolar nasal) take -rta (rt is an apico-postalveolar stop), stems with final ny 

or nh (lamino-palatal or lamino-dental nasal) take -tha, (th is a lamino-dental stop) and 

stems with a final rhotic or lateral take -a (Dench 1995: 64). The effector suffix has the 

same shape as the locative in all of these contexts, except that in each case, the effector 

suffix has /u/ rather than /a/. The sensitivity to mora count seen here will be taken up in 

chapter 4; for our present purposes, we will focus on the allomorphy that relates to 

whether the stem ends in a nasal or a liquid. Examples are given below (with page 

numbers from Dench 1995 in parentheses), though no examples of the -rta or -tha form 

are provided. 

(59) panyu-ngka-a yartapalyu-la 

good-LOC-ACC other.group-LOC 

‘in a better (tree)’ (73) ‘...out of that other mob’ (75) 

kalyaran-ta-a kuyil-a 

tree-LOC-ACC bad-LOC 

‘in a tree’ (73) ‘on that bad (hill)’ (94) 

The aspect of this pattern of interest here is the -ta ~ -rta ~ -tha ~ -a alternation in the 

locative suffix, and the corresponding -tu ~ -rtu ~ -thu ~ -u pattern of the effector suffix. 

It seems possible that /t/ is not part of the locative or effector suffixes, but some kind of 

stem-former or empty morpheme that can be used with the locative and effector suffixes; 

regardless of whether it belongs to the suffixes or the stem, the alternation exhibited in 

91

the consonant in this position should be explained. Dench (1995: 41) proposes an 

assimilation rule to account for the first three of each of these allomorph sets, leaving us 

with two basic locative allomorphs, /-ta/ and /-a/, and two basic effector allomorphs, /-tu/ 

and /-u/. Assuming Dench’s assimilation rule, we still need to explain the -V form of the 

suffixes after a rhotic or lateral. This could be handled via a rule specific to this affix that 

deletes /t/ following a liquid, but the rule would be unmotivated since if anything, liquids 

seem to make better codas than other consonants. The result of this case of PCSA can 

therefore be said to be phonetically unnatural. Here the feature that distinguishes nasals 

from other sonorant consonants (whether it is [±liquid], [±nasal], [±continuant], or some 

combination of these is not important here) causes an allomorph to be used that has no 

initial consonant; this is parallel to a phonological rule deleting the initial consonant of 

the suffix after a [liquid] consonant. We could analyze this in terms of a constraint such 

as *LT banning voiceless consonants after a liquid, an extension of the *NT constraint 

(referred to as *NC̥) used by, e.g., Pater (1999). Paster (2004: 389) proposed a constraint 

*NT/lT banning voiceless consonants after nasals and /l/, but the use of that particular 

constraint for Martuthunira would not work since in that case we would expect the 

vowel-initial suffix allomorph to occur after nasals as well as liquids. We would therefore 

need a more specific constraint *LT referring only to liquids and not nasals; we would 

also need to explain why *NT was not also active in a language with *LT, since *NT 

effects seem to be common and are phonetically well-motivated, probably by the same 

mechanism that would be responsible for *LT. 31 It may thus be possible to account for 

this pattern if we propose a new constraint, but it is not at all clear that the overall pattern 

31 See Pater 1999 for further discussion of *NT. 

92

is optimizing. Certainly, this example is not like the other Martuthunira examples seen in 

the previous section that appear to be driven by syllable contact considerations. Since 

nasals are considered to be less sonorous than /l/, we might expect that the sequence l.t 

should be preferred over n.t and therefore that the t-initial suffix allomorphs should occur 

after /l/ and not after nasals. Viewed in this way, we might almost say that the pattern of 

allomorphy here is the exact opposite of what we would expect if it were optimizing. 

A second example of apparently non-optimizing PCSA conditioned by consonant 

features is found in Nishnaabemwin (Algonquian, Ontario; Valentine 2001), where the 

conjunct order third person suffix takes the form -g following a nasal-final stem and -d 

following other stems (except in some conjunct negatives) (Valentine 2001: 227). Some 

examples are provided below. 32 

(60) dgoshin-g ‘if/when ANsg arrives’ 

boodwe-d ‘if/when ANsg builds a fire’ 

dgoshnini-d ‘if/when ANsg (obv.) arrives’ 

There is no indication of a general rule of d → g elsewhere in the language, and there do 

exist nd sequences, including coda clusters, as in the examples below (nd sequences 

appear in bold; page numbers are from Valentine 2001). 

(61) bbaamaandwe ‘climb around’ (375) 

gaandnan ‘push X (with an instrument’ (427) 

naagdawenmind ‘(CONJ) ANsg is taken care of’ (563) 

niinwind ‘we/us/our’ (574) 

aanind ‘some’ (575) 

Although the -g and -d allomorphs have a similar phonological shape, an item-specific 

rule changing d to g after nasals would be unnatural since there is no apparent phonetic 

32 Valentine indicates (2001:74) that final ng reduces to [ŋ]; it is not clear whether this 

applies only to ng or to final nasal+stop sequences in general. A general postnasal stop 

deletion rule is not included in the discussion of phonological processes applying to 

consonants (Valentine 2001: 73-89). 

93

motivation for such a rule. This is an example where PCSA has a ‘neutral’ effect (i.e., no 

effect) on the well-formedness of words, since there is no phonological optimization in 

this pattern, and the distribution of allomorphs could just as easily be reversed or reduced 

to a single allomorph in all environments without any change in the number of violations 

of phonological well-formedness constraints. 33 

In Kashaya, the absolutive involves suppletion conditioned in part by the final 

consonant of the stem (Buckley 1994: 336, Oswalt 1960: 265). Buckley states that the 

allomorphs of the absolutive are distributed as follows: the suffix -w occurs after a vowel 

((62)a), -i occurs after n’ (and later undergoes [+round] insertion, surfacing as [u]) 

((62)b), and -ʔ occurs elsewhere ((62)c). 34 Examples are from Oswalt 1960: 265, except 

where noted. 

33 

Another example involving conditioning by a stem-final nasal is found in Northern 

Sotho (Bantu, South Africa; Kosch 1998). According to Kosch (1998: 35-36), Northern 

Sotho exhibits phonologically conditioned allomorphy in the past tense suffix -ile / -e. 

The past tense is marked with -ile (e.g. -rut-ile ‘teach-past’) except after certain verb 

roots, all of which end in a nasal (e.g. -em-e ‘stand-past’ and -bon-e [no gloss]). This 

suggests that the selection of the -e allomorph may be conditioned by the presence of a 

root-final nasal, though Kosch acknowledges that there are some cases in which a nasalfinal 

root takes -ile (e.g. -rom-ile ‘send-past’). It is unclear how many nasal-final roots 

take -ile, but Kosch implies that there are several. However, Kosch claims that the 

allomorphy was strictly phonologically conditioned at an earlier stage of the language. 

The allomorph distribution seems to have nothing to do with the conditioning factor, 

namely the presence of a final nasal. One possibility is that there is a dispreference for a 

high vowel following a nasal, though there does not appear to be any such dispreference 

exhibited elsewhere in the language. Thus, ‘pre-Northern Sotho,’ if it did indeed exhibit 

exceptionless phonological allomorph distribution, would constitute another example 

where allomorphy is not obviously optimizing and the nasal feature that triggers the 

allomorphy has no clear relationship with the shape of the allomorphs. 

34 

This is essentially the same as Oswalt’s (1960: 265) characterization except that 

Buckley’s n’ corresponds to Oswalt’s d, and Oswalt assumed that the absolutive 

allomorph for stems ending in this segment was /-u/ rather than /-i/ with [+round] 

insertion. 

94

(62) a. mo-ˑla-w ‘to run down’ 

bahcú-w ‘to jump’ (Buckley 1994: 61) 

ca-hcí-w ‘to sit down’ (Buckley 1994: 61) 

b. mo-laˑmed-u ‘to keep running down’ 

c. mom-’ ‘to run across’ 

c’ihwín-’ ‘get red hot’ (Oswalt 1960: 171, via Buckley 1994) 

dahal-’ ‘dig (a hole)’ (Oswalt 1960: 171. via Buckley 1994) 

Under an analysis without [+round] insertion, in which the underlying form of the 

absolutive suffix was /-u/, the u ~ w distribution could be accounted for by post-vocalic 

glide formation (u → w). However, [u] and [w] do not easily relate to [’], and there is no 

apparent reason why /d/ should behave differently from other consonants in not triggering 

a rule of u → ʔ. Even if we assume, with Buckley, that [d] corresponds to /n’/, we cannot 

explain its behavior as being related to its already being glottalized, since this segment 

still behaves differently from other glottalized segments, in that stems ending in other 

glottalized consonants do not take -u. In fact, according to Buckley (1994: 73), in general 

when a glottal stop occurs to the right of an already glottalized segment, the glottal stop 

simply deleted. The distribution of w vs. ’ might be argued to optimize syllable structure 

since adding /w/ to the end of a consonant-final stem would create a complex coda that 

would have to be repaired via epenthesis (Buckley 1994: 249), where as adding /’/ only 

glottalizes the consonant. However, the behavior of d-final stems is arbitrary and 

puzzling, and it seems that we cannot account for it based on any well-known 

phonological constraints. 

A final example of seemingly arbitrary, non-optimizing PCSA is found in Coptic 

(Kramer 2005, to appear). In this unusual example, allomorphy appears to be driven by 

the number of consonants in the stem. According to Kramer (2005: 14), the quality of the 

vowel that marks the stative in Coptic root and pattern morphology can be predicted 

95

ased on the number of consonants making up the root of the verb. Roots with a single 

consonant ((63)a) take eu, roots with two consonants ((63)b) take e, roots with three 

consonants ((63)c) take ɔ, and roots with four consonants ((63)d) or five consonants 

((63)e) take o. 35 Examples of stative forms are shown below (Kramer 2005: 4, 9, 14-15, 

23; capital letters indicate syllabic consonants, and periods indicate syllable boundaries). 

(63) a. seu ‘is sated’ 

tʃeu ‘is sown’ 

weu ‘has become distant’ 

ʃeu ‘is measured’ 

b. pet ‘has run’ 

keβ ‘has been made cool’ 

seh ‘is written’ 

ket ‘is built’ 

c. sɔ.tM ‘to be listened to’ 

kɔ.nS ‘to be stinky’ 

mɔ.səɁ ‘to be born’ 

sɔ.βK ‘to be few’ 

d. wS.ton ‘is broadened’ 

tN.ton ‘is likened’ 

sL.sol ‘is consoled’ 

ʃF.ʃof ‘has burrowed’ 

e. srM.rom ‘is dazed’ 

ʃtR.tor ‘is disturbed’ 

k j lM.lom ‘is twisted’ 

slK j .lok j ‘is made smooth’ 

This example is unusual and difficult to characterize. It is undesirable to allow the 

grammar to refer to the number of root consonants in selecting the stative vowel, and yet 

the pattern is straightforward, and we would like to capture it rather than say that the 

stative vowel quality is lexically determined (in which case the generalization about the 

35 

Though most quadriconsonantal stems and all quinquiconsonantal roots look like 

reduplicants on the surface, Kramer argues persuasively that they do not result from any 

synchronic reduplication process since the form and meaning of the apparent reduplicants 

is not consistent, and in some cases there is no shorter root that could have been the base 

for reduplication. 

96

number of root consonants would be lost). A possible way around this problem is to refer 

to the number and/or shape of syllables in the output form. Though it does not appear that 

this would allow us to explain the particular vowel that is used with each root type, we at 

least have a way to characterize the pattern without allowing the grammar to ‘count’ root 

consonants. In the case of monoconsonantal roots, the use of eu does make some sense 

phonologically, since the output form ends up consisting of a single syllable, and perhaps 

using eu allows the word to be bimoraic. Similarly, for the biconsonantal roots, we could 

say that the e allomorph occurs when the word will consist of a closed syllable. Likewise, 

the ɔ allomorph occurs in an initial open syllable, and the o allomorph occurs in a final 

closed syllable in words with two syllables. Note that except in the case of the diphthong 

used with monoconsonantal roots, there is no apparent principled relation between the 

quality of the vowel and the number of root consonants or the CV structure of the word. 

A particularly interesting contrast is between stems with two vs. four or five consonants. 

In both cases, the vowel ends up in a word-final CVC syllable, yet with biconsonantal 

roots it surfaces as e, while with quadri- or quinquiconsonantal roots it surfaces as o. 

Thus, although the pattern can be described in terms of syllable structure and syllable 

count (rather than consonant count), it does not optimize syllable structure, with the 

possible exception of words with monoconsonantal roots. We can therefore capture the 

pattern in phonological terms, but we cannot do this by using regular phonological well- 

formedness constraints in OT. This case is therefore not easily handled using P >> M. 

In this section, we have seen numerous examples of PCSA that do not maximize 

well-formedness, and some in which the distribution of allomorphs actually makes words 

less well-formed. As has been discussed throughout, these examples are problematic for 

97

the P >> M approach, because that approach seeks to account for phonological effects in 

morphology using the same constraints that account for purely phonological, non- 

suppletive allomorphy, and yet unlike phonology, which is usually ‘natural’ (though cf. 

Anderson 1981), many cases of PCSA are not at all natural. A model in which suppletive 

and non-suppletive allomorphy are analyzed in exactly the same way is incapable of 

accounting for this asymmetry. 

2.1.2.6 Opaque conditioning 

There is another class of examples of PCSA that are problematic for the P >> M 

model. This survey revealed a number of cases in which the phonological element that 

conditions the distribution of suppletive allomorphs does not appear in the surface form, 

rendering opaque the phonological condition on PCSA. Though the original version of 

OT was intended to be purely surface-based, later versions incorporated two-level 

constraints, sympathy, conjoined constraints, and other strategies to account for opaque 

phonological interactions. Thus, by incorporating some of these mechanisms, current OT 

is equipped to model opacity and, by extension, to model opaque interactions in PCSA. 

Given this, the P >> M model does predict the existence of opaque conditioning in 

PCSA. However, just as opaque phonological processes have been problematic for OT in 

making it necessary to continually add new mechanisms to the model, opaque conditions 

on PCSA are equally problematic. This survey revealed several such examples, to be 

discussed below. 

In Turkish (Lewis 1967), the 3rd person possessive suffix exhibits allomorphy 

whose conditioning is rendered opaque by a regular process of velar deletion. The suffix 

98

takes the form -I following a consonant-final stem and -sI following a vowel-final stem, 

as seen in the examples below (examples are from Aranovich et al 2005 and from Gizem 

Karaali, p.c.; note that vowel alternations are due to regular vowel harmony). 

(64) bedel-i ‘its price’ fire-si ‘its attrition’ 

ikiz-i ‘its twin’ elma-sɪ ‘its apple’ 

alet-i ‘its tool’ arɪ-sɪ ‘its bee’ 

A proponent of the P >> M approach might cite this as an example of syllable structure 

optimization since the distribution of allomorphs serves to avoid both vowel hiatus and 

syllable codas. However, as pointed out by Aranovich et al (2005), there are examples in 

which the distribution of allomorphs is rendered opaque by a regular phonological 

process. Turkish exhibits a regular process of Velar Deletion (see, e.g., Sezer 1981), 

which deletes /k/ in intervocalic position. This interacts with the third person possessive 

suffix allomorphy in an interesting way: underlyingly /k/-final stems take the /-i/ suffix 

allomorph since the stems are underlyingly consonant-final, but then because of the 

addition of the vowel suffix, the /k/ is in intervocalic position and is therefore deleted. 

Thus, as seen in (65), the interaction of PCSA and Velar Deletion produces the very 

vowel hiatus that the distribution of /-i/ vs. /-si/ was supposed to avoid. 

(65) açlɪ-ɪ ‘its hunger’ (cf. açlɪk ‘hunger’) 

bebe-i ‘its baby’ (cf. bebek ‘baby’) 

gerdanlɪ-ɪ ‘its necklace’ (cf. gerdanlɪk ‘necklace’) 

ekme-i ‘its bread’ (cf. ekmek ‘bread’) 

The explanation for this situation seems to be that suppletive allomorph selection 

takes place first in the morphology, and then the regular phonology of the language 

(including Velar Deletion) applies to the output of morphology, in some cases rendering 

the conditions on PCSA opaque. This is easy to capture in the subcategorization model, 

since we can say that /-i/ subcategorizes for stems that are consonant-final in the input, 

99

and then the output of morphology is the input to phonology. It is much more difficult to 

capture using P >> M because the constraints on allomorph distribution are not surface- 

true. The distribution cannot be stated using only output-based constraints in P >> M 

because that model does phonology and morphology together in parallel; it is crucial that 

morphology apply before phonology in this case. 

In Sa’ani Arabic (Egypt; Watson 2002), there is another pattern of PCSA in which 

the result appears to be neutral with respect to phonological well-formedness, and in 

which the conditions on PCSA are not transparent on the surface. In this language, some 

subject suffixes have phonologically conditioned allomorphy determined by whether the 

stem ends in a consonant or a vowel. According to Watson (2002: 178, 184), the perfect 

aspect 3pl masculine suffix is -ū after consonant-final verb stems and -aw after vowel- 

final verb stems. Some examples are shown below (Watson 2002: 185). Note that the 

stems taking -aw have underlying final vowels that are lost. 

(66) katab-ū ‘they (masc.) wrote’ ram-aw ‘they (masc.) threw’ 

libis-ū ‘they (masc.) wore’ mall-aw ‘they (masc.) filled’ 

Because of the deletion of stem-final vowels, the condition on allomorph distribution is 

lost on the surface in these examples. 

Also in Sa’ani Arabic, The perfective aspect 3sg feminine suffix has the form -at 

following consonant-final roots, and -it following vowel-final roots. Examples are given 

below (Watson 2002: 185); again, the forms taking the postvocalic allomorph have root- 

final vowels that are lost. 

(67) katab-at ‘she wrote’ ram-it ‘she threw’ 

libis-at ‘she wore’ mall-it ‘she filled’ 

100

Here, the stem-final vowel that triggers the use of -it is lost, rendering the condition on 

PCSA opaque. In this case, even if the stem-final vowel had been retained, there is no 

immediately apparent way in which the allomorphy is optimizing, since both allomorphs 

have the structure -VC. 

Thus, in both of the Sa’ani Arabic examples, we see that PCSA can be 

conditioned by a phonological element that is lost on the surface via the application of a 

regular phonological rule or constraint. This type of example is problematic for the P >> 

M model. In OT, phonological constraints generally either optimize output well- 

formedness or maximize faithfulness to the input. In the P >> M model, the same 

phonological constraints that are active in the phonological grammar are also responsible 

for PCSA, and therefore we expect cases of PCSA either to be optimizing or to maximize 

faithfulness (the latter being somewhat improbable since faithfulness to the input form of 

an affix would have the effect of preventing allomorphy, not producing it). In Sa’ani 

Arabic, PCSA does not seem to result in surface optimization in any way, and therefore 

modeling it using regular phonological constraints is problematic. We can account for it 

by writing specific constraints that specify which allomorph to use in which environment, 

but this defeats the purpose of modeling PCSA using P >> M, since the P >> M model 

was designed to account for examples where PCSA can be said to follow from general 

phonological constraints. 

Another example that exhibits PCSA conditioned by an element that is later 

deleted is found in Kimatuumbi (Bantu, Tanzania; Odden 1996). According to Odden 

(1996), the perfective has four allomorphs: -i̹te, -i̹i̹le, -i̹i̹ye, and the infix -i̹-. The 

allomorphs are distributed according to both segmental and syllable count-based factors 

101

as follows (Odden 1996: 51-53). Monosyllabic verbs (of shape (C)V(V)C), as well as 

stems whose final vowel is long, take -i̹te. Polysyllabic stems (except those with a final 

long vowel) take the -i̹- infix, which is inserted into the stem-final syllable; these forms 

also take the final vowel e rather than a. Stems ending in yaan or waan take the -i̹i̹ye 

suffix variant, and of particular interest here is the fact that glide-final roots take the -i̹i̹le 

allomorph, as shown below in (68)a-b (Odden 1996: 52) and that root-final [y] undergoes 

deletion in the examples in (68)b. 

(68) a. naa-bóyw-i̹i̹le ‘catch’ ni-chékw-i̹i̹le ‘shave’ 

b. tu-sák-i̹i̹le ‘germinate’ naa-égel-i̹i̹le ‘approach’ 

The overall generalization is as follows: among polysyllabic stems, glide-final stems take 

-i̹i̹le, and all other polysyllabic stems take -i̹- unless a more specific factor selects some 

other allomorph (such as a final long vowel, or as in the case of stems in yaan/waan). The 

post-glide allomorph does not readily derive from the same underlying form as the 

general allomorph that occurs with polysyllabic stems (-i̹-). Therefore, this appears to be 

a case of suppletion conditioned by the presence of a stem-final glide. What is 

particularly interesting is that the glide that conditions the selection of the -i̹i̹le allomorph 

is not actually present in some examples, as in examples such as naa-égel-i̹i̹le ‘approach’, 

corresponding to the infinitive form égelya. The perfective form undergoes glide deletion 

in this example through a general rule of the language, Homorganic Glide Deletion 

(Odden 1996: 108), which deletes glides before a homorganic vowel. The fact that the 

glide that conditions suppletion is itself not present in the surface form presents a 

problem for any approach that would attempt to account for allomorph distribution solely 

102

through surface constraints. This can only be handled in the P >> M approach if we make 

our P constraints refer to underlying glides that do not surface. 

To summarize, in this section we have seen examples in which the conditions on 

allomorph distribution are rendered opaque by other phonological processes. These 

examples cannot be handled by surface well-formedness constraints. Since they do not 

produce better-formed outputs, these examples can be said to be non-optimizing just like 

the examples seen in the preceding section, the only difference being that in some cases 

we can account for the opaque examples using regular phonological constraints, provided 

that we allow these constraints to refer to underlying elements that do not surface. As has 

been pointed out throughout this section, the existence of opaque PCSA is problematic 

for P >> M. However, it is predicted and easily accounted for in the subcategorization 

approach that I advocate here, since in this approach, the selection of affixes for stems 

takes place in morphology, which precedes the application of phonological processes at 

each level. 

2.1.3 Summary 

In the preceding discussion, we have seen examples of PCSA resulting in 

assimilation, dissimilation, syllable structure optimization, and syllable contact 

optimization (perhaps). We have also discussed a number of examples that do not 

optimize surface forms, including examples in which the conditions on allomorph 

distribution fail to surface, due to regular phonological processes. As has been discussed, 

the C/V distinction as well as a range of phonological features can condition PCSA, and 

in some cases the results look similar to the results of phonological rules. This is a pattern 

103

that we would predict under the P >> M model of PCSA. On the other hand, we have 

seen a number of examples that do not conform well to the predictions of P >> M. For 

example, in Bari, we found a number of affixes exhibiting PCSA that resulted in vowel 

harmony or disharmony or a combination of both. In some cases the pattern for one 

allomorph was the opposite of that for another allomorph, making it virtually impossible 

to capture the whole pattern using a single set of P constraints for the language as a 

whole. We also saw a number of non-optimizing PCSA examples. These are also 

problematic for the P >> M model, which was designed to account for optimizing PCSA. 

Though we can still analyze at least the neutral examples using P >> M, it is not clear 

why we would want to do this, since we lose the original motivation for analyzing PCSA 

using the same phonological constraints that are used to drive purely phonological 

processes. In the following section, I propose an analysis for these cases in terms of 

subcategorization frames. As will be seen, we can account for PCSA straightforwardly 

using such an approach, and we avoid the problems that come up when we try to model 

PCSA using phonological constraints. 

2.2 Analysis 

In the preceding section, I presented examples of suppletive allomorphy 

conditioned by segments or their features. Here, I outline a general strategy of analysis 

for examples of this type. The analysis is in terms of subcategorization. The 

subcategorization approach to morphology has been advanced in various forms by Lieber 

1980, Kiparsky 1982b, Selkirk 1982, Orgun 1996, and Yu 2003, among others. In this 

approach, affixation is handled through subcategorization frames, which include 

104

specifications for the type of element an affix will attach to (whether the element is a 

stem of a certain category or having certain features or properties, or a phonological 

element, or any combination of these). Affixes can subcategorize for any element in the 

phonological representation of the stem (as well as for elements of morphological, 

lexical, and other aspects of the stem), the only apparent restriction being that the affix in 

question must end up adjacent to the phonological element for which it subcategorizes. 36 

Affixation is the selection of affixes for stems in such a way as to satisfy the 

subcategorizational requirements of each morpheme, including phonological 

requirements. 

Taking Hungarian 2sg suffix allomorphy as a representative example, I will now 

demonstrate how a subcategorization approach can be used to analyze cases of suppletive 

allomorphy conditioned by features of the stem-initial or stem-final consonant. Recall 

from §2.1.2.2 that in Hungarian in the present indefinite, the 2sg is marked by /-El/ suffix 

when the root ends in a sibilant; otherwise, it is marked by /-s/. A set of subcategorization 

frames for these suffix allomorphs is shown below. 

(69) Hungarian 2sg present indefinite construction A 

[[ [sibilant]#] stem El 2sg pres indef suffix ] 2sg pres indef word 

Hungarian 2sg centripetal construction B (‘elsewhere’) 

[[ ] stem s 2sg pres indef suffix ] 2sg pres indef word 

The subcategorization frames are interpreted as follows. In construction A, which uses 

the suffix allomorph found in the more specific context, -El will attach to the right edge 

of any stem (modulo the non-phonological requirements, which I omit here) that ends in 

a segment that is specified [sibilant]. The result of the affixation of -El to the stem is a 

36 This restriction is predicted by the Generalized Determinant Focus Adjacency 

Condition (Inkelas 1990), to be discussed later in this section. 

105

word that is marked as 2sg, present indefinite. In construction B, the ‘elsewhere’ case, 37 

-s attaches to the right edge of any eligible stem regardless of its phonological shape, and 

the result of the affixation is again a word that is marked as 2sg. present indefinite. 

Suppletive allomorphy conditioned by the stem-initial or stem-final vowel can be 

handled in the same way. For example, in §2.1.2.5, I described a situation in Yucunany 

Mixtepec Mixtec where 3sg is marked by -à on i-final stems and by -ì elsewhere. Under 

the subcategorization approach, we can say simply that /-à / selects for a stem with a final 

/i/, while /-ì/ selects for any stem. This is schematized below. 

(70) Yucunany Mixtepec Mixtec 3sg construction A 

[[i#] stem à 3sg suffix ] 3sg word 

Yucunany Mixtepec Mixtec 3sg construction B (‘elsewhere’) 

[[ ] stem ì 3sg suffix ] 3sg word 

Construction B involves the further complication that the -ì suffix replaces the final 

vowel of the stem, but this can be handled in the phonology and therefore is not 

problematic for the simple analysis provided here. 

37 

The notion of the ‘elsewhere’ case may be implemented in any number of different 

ways; for example, in an OT model of morphology, some morphological constraint 

requiring 2sg to be marked by -s could be outranked by another morphological constraint 

requiring 2sg to be marked by -El, which could in turn be outranked by an undominated 

(and perhaps inviolable) constraint prohibiting subcategorization frames from being 

disobeyed. Under such a ranking, a hypothetical stem that satisfies both the phonological 

subcategorization requirements for the more specific -El allomorph and the (nonexistent) 

requirements for the -s allomorph can have two 2sg candidates that satisfy the highest 

ranked constraint against violation of subcategorization frames, but the candidate 

satisfying the highly ranked 2sg=-El constraint will win out over the candidate satisfying 

2sg=-s, and therefore the allomorph with the more specific environment will always be 

used when that environment is met. In a different hypothetical case where the stem does 

not meet the requirements for affixation of the more specific allomorph, the inviolable 

constraint against violating subcategorization frames will filter out candidates that 

incorrectly use the more specific allomorph. 

106

This approach can also easily handle suppletive allomorphy conditioned based on 

whether the stem-initial/final segment is a consonant or vowel. Ergative allomorphy in 

Warrgamay (Dixon 1980), discussed in §2.1.2.3, is an example of this. As described 

above, in Warrgamay the ergative is marked by -ŋgu after a vowel-final stem, or by -du 

after a consonant-final stem. Subcategorization frames for these two suffixes are shown 

below. Note that I have not characterized either suffix as the ‘elsewhere’ form since the 

two allomorphs exhibit perfect complementarity and neither occurs in a more specific 

environment than the other. 

(71) Warrgamay ergative construction A 

[[V#] stem du ergative suffix ] ergative word 

Warrgamay ergative construction B 

[[C#] stem ŋgu ergative suffix ] ergative word 

Because the subcategorization frames do not encode the apparent optimization that 

results from some examples of C/V-conditioned PCSA, the exact same analysis can be 

used with equally good results for both the optimizing and non-optimizing examples of 

this type. 

So far in this discussion we have not addressed one of the most interesting 

properties common to all of the examples presented in this chapter: namely, that the 

affixes involved in the suppletive relationship are always adjacent to the conditioning 

segment or feature of the stem. This is result of the survey is of theoretical interest for 

two reasons. The first reason is that it confirms a prediction inherent in a claim made by 

Inkelas (1990) building on Poser 1985, relating to a different phenomenon (namely, 

extrametricality). Extending Poser’s (1985) Determinant Focus Adjacency Condition, 

Inkelas (1990) proposed a Generalized Determinant Focus Adjacency Condition 

107

(GDFAC), given below (this was also discussed in chapter 1). 



If the GDFAC is a true principle of grammar, then the adjacency phenomenon 

observed in the survey can be said to follow from this principle. However, there is 

another possible explanation for the fact that PCSA always involves local conditioning. 

This relates to the historical development of PCSA, which I hypothesize to arise, in many 

cases, from a general phonological rule whose domain of application somehow becomes 

restricted to a single morphological context. This hypothesis is supported by the 

difficulty, pointed out above, in distinguishing suppletive allomorphy from allomorphy 

derived via phonological rules specific to particular morphological environments. If it is 

true that PCSA often or usually arises via the restriction of a phonological rule to a 

particular morpheme, then this could explain why allomorphs are adjacent to the edge of 

the stem that determines their distribution. <strong>Phonological</strong> rules generally apply locally (if 

we expand the term ‘local’ to refer not only to strictly adjacent segments but to features 

that are tier-adjacent), so the morphological patterns of suppletion that arise from the 

allophony derived by these rules should also occur at the same edge of the stem. Thus, 

the GDFAC may not be the ultimate explanation for the adjacency effect observed in 

PCSA, but as a principle of grammar it is consistent with the results of this survey, and it 

provides us with a way of accounting for the adjacency effects within the 

subcategorization approach. 

We have thus seen that the subcategorization approach is well equipped to model 

the locality generalization observed in the data. Under the P >> M approach, on the other 

hand, examples of non-locally conditioned PCSA are predicted to exist. For example, an 

108

OCP constraint banning multiple instances of a particular feature in a word (e.g., [labial]) 

could force the grammar to choose a different allomorph if an -m suffix were to be 

attached to a labial-initial stem. This type of example is unattested in the surveyed 

languages, so in this domain it appears that the P >> M model overpredicts and is too 

powerful. This is therefore an important difference between the two models, since as 

discussed above, the subcategorization rules out the unattested non-local conditioning of 

PCSA. 

2.3 Conclusion 

In this chapter we have seen a number of examples in which suppletive 

allomorphy is conditioned by segments and/or their features. Some observations that 

were made about these examples were as follows. First, we have seen that conditions on 

PCSA are local, in that the condition is always at the edge of the stem where the affix 

allomorphs attach. Second, it was demonstrated that a wide range of features, in addition 

to the C/V distinction, can determine allomorph distribution. Third, we have seen many 

examples that are not obviously optimizing. Finally, we have seen examples of opaque 

conditions on PCSA, which show that allomorph distribution is sensitive to input 

properties of the stem, not output properties. 

An analysis of these examples based on subcategorization for phonological 

elements was proposed in §2.2. We have seen that subcategorization can handle the 

examples described above without difficulty. As has been discussed, the P >> M model 

does not fare as well in dealing with some of the cases presented here. P >> M was 

conceived in order to account for cases of phonologically optimizing PCSA that follows 

109

from well-known phonological constraints. When PCSA seems not to be optimizing or 

would require stipulating unnatural or language-specific constraints, we lose the 

advantage of generality and explanatory power that has compelled researchers to use the 

P >> M model to account for PCSA in the past. This problem is avoided in the 

subcategorization model, which does not rely on phonological constraints to account for 

phonological effects in morphology. 

In chapters 3 and 4, I discuss further results of the cross-linguistic survey 

involving suppletive allomorphy conditioned by other types of phonological factors, and I 

propose analyses for those examples along the same lines as the analysis outlined in this 

chapter. As we will see, the other types of PCSA to be discussed pose similar challenges 

for P >> M and lend themselves equally well to subcategorization-based analyses. 

110

Chapter 3: Tone/stress conditioned suppletive allomorphy 

In this chapter, I consider cases of phonologically conditioned suppletive 

allomorphy (PCSA) where the conditioning factor is the tone or stress pattern of the stem. 

This survey revealed few such cases in comparison to the other types of PCSA described 

in chapters 2 and 4, but each case is robust enough so that there can be no question that 

the phenomenon exists. This chapter is structured as follows. First, in §3.1, I give 

examples of PCSA conditioned by tone or stress. In §3.2, I give a subcategorization- 

based analysis of this phenomenon along the same lines as the analyses provided in 

chapters 2 and 4 for different types of PCSA. I conclude in §3.3. 


In this section, I present examples of PCSA that involve conditioning by stress or 

tone. These examples were revealed by a cross-linguistic survey of PCSA, the details of 

which were described in chapter 1. A search of over 600 grammars revealed 137 cases of 

PCSA in 67 different languages. Of these, eight cases (from seven different languages) 

are stress- or tone-conditioned; these are the cases that I will discuss here. The relatively 

small number of cases of this type in comparison to other types of PCSA is somewhat 

puzzling. Languages with tone or a prominent stress pattern often exhibit several 

phonological rules or constraints resulting in tone or stress pattern effects (such as high 

tone dissimilation, tonal plateauing, tone spreading, and stress clash and lapse 

avoidance), so we might expect this to extend to PCSA as well. 

Contrary to this expectation, few cases of PCSA conditioned by stress or tone 

were found. There is the possibility that some cases were missed by the survey, but given 

111

the efforts to make the survey large and balanced, it seems unlikely that such a significant 

class of examples could have been systematically neglected. One possible factor in the 

apparent lack of cases of stress-conditioned allomorphy is that there is some overlap 

between these cases and the cases of foot-conditioned PCSA to be described in chapter 4. 

Some cases that we might want to describe as being stress-conditioned may have ended 

up in chapter 4 since they could also be viewed as being foot-conditioned. The lack of 

cases of stress-conditioned allomorphy may therefore be due in part to the (somewhat 

artificial) division between the stress-conditioned allomorphy described in this chapter 

and the foot-conditioned allomorphy to be described in chapter 4. 

However, this does not explain the lack of cases of tone-conditioned allomorphy, 

which remains puzzling. Only two examples of tone-conditioned PCSA were revealed by 

the survey, despite the fact that over half of the world’s languages have tone and a 

significant number of these languages must exhibit some tonal phonology. I will leave 

this problem for future consideration. 

This section is organized as follows. I begin in §3.1.1 with a general discussion of 

the expected and attested types of stress- and tone-conditioned PCSA. In section §3.1.1.1, 

I present examples of stress-conditioned allomorphy conditioned by stress. In §3.1.1.2, I 

discuss examples of stem allomorphy (tone- and stress-conditioned). In §3.1.1.3, I present 

an example of tone-conditioned PCSA. In §3.1.1.4, I discuss an example of stress- 

conditioned allomorphy that appears not to be optimizing. I conclude in §3.1.2 with a 

summary of these examples before moving to the analysis in §3.2. 

112


3.1.1.1 Stress effects 

Many languages exhibit stress assignment rules/constraints that seem to maximize 

alternating patterns of stressed and unstressed syllables in a word. Constraints such as 

CLASH and LAPSE have been devised to account for the apparent universal dispreferences 

for sequences of two stressed or two unstressed syllables, respectively (see e.g. Prince 

1983, Selkirk 1984). In phonology, violations of CLASH and LAPSE are most commonly 

repaired via stress shift, so that the underlying stress pattern (if any) of a morpheme or 

the stress pattern assigned to it in citation form is changed in certain contexts to optimize 

the overall stress pattern of the word. In PCSA, we might imagine that highly-ranked 

CLASH or LAPSE could cause an inherently stressless affix allomorph to be selected when 

it occurs next to a stressed stem syllable, or vice versa. Of course, in order for us to know 

that a given example involved suppletive allomorphy rather than purely phonological 

allomorphy, the affix allomorphs in question would have to differ in more than just their 

stress pattern; i.e., they would probably also need to be different in their segmental 

makeup. Three such examples are attested in the survey, to be discussed below. 

Aside from alternating stress patterns, another stress-related consideration is the 

notion of the optimal stress-bearing syllable. In many languages, stress is ‘attracted’ to 

certain syllable types, usually heavy syllables. In some cases, it has been claimed that 

stress can also be attracted to syllables having certain vowels, not only in terms of their 

length, but in some cases their quality (though this may relate to duration if high vowels 

are inherently shorter than low vowels). In this section, I discuss one case where it has 

113

een claimed that PCSA results in stress occurring on syllables containing a vowel that is 

an optimal bearer of stress. 

An example of PCSA resulting in an alternating stress pattern is found in Dutch 

(Booij 1997), where the plural is marked by -s or -en, depending on whether the final 

syllable of the base word is unstressed or stressed, respectively. Examples are given 

below (Booij 1997: 272-273; stem-final doubled consonants are orthographic only). 

(1) a. dam ‘dam’ dámm-en ‘dams’ 

kanón ‘gun’ kanónn-en ‘guns’ 

kanáal ‘channel’ kanáal-en ‘channels’ 

lèdikánt ‘bed’ lèdikánt-en ‘beds’ 

ólifànt ‘elephant’ ólifànt-en ‘elephants’ 

b. kánon ‘canon’ kánon-s ‘canons’ 

bézəm ‘sweep’ bézəm-s ‘sweeps’ 

tóga ‘gown’ tóga-s ‘gowns’ 

proféssor ‘professor’ proféssor-s ‘professors’ 

If the -en suffix is inherently unstressed, then using it with the stems in (1)b would have 

resulted in a sequence of unstressed syllables. It could therefore be said, in P >> M terms, 

that the allomorphy is driven by LAPSE. Of course, an alternative way to avoid a LAPSE 

violation if -en were used with unstressed-final stems would be to stress the syllable 

containing the -en suffix. However, this would require us to posit a monosyllabic foot at 

the end of the word; the allomorphy can therefore be said to optimize foot structure as 

well (or, to help avoid the creation of subminimal feet). 1 

This Dutch example is interesting for another reason, namely that it appears to be 

sensitive to derived phonological information since stress when predictable is usually 

1 

For this reason, the Dutch example also fits into the category of examples discussed in 

chapter 4 (§4.1.1.1) in which allomorphy appears to be motivated by foot structure. I 

have discussed the example here instead of in chapter 4 since Booij (1997) characterizes 

it as being driven by stress. 

114

assumed to be assigned by the grammar rather than being lexically specified. Sensitivity 

to derived properties is predicted for PCSA both by the P >> M model and by any 

approach such as Lexical Phonology (Kiparsky 1982a, Mohanan 1986) where 

morphological processes apply cyclically and where phonological and morphological 

processes are ‘interleaved’ in grammar. However, it is not entirely clear that Dutch 

provides an example of sensitivity to derived properties, because stress is not entirely 

predictable, as acknowledged by Booij and Lieber (1993: 41) in their treatment of the 

Dutch adjectival suffix (to be discussed in §3.1.1.4). Assuming that stress normally falls 

on the penultimate syllable, the stems in (1)a would need to be lexically specified as 

having final stress. Thus, if we claim that -en occurs with final-stressed stems and -s 

occurs ‘elsewhere’, then we do not have to refer to any derived properties. On the other 

hand, if -s is considered to be the more specific allomorph, then its distribution would 

need to refer to a derived property, namely the default penultimate stress that is assumed 

to be assigned by a general rule of the language. The problem is that it is not entirely 

clear which is the more specific allomorph and which is the ‘elsewhere’ allomorph here. 

Another possible case of reference to a derived property is found in Spanish, 

where the feminine definite article has two allomorphs whose distribution is stress- 

sensitive (Kikuchi 2001). The most commonly used allomorph is la, but the form el 

occurs before stems that begin with a stressed /a/. Some examples involving a-initial 

stems are shown below (Kikuchi 2001: 50). Note that all of these nouns have feminine 

gender. 

(2) a. la alméja ‘the clam’ la anguíla ‘the eel’ 

la aréna ‘the arena’ la amíga ‘the friend’ 

b. el álma ‘the soul’ el águila ‘the eagle’ 

el área ‘the area’ el áma ‘the mistress’ 

115

Kikuchi argues that this pattern is driven by positional faithfulness (Beckman 1998). 

When la occurs before an unstressed /a/, the /a/ vowels of the article and stem coalesce 

into a single [a]. Kikuchi claims that el is used when the stem-initial /a/ is stressed in 

order to avoid coalescence of a stressed vowel with another vowel. Thus, the initial 

stressed /a/ is ‘protected’ by the use of el. 

An interesting aspect of this pattern is that the article that is used can change with 

the addition of certain affixes to the stem if this causes its stress to shift. Some examples 

are shown below (Kikuchi 2001: 50, except where noted). 

(3) el água ‘the water’ la aguáda ‘the water supply’ 

el árma ‘the weapon’ la armadúra ‘the weaponry’ (Kikuchi 2000) 

If the examples in (3) are indeed productively suffixed forms, then this constitutes 

an example of PCSA that makes reference to a derived phonological property. The 

determination of which allomorph to use for the definite article must be carried out after 

the attachment of affixes in (3). These affixes trigger a shift of the stem stress one 

syllable to the right, so that the initial /a/ is no longer stressed. Then, when the definite 

article is added, the more specific (el) allomorph ‘checks’ the stem for a stressed initial 

/a/. Due to the stress shift triggered by affixation, aguáda and armadúra do not satisfy 

this condition, so el is not used; the more general allomorph la is used instead. Thus, the 

distribution of allomorphs refers to a stress pattern that was not present in the underlying 

form of the stem. 

Although this seems on its surface to be a good example of PCSA that refers to a 

derived property, there is a problem with the example, which is that in some cases a stress 

shift does not result in a change to the form of the definite article as expected. Kikuchi 

116

(2000) points out that diminutive forms whose stems have initial stressed /a/ take the el 

article even though their stress shifts, as in the examples below. 2 

(4) el áma ‘the mistress’ el amíta ‘the mistress (dim.)’ 

el álma ‘the soul’ el almíta ‘the soul (dim.)’ 

Furthermore, el is also used with compounds, which have stress on the rightmost element, 

when the leftmost element has initial stressed [a] in its citation form. 

(5) el água ‘the water’ el aguafuérte ‘etching’ el aguamála ‘jellyfish’ 

The fact that el occurs unexpectedly in these examples suggests that the definite 

article is sensitive to the stress pattern of the stem rather than the word. This is somewhat 

puzzling since the morphological bracketing should be such that the definite article 

(which is a clitic) occurs outside the stem+affixes, as follows: [article [[stem]-suffix]]. 

There are two possible solutions to this puzzle. The first is to assume that stress is 

assigned at the stem level and that stress shift in compounds and affixed forms occurs not 

immediately after compounding or affixation has applied, but rather at the level of the 

phonological word, i.e. after cliticization. This would allow the clitic to refer to the stress 

pattern of the stem before stress shift takes place. A second possibility is that there is a 

mismatch between the morphological and phonological bracketing of words with the 

definite article, so that the article and the word-initial stem form a phonological 

constituent to the exclusion of what follows (i.e., a suffix or the second member of a 

compound). In either case, we have a way around the problem in the subcategorization 

approach to PCSA, but this example is difficult in the P >> M model because the 

condition that determines the allomorph distribution is rendered opaque by stress shift. 

2 Much of Kikuchi (2000) is duplicated in Kikuchi (2001), but the diminutive and 

compound forms discussed here appear only in Kikuchi (2000). 

117

Because OT (and, in particular, the P >> M model) is surface-oriented, the P >> M model 

predicts that we should not find examples of PCSA whose conditions are rendered 

opaque by ‘later’ phonological processes. Some other examples of this type were seen in 

chapter 2 as well and were argued to provide evidence in support of the subcategorization 

approach to PCSA. 

An example of a different type of stress-conditioned allomorphy is found in 

Shipibo (Panoan, Peru; Elías-Ulloa 2004). This case seems to involve the notion of 

optimal stress-bearing syllables or vowels. In Shipibo, the suffix meaning ‘again’ takes 

the form -ribi when the second syllable of the suffix is an even-numbered syllable 

(counting from left to right) and -riba when the second syllable of the suffix is in an odd- 

numbered syllable. Some examples are provided below (Elías-Ulloa 2004: 2-3; examples 

are originally from Lauriault 1948 except where noted). 

(6) yono-ribi-kɨ ka-ma-ribi-kɨ 

command-Again-Past go-Caus-Again-Past 

‘commanded again’ ‘went again’ 

pi-riba-kɨ yomɨtso-riba-kɨ 

eat-Again-Past steal-Again-Past 

‘ate again’ ‘stole again’ 

ka-yama-riba-kɨ 

go-Neg-Again-Past 

‘do not go again’ (from Elías-Ulloa field notes) 

Elías-Ulloa relates this to the stress pattern of the language, which stresses the second 

syllable if it is closed, or otherwise the first syllable. He accounts for the -ribi ~ -riba 

allomorphy using the constraint rankings *i/Head >> *a/Head and *a/NonHead >> 

*i/NonHead. These constraints are claimed to account for why -riba allomorph occurs 

when its second syllable will be stressed, while -ribi occurs when its second syllable will 

118

not be stressed: [a] is more sonorous than [i] and is therefore a better bearer of stress than 

[i]. While it seems desirable to relate the allomorphy to the stress system, Elías-Ulloa 

does not consider the possibility that /-ribi/ surfaces as -riba when the second syllable is 

stressed. This would require a rule, specific to the ‘again’ suffix, changing i to a in a 

stressed syllable. Though this rule would be item-specific, it would still be a phonetically 

motivated rule, using Elías-Ulloa’s argument that [a] is a better stress-bearer than [i]. 

Thus, we cannot be sure that this is a case of PCSA. Nonetheless, it does seem to be an 

example of allomorphy that results in stress occurring on a more ‘stressable’ syllable than 

it would if there were no allomorphy. 

In this section we have seen three examples of stress-related allomorphy. One 

example, from Dutch, involved PCSA that seemed to maximize an alternating stress 

pattern while preventing the creation of monosyllabic feet. Another example, the case of 

the Spanish feminine definite article, involved the preferential retention of a stressed 

vowel in surface forms. The third example, from Shipibo, appeared to ensure that stress 

occurred on syllables containing /a/, which is considered to be a good stress-bearing 

vowel since it is highly sonorous. Thus, we have seen three different ways in which 

PCSA can be argued to be driven by phonological optimization. This is consistent with 

the P >> M model, though note that it is also consistent with the subcategorization 

approach (as will be demonstrated in §3.2). 

3.1.1.2 Stem allomorphy 

In the preceding section, we saw examples in which the distribution of allomorphs 

of an affix or clitic was determined by stress. So far, we have not discussed any 

119

examples in which a stem exhibits PCSA; even in chapter 2, where 74 cases of 

segmentally conditioned PCSA were discussed, we did not find any examples of stem 

allomorphy. This is a striking result. It raises the question of why there are not more 

examples, and also why the attested examples are not segmentally conditioned but rather 

conditioned by more ‘global’ or suprasegmental factors such as stress and tone. The P >> 

M model of PCSA does not answer these questions. 

The subcategorization approach, on the other hand, allows us to account for this 

cross-linguistic pattern as follows. First, since we are assuming that words are built ‘from 

the inside out,’ we predict that allomorphy in a root should never be sensitive to 

properties of an affix. As will be seen, this prediction can be maintained based on the 

present survey. We furthermore predict that allomorphy in the ‘stem of affixation’ should 

also never be sensitive to properties of any affix, with the following exception: if what 

appears to be the stem is morphologically complex, then in some cases the material that 

occurs next to the root could possibly be an infix, in which case its distribution could be 

conditioned based on some property of an affix that attaches ‘before’ the infix but ends 

up farther away from the root in terms of surface linear order. Since this is a very specific 

and complex scenario, we expect that it should not occur very often, and therefore that 

examples of this type should be rare. It is also crucial that in such cases, the root should 

be extractable from the ‘pseudo-stem’ that is formed by the addition of the infix; if all of 

the root’s segments are not contained in this pseudo-stem, then we must conclude that we 

are dealing with root allomorphy rather than stem allomorphy; this then would contradict 

a prediction of the subcategorization model as presented here. The subcategorization 

approach severely restricts the possibility of affix-conditioned stem allomorphy, but note 

120

that other types of phonological conditioning on stems may still be permitted if the 

phonological condition is imposed from outside of the word. The model that I am using 

here is only meant to account for morphological processes and does not make predictions 

for or set limitations on phonological interactions above the level of the word. 

While P >> M predicts that stem allomorphy conditioned by phonological 

properties of affixes should be perfectly acceptable, the subcategorization approach 

severely restricts the instances in which stem allomorphy can occur, as discussed above. 

Nonetheless, some types of stem allomorphy are still allowed and able to be captured in 

the subcategorization approach, and in fact, as mentioned above, some such cases are 

attested. In this chapter, I discuss two cases of PCSA involving stem allomorphy: one 

conditioned by stress, and the other conditioned by tone. 

An example of stress-conditioned stem allomorphy occurs in Italian (Hall 1948). 

In Italian, some stems have allomorphs ending in /isk/ that occur only in morphological 

contexts where the word stress falls on the root-final syllable, namely, in the present and 

subjunctive 1sg, 2sg, 3sg, and 3pl and in the 2sg imperative (Hall 1948: 25, 27). One 

such root is fin- ‘finish’; some examples are shown below (Hall 1948: 214). 

(7) Present 

finísk-o ‘I finish’ fin-iámo ‘we finish’ 

finíšš-i ‘you (sg.) finish’ fin-íte ‘you (pl.) finish’ 

finíšš-e ‘s/he finishes’ finísk-ɔno ‘they finish’ 

Subjunctive 

finísk-a ‘that I finish’ fin-iámo ‘that we finish’ 

finísk-a ‘that you (sg.) finish’ fin-iáte ‘that you (pl.) finish’ 

finísk-a ‘that s/he finish’ finísk-ano ‘that they finish’ 

Imperative 

fin-iámo ‘let’s finish’ 

finíšš-i ‘(you (sg.)) finish!’ fin-íte ‘(you (pl.)) finish!’ 

121

Other roots of this type include aǧ- ‘act’ (Hall 1948: 43), argu- ‘argue’ (1948: 44), dilu- 

‘add water’ (1948: 45), mɛńt- ‘lie’ (1948: 45), diminu- ‘diminish’ (1948: 52), and 

ammon- ‘admonish’ (1948: 61). There appears not to be any semantic or phonological 

generalization regarding which roots pattern with this class. A possible motivation for 

this pattern of allomorphy is stress clash avoidance. If we assume that /isk/ is inherently 

stressed, then its occurrence next to an inherently stressed suffix would create a stress 

clash, so /isk/ is avoided in that context. 

This example is particularly interesting and important because it is the first case 

of apparent stem allomorphy that we have seen. As noted above, no examples were 

discussed in chapter 2, which described cases of segmentally conditioned allomorphy. 

The subcategorization approach allows us to explain the relative rarity of stem 

allomorphy, but it also requires us to assume that /isk/ here is an infix. This may provoke 

some skepticism, but in modern analyses of Italian stem allomorphy (e.g., DiFabio 1990, 

Schwarze 1999), -isc is considered to be an affix or ‘stem extension’ (Schwarze 1999). 

Therefore, it is already assumed that stems containing -isc do not exhibit root 

allomorphy, but rather the presence or absence of -isc is determined by the stress pattern 

of the word. And, in fact, DiFabio (1990) analyzes -isc- as an infix. 3 Assuming that -isc- 

3 

The fact that separate stems fin- and finisc- co-existed at an earlier stage in the history 

of Italian does not preclude an infixation analysis, since speakers could have created the 

infix -isc- on the basis of the existence of other such pairs of verbs in the language. 

However, if each of the two separate stems previously was able to occur in all of the 

morphological/phonological environments in (7), then we are faced with the question of 

how the stem allomorphs came to be distributed in the modern language (after the stems 

collapsed into a single suppletive paradigm) in just such a way so as to avoid stress clash. 

One possibility is that forms of each stem with every possible inflectional suffix coexisted 

at some stage, and as the separate stems merged, perhaps via the creation of the - 

isc- infix, the forms without -isc- preferentially survived in the stressed suffix 

environment because they had a more harmonic rhythmic pattern than those with -isc- 

122

is an infix explains how it can be sensitive to a property of the stem that is determined by 

affixes that occur later in the word (in terms of their linear order), without posing any 

problem for the ‘inside-out’ direction of word formation in the subcategorization 

approach advocated here. This move may seem to be a clever trick designed to uphold the 

prediction that affixes cannot condition stem allomorphy. However, as discussed earlier, 

the lack of cases of phonologically conditioned stem allomorphy is striking given the 

number of cases involving affix allomorphy. Furthermore, it is not the case that any 

possible putative case of stem allomorphy could be explained away. The infixation 

account is possible for Italian because isc is present in all of the extended stems, but 

imagine a language in which some stems had an etymologically unrelated allomorph with 

an extra syllable that was used in a particular stress context. For example, if complet- (lit. 

‘complete’) were used instead of finisc- where it occurs in the paradigm of fin-, it would 

be impossible to claim that the allomorphy resulted from the addition of a stem extension, 

and we would have to conclude that this was a counterexample to our model. No such 

cases were revealed by the present survey. 

Another case of stem allomorphy is found in Zahao (Chin, Burma; Osburne 

1975). According to Yip (2004), Zahao exhibits suppletive allomorphy in verb stems that 

involves tonal alternations. In particular, verbs in Zahao have two stems, a ‘primary’ stem 

(this would be true whether -isc- had inherent stress or not; for example, the modern form 

fin-íte ‘you (pl.) finish’ has an alternating stress pattern, whereas its hypothetical 

competitors finísk-íte and finisk-íte do not). Note, however, that this historical scenario 

does not require that speakers ever utilized a phonological constraint in determining when 

fin- vs. fin-isc- would be used. In the P >> M model, P constraints are used to determine 

which affixes will be allowed to combine with which stems, not to choose between 

multiple pre-existing semantically equivalent words. Perhaps the choice between words 

already existing in a language is more a conscious stylistic decision than one that is 

driven by constraints or rules of grammar. 

123

and a ‘secondary’ stem. The primary stem is used in main clauses, while secondary stems 

are used in subordinate clauses. Though this can be stated in syntactic terms, there is 

some interaction with focus, and Yip suggests that the conditioning is prosodic. Some 

examples of primary-secondary stem pairs are shown below (following Yip 2004, tones 

are given next to the stem; stems with no tone are underlyingly toneless). 

(8) Primary stem Secondary stem Gloss 

hmaan L hmaan L ‘be correct’ 

hreen L hren H ‘lock up, close’ 

laam LH laam L ‘dance’ 

hmaan L hmaan L ‘be correct’ 

ŋaan H ŋaan L ‘write’ 

ree LH reet H ‘insert’ 

khur L khurɁ ‘shiver’ 

cat H caɁ ‘cut off’ 

Yip points out that in each pair, the secondary stem is equally or less ‘marked’ 

than the primary stem in terms of its laryngeal specification (tone or final glottal stop): if 

the primary stem has a contour tone, then the secondary stem has a level tone; if the 

primary stem has a level tone, then the secondary stem has either a level tone or a final 

glottal stop, which is considered to be less marked than a level tone. Yip proposes a 

harmonic scale Ɂ > H,L > LH, where the glottal stop is most harmonic and least marked, 

the H and L tones are somewhat less harmonic and more marked (but equivalent to each 

other on the scale), and LH is the least harmonic and the most marked. With respect to 

the primary stem, a secondary stem can either stay at the same level on the scale or move 

up by one degree (but not two). As seen in the examples above, in some cases there is no 

change between the primary and secondary stem. In Yip’s analysis, the allomorphy is 

driven by several different factors, including: first, the existence in the lexicon of 

124

multiple underlying forms for stems as well as a constraint MAXPARADIGM that requires 

every underlying stem form to be realized in some environment; second, a constraint 

requiring that every stem must have a laryngeal specification; third, conjoined MAXH and 

MAXL constraints that prevent the secondary stem from differing from the primary stem 

by more than one degree on the harmonic scale given above; and finally, a constraint 

prohibiting contour tones in a secondary stem. In a P >> M terms, all of these constraints 

would outrank UNIFORMEXPONENCE, which is the morphological cosntraint that requires 

morphemes to be realized consistently regardless of the environment in which they occur. 

Note, though, that the existence of this example of stem allomorphy does not 

contradict any of the predictions made by the subcategorization model as discussed 

above, and therefore this need not be taken as evidence in favor of the P >> M model 

simply because it can be analyzed in those terms. As noted above, stem allomorphy that 

is not conditioned by an affix does not constitute a counterexample to the 

subcategorization approach that I am using, especially given that in this particular case, 

as acknowledged by Yip (2004), the distribution of stem allomorphs may reduce to 

syntactic conditioning, which is outside the scope of what is covered by either model of 

PCSA being contrasted here. 

In summary, in this section we have seen two examples of stem allomorphy, one 

conditioned by stress and one conditioned by syntax and/or prosodic factors and 

involving tonal alternations. As has been discussed, the small number of examples is 

problematic for the most basic form of the P >> M model, which does not take into 

account morphological constituency, since in principle this means that stems can exhibit 

PCSA just as easily as affixes, which seems not to be the case. On the other hand, as 

125

discussed above, the subcategorization approach allows us to explain why stem PCSA is 

relatively rare, while still not ruling out the examples that are attested in the present 

survey. 

3.1.1.3 Tone effects 

In addition to the Zahao example described above, one other example of PCSA 

involving tone was found in this survey. The fact that only two examples of tone-related 

PCSA were found is somewhat surprising given the fact that phonological tone rules are 

so common in the world’s languages. We might have expected to find a number of 

examples of PCSA resulting in the same surface patterns that result from these tone rules: 

tone spread, dissimilation of adjacent H tones, and tonal plateauing, to name a few. The 

OT constraints CLASH and LAPSE, originally formulated to account for stress patterns, are 

now also used for tone (as advocated by Zoll 2002), so we might expect these constraints 

to be among those that could be ranked above a morphological constraint (in a P >> M 

analysis), resulting in PCSA. No such examples are found. Other constraints that have 

been used for tone include the OBLIGATORY CONTOUR PRINCIPLE (OCP; Leben 1973, 

McCarthy 1986) for dissimilation and *H to capture the idea that a high tone is ‘marked’, 

especially in languages with an underlying H vs. Ø tone contrast. In the P >> M 

approach, we would expect that these constraints could play a role in PCSA as well. 

The P >> M also predicts the existence of some non-local tone effects. For 

example, a conjoined constraint *H&*H ranked over UNIFORMEXPONENCE could result 

in long-distance tonal dissimilation such that a H-toned affix allomorph could fail to be 

selected if there is a H tone anywhere in the stem. The subcategorization approach that I 

126

take here, which includes a requirement that affixes must be adjacent to the phonological 

elements for which they subcategorize, predicts that any such effect should be purely 

local, so that the triggering H tone would have to be at the edge of the stem (or, at the 

edge of the stem on the tonal tier). This difference in predictions turns out to be 

somewhat of a moot point since no cases of PCSA resulting in tone dissimilation of any 

kind are found. However, the fact that the subcategorization approach is more restrictive 

regarding types of tone dissimilation is a point in its favor given the lack of any cases at 

all. The possibility of constraint conjunction vastly increases the number and type of 

possible phenomena predicted by P >> M, and contrary to any argument that constraint 

conjunction is not an accepted mechanism in modern mainstream OT, Yip’s (2004) 

analysis of Zahao discussed above demonstrates that constraint conjunction is still very 

much in use in modern OT analyses. 

One example of tone-conditioned PCSA is attested in the survey, though it does 

not fit into any of the predicted types described above. The example is found in the 

Yucunany dialect of Mixtepec Mixtec (Otomanguean, Mexico; Paster and Beam de 

Azcona 2005). This language exhibits PCSA in the first person singular clitic (subject or 

possessor). The 1sg is marked by a floating low tone that associates to the end of the verb 

or noun phrase, except in those cases where the final tone of the verb or noun phrase is 

low, in which case the first person singular is marked by the yù allomorph. Examples M- 

or H-final stems taking the floating L tone allomorph are shown below (examples are 

from my field notes). 

127

(9) nàmá ‘soap’ nàmáà ‘my soap’ 

kwíìí ‘narrow/thin’ kwíìíì ‘I am narrow/thin’ 

vílú ‘cat’ vílúù ‘my cat’ 

tìinà ncháá ‘blue dog’ tìinà nchááà ‘my blue dog’ 

tzàáku ‘corral’ tzàákuù ‘my corral’ 

yùúti ‘sand’ yùútiì ‘my sand’ 

sì’i ‘leg’ sì’iì ‘my leg’ 

kwà’a ‘man’s sister’ kwà’aà ‘my sister’ 

As seen in the examples below, the yù allomorph occurs following a L tone. 

(10) sòkò ‘shoulder’ sòkò yù ‘my shoulder’ 

tutù ‘paper’ tutù yù ‘my paper’ 

chá’à ‘short’ chá’à yù ‘I am short’ 

ve’e nchá’ì ‘black house’ ve’e nchá’ì yù ‘my black house’ 

This case likely involves suppletion since the two different forms of the 1sg are not 

phonologically similar enough to warrant relating them to a single underlying form. Even 

if they were, we would want the underlying form to resemble the ‘elsewhere’ allomorph 

(the floating L that occurs after H- and M-final roots) rather than the allomorph that 

occurs in the more specific environment (the yù form that occurs with L-final roots), and 

yet this would require us to posit insertion of [yu] following a L-toned root. This seems 

implausible. I conclude that this example involves genuine suppletive allomorphy, 

conditioned by the final tone of the root. 

This pattern, where the floating L tone allomorph occurs in all environments 

except where preceded by a L tone, serves to maintain a distinction between 1sg and 

unmarked forms. Without the yù allomorph, 1sg forms would be homophonous with the 

unmarked forms of verbs and noun phrases with final L tone. One possible way of 

modeling this is to use a phonological constraint against homophony among forms in a 

128

paradigm. However, we can explain the apparent homophony avoidance without positing 

a constraint that generates it. In a different dialect of Mixtepec Mixtec described by Pike 

and Ibach (1978), yù is a 1sg polite marker, while the floating low marks the informal 

1sg. This is most likely the conservative dialect since it would be difficult to develop the 

semantically conditioned allomorphy from the phonological condition. Furthermore, the 

Yucunany dialect does exhibit an informal vs. polite distinction in the 2sg and 3sg, 

suggesting that it had such a distinction in the 1sg at an earlier stage. 

The tables below show the proposed stages in the development of the modern 

pattern of allomorphy in the Yucunany dialect. (11)a shows the original stage, where the 

two 1sg markers are distributed based on the informal vs. polite distinction. This pattern 

is retained in the modern dialect described by Pike and Ibach (1978). (11)b shows a 

proposed intermediate stage in the Yucunany dialect, where the semantic distinction 

between informal and polite in the 1sg has been lost, and both forms of the 1sg marker 

still exist in free variation. At this stage, each type of root has two possible 1sg forms, but 

L-final stems have only one form that is not homophonous with the root. In some 

contexts where a L-final root is marked with a redundant final L tone, the intended 1sg 

form may be mistaken for a plain form if the 1sg meaning is not of critical relevance in 

the discourse. Therefore, assuming that the two allomorphs are used by speakers with 

equal frequency, the majority of underlyingly L-final stems that are understood by the 

listener to be 1sg forms will have the yù allomorph rather than the floating L tone. Since 

it is used more frequently than the floating L tone, the yù allomorph ultimately ‘wins 

out’, becoming the only 1sg marker to be used with L-final roots, as shown in (11)c. At 

this stage, one possible development is for the 1sg of M- and H-final roots to be marked 

129

only by yù by analogy with L-final roots. Instead, in the modern dialect, the M- and H- 

final roots converge on the floating L tone as the marker of 1sg, as seen in (11)d. One 

explanation for this is that speakers picked up the discrepancy between the existence of 

the L-final 1sg forms for M- and H-final roots on the one hand, and the lack of L-final 

1sg forms for L-final roots on the other hand. This could have led to the generalization 

that yù is used with L-final roots while the floating L tone is used with M- and H-final 

roots. 

(11) a. Mixtepec Mixtec 

Root type Plain form 1sg informal 1sg polite 

final L final L final L =yù 

final M final M final L =yù 

final H final H final L =yù 

b. Early Yucunany Mixtepec Mixtec 

Root type Plain form 1sg 

final L final L final L ~ =yù 

final M final M final L ~ =yù 

final H final H final L ~ =yù 

c. Intermediate Yucunany Mixtepec Mixtec 


final L final L =yù 

final M final M final L ~ =yù 

final H final H final L ~ =yù 

d. Modern Yucunany Mixtepec Mixtec 


final L final L =yù 

final M final M final L 

final H final H final L 

Thus, the pattern of tone-conditioned suppletive allomorphy emerged in Yucunany 

Mixtepec Mixtec without necessarily being driven by homophony avoidance. We are 

therefore free to analyze the pattern in a way that captures the generalization without 

necessarily encoding the anti-homophony concept in our analysis. A subcategorization 

130

analysis is thus well-suited to this example, as will be demonstrated in §3.2. 

In this section, we have discussed one example of tone-conditioned PCSA. 

Including the Zahao example discussed in the preceding section, there were two tone- 

related examples of PCSA found in this survey. 


In addition to the examples described above, which can each be viewed as 

phonologically optimizing in some way, the survey also revealed two examples of stress- 

conditioned allomorphy where the distribution of allomorphs does not appear to optimize 

the overall phonological pattern of the word. By now, such cases will not be surprising 

since a number of examples were discussed in chapter 2. However, it should be pointed 

out once again that examples of this kind are problematic for the P >> M model, which 

reduces PCSA to resulting from regular phonological well-formedness constraints. The 

subcategorization approach, on the other hand, predicts and accounts straightforwardly 

for non-optimizing PCSA. 

One example of apparent non-optimizing PCSA that relates to stress is found in 

Woleaian (Trukic, Micronesia; Sohn 1975, Sohn and Tawerilmang 1976). According to 

Kennedy (2002), the ‘denotative’ marker (which converts intransitive verbs from 

transitive verbs) has two allomorphs. It can either appear as a suffixed reduplicant 4 

4 

The characterization of this pattern as suffixing (rather than infixing before the stemfinal 

foot) is Kennedy’s. Kennedy does not seem to entertain the infixing analysis, 

focusing instead on the choice between prefixing vs. suffixing reduplication: ‘Note that 

the order of stem and affix is not immediately certain for forms like fatifeti and many 

others, in which a complete bivocalic stem is reduplicated. I treat these forms as suffixed 

to be uniform with trivocalic denotatives in this group [which] are clearly suffixed, as in 

131

consisting of the final two syllables of the stem, or as gemination of the initial 

consonant. 5 Examples are shown below (Kennedy 2002: 4). 6 

(12) a. feragi ‘spread’ fferagi ‘to be spread’ 

ŋüsü-ri ‘snort it’ ŋŋüsü ‘to snort’ 

pilegü-w ‘bundle it’ ppilegü ‘to be bundled’ 

sawee-y ‘go along side of it’ ssawe ‘to go along side of’ 

tabee-y ‘follow it’ ttabe ‘to follow’ 

b. βugo-si ‘tie it’ βugo-βugo ‘to tie’ 

faŋošo ‘current’ faŋošo-ŋošo ‘to have a little current’ 

file-ti ‘stir it’ file-file ‘to stir’ 

masowe ‘hard’ masowe-sowe ‘to be strong’ 

perase ‘to splash’ perase-rase ‘to scatter’ 

In Kennedy’s analysis, the location of the denotative affix (prefix or suffix) is lexically 

determined. Then, based on its position the shape of the allomorph is determined by 

phonological constraints, such that prefixes consist of gemination of the initial consonant, 

while suffixes consist of a reduplication of the final two syllables of the stem. In order to 

get the shape of the allomorphs, Kennedy first assumes that stems that take the suffixing 

form are lexically specified as having a ‘flag’, whose surface correlate is stress, and 

which is required to be anchored at the left edge of the word. In effect, this means that the 

leftmost syllable of the stem will be the first stressed syllable in the word if the stem has a 

flag. By using constraints such as LAPSE and ALLFEETRIGHT, this analysis ensures that 

perase-rase.’ (2002: 4). Sohn (1975) does not say explicitly whether reduplication is 

infixing or suffixing, and the examples do not include morpheme breaks. 

5 

Though one may be tempted to suggest that initial gemination and reduplication are two 

different morphological processes, Sohn implies (1975: 132) that they are the same in 

stating that ‘transitive stem minus thematic element + reduplication equals neutral verb. 

(Here reduplication in neutral verbs includes that of a whole-stem, a part, and consonantdoubling)’ 

(emphasis mine). 

6 

According to Sohn (1975: 132-133), these are not the only two ways to create an 

intransitive verb. For example, some verbs undergo truncation and/or vowel changes 

(e.g., petangi ‘land on it’ → pat ‘to land’, fangiuli ‘wake him up’ → fang ‘to wake up’). 

Others appear to undergo prefixing reduplication (e.g., torofi ‘catch it’ → tottor ‘to 

catch’, telati ‘free it’ → tettal ‘to free’). 

132

the suffix allomorph can only be a two syllable reduplicant. Similarly, the constraints 

ALLFEETRIGHT and *SEGMENT (similar to *STRUC) cause the prefix always to surface as 

initial gemination. 

Thus, in an indirect way, the shape of the denotative affix in Woleaian is 

determined by stress. This example does not fit neatly into any particular category of 

PCSA, nor is it of a type predicted by the P >> M model: even though the OT analysis is 

perfectly compatible with the P >> M approach, the ‘flag’ and the constraint that aligns it 

to the left edge of the word is stipulative and language-specific, so even though it can be 

analyzed using phonological constraints in OT, it is not exactly ‘optimizing’ in the way 

that proponents of the P >> M model view PCSA to be. It could only be argued to be 

optimizing in a very language-specific sense. 

Another example of stress-related PCSA that does not appear to be 

phonologically optimizing is found in Dutch. In Dutch, two suffixes that are commonly 

used to derive adjectives from nouns are -isch /is/ and -ief /iv/ (Booij and Lieber 1993). 

For nouns ending in ie /i/, the distribution of the adjectival suffix allomorphs is 

determined by the stress pattern of the stem, as follows: -isch is used if the stem has final 

stress, while -ief is used if the stem-final syllable is unstressed. 7 Examples are shown 

7 

It is possible that the different patterning of forms in (13)a vs. (13)b is based in some 

way on the origin of the stems rather on their stress pattern, since those in (a) have a 

Greek origin while those in (b) have a Latin origin. This generalization is not 

contradicted among the nouns in ie provided by Booij and Lieber (1993), though as will 

be seen in (14), there are other, non-ie nouns taking -isch that are not of Greek origin, e.g. 

álgebra (from Arabic). In fact, additional forms found in Booij (2001: 128) uphold the 

alternative generalization: agressief ‘aggressive’, demonstratief ‘demonstrative’, imitatief 

‘imitating’, and suppletief ‘suppletive’ are based on stems of Latin origin, as is 

conservatief ‘conservative’ (Booij 2001: 106); filosofisch ‘philosophical’ (Booij 2001: 

108) is of Greek origin. Thus, the trigger of allomorphy could actually be a diacritic 

indicating the perceived origin of the stem, rather than its stress pattern. Already Booij 

133

elow (Booij and Lieber 1993: 25). 8 

(13) a. sociologíe ‘sociology’ sociolog-isch ‘sociological’ 

blasfemíe ‘blasphemy’ blasfem-isch ‘blasphemous’ 

allergíe ‘allergy’ allerg-isch ‘allergic’ 

b. prevéntie ‘prevention’ prevent-ief ‘preventive’ 

constrúctie ‘construction’ construct-ief ‘constructive’ 

integrátie ‘integration’ integrat-ief ‘integrating’ 

Here, there is no indication that the pattern of allomorphy is optimizing, since both 

allomorphs have the shape /iC/ and therefore would not affect the number of syllables for 

the creation of stress feet. The allomorphy therefore appears to be non-optimizing, unlike 

the Dutch plural example described earlier. 

Also unlike the plural, this pattern seems crucially to refer to a derived 

phonological property. Recall that the distribution of plural suffixes was able to be stated 

in such a way that we did not have to refer to any derived properties, since the stress in 

one class of words had to be lexically specified. In this case, however, it is more difficult 

to avoid reference to a derived property. As discussed by Booij and Lieber (1993: 25,41), 

the -ief suffix must be the more specific allomorph because it only occurs productively 

(2001: 76) has proposed a feature [-native] to mark non-native stems, since these are the 

only stems to which non-native suffixes can attach (Booij 2001: 75), and also features 

[+French] and [+Germanic] ((2001: 104). Perhaps other features could be used to 

distinguish the Latin and Greek vocabulary. Under such an analysis, distrubution of -isch 

vs. -ief would not be an example of phonologically conditioned allomorphy. Thanks to 

Andrew Garrett for pointing out this alternative generalization. 

8 Booij and Lieber (1993) do not mark stress on the derived forms. Based on Booij (2001: 

114), it may be inferred that forms with the -isch suffix assign stress to the last stressable 

syllable before the suffix, but this is a generalization referring to the ‘native’ suffix -isch, 

while the -isch that attaches to non-native stems is listed separately (Booij 2001: 76) as a 

non-native suffix. Therefore, Booij may be assuming there are two -isch suffixes (though 

it is not clear how one would distinguish them). Stress is marked on one form with -ief: 

cònservatíef ‘conservative’ (Booij 2001: 106), suggesting that -ief gets the main stress in 

words where it occurs, but there are not enough examples to demonstrate that this 

generalization holds for all forms with -ief. 

134

with stems ending in ie, whereas -isch also occurs with other types of stems, as in the 

examples below (Booij and Lieber 1993: 25). 

(14) proféet ‘prophet’ profet-isch ‘prophetical’ 

álgebra ‘algebra’ algebra-ïsch ‘algebraic’ 

organisátor ‘organizer’ organisator-isch ‘organizational’ 

Therefore, as Booij and Lieber argue, -ief must be the allomorph that imposes a 

phonological requirement on stems, and since this is the allomorph that occurs with stems 

having the predictable (penultimate) stress pattern, this means that the distribution of -ief 

refers to a derived phonological property. As mentioned earlier, the existence of such 

examples is predicted both by P >> M and by the subcategorization approach. Although 

the fact that this type of pattern is attested does not help us to choose between the 

competing ways of modeling PCSA, it is nonetheless important to document the fact that 

this type of example does in fact exist. 

In this section we have seen two examples of apparently non-optimizing stress- 

related PCSA. These cases are significant in that they fail to uphold the prediction of the 

P >> M model that PCSA is phonologically optimizing, unless we extend the notion of 

‘optimizing’ to such a degree that it is virtually meaningless. Another important point 

was made by the Dutch example, which is that stress-conditioned PCSA can involve 

reference to derived phonological properties. 

3.1.2 Summary 

In §3.1, we have seen eight examples of PCSA involving stress or tone. In some 

cases, in particular in the non-optimizing examples discussed in §3.1.1.4, the results of 

the survey contradicted a claim of the P >> M approach, namely that PCSA is 

135

phonological optimization. The failure of this claim to be upheld in the survey data is by 

now a recurrent theme, since this was also demonstrated in chapter 2. In general, then, it 

can be said that although not very many examples of tone- or stress-conditioned PCSA 

were attested in the survey (for the possible reasons discussed earlier), those examples 

that are attested tend to favor the subcategorization approach. In the following section, I 

demonstrate how this approach applies to some of the examples described above. 

3.2 Analysis 

Tone- and stress-conditioned suppletive allomorphy are easily modeled using the 

subcategorization approach as described and implemented in chapter 2. I will reiterate the 

basics of this approach here; for a fuller description, see chapters 1 and 2. 

Subcategorization is a mechanism in which an affix is prespecified with 

properties that are required to be present in any stem to which it will attach. These 

properties may be morphological, lexical, or otherwise; crucially, they may also be 

phonological. Subcategorization approaches have been advanced by Lieber 1980, 

Kiparsky 1982b, Selkirk 1982, Orgun 1996, and Yu 2003, among others. The basic 

approach here is the same; the one stipulation that seems to be necessary (though see 

chapter 2 for more discussion) is that an affix must be adjacent to the phonological 

element that it subcategorizes for. This requirement is covered by the Generalized 

Determinant Focus Adjacency Condition (GDFAC), which was proposed by Inkelas 

1990 building on a proposal of Poser (1985). The GDFAC is given below. 



136

With this in mind, let us move to a demonstration of how the subcategorization 

approach works for cases of tone- and stress-conditioned PCSA. Tone-conditioned 

allomorphy is modeled in the same way as feature-conditioned allomorphy discussed in 

chapter 2, except that the features that are subcategorized for are on the tonal tier rather 

than being features belonging to particular segments. Below I give subcategorization 

frames for each of the two 1sg allomorphs in Mixtepec Mixtec, accounting for their 

distribution. 

(16) Mixtepec Mixtec 1sg construction A 

[[ L#] stem yù 1sg clitic ] 1sg word 

Mixtepec Mixtec 1sg construction B (‘elsewhere’) 

[[ ] stem L 1sg clitic ] 1sg word 

An alternative approach would encode homophony avoidance directly into the synchronic 

account of the allomorphy via an OT constraint that penalizes any word that is 

homophonous with another word in the same paradigm, such as Crosswhite’s (1999) 

ANTI-IDENT, shown below. 

(17) ANTI-IDENT: For two forms, S1 and S2, where S1 ≠ S2, ∃α, α∈S1, such that 

α ≠ ℜ(α). 

One could account for the allomorphy by ranking ANTI-IDENT (a phonological constraint) 

ahead of the morphological constraint that requires morphological categories to be 

expressed uniformly on all stems. This would be an example of the P >> M ranking 

schema (McCarthy and Prince 1993a,b) discussed earlier. Such an alternative analysis 

would be problematic because, as shown above, it is possible to explain the apparent 

homophony avoidance effect based on its historical development. Therefore, the P >> M 

analysis provides an unnecessary and contradictory explanation. The subcategorization 

approach, on the other hand, does not include a synchronic explanation for the pattern, 

137

and therefore it accounts for the distribution of allomorphs without duplicating or 

contradicting the historical explanation for the pattern. 

Stress-conditioned suppletive allomorphy under this approach amounts to 

subcategorization for a stressed syllable, as I will demonstrate here for the Dutch 

example. A different option would be to view this as subcategorization for a foot, so that 

Dutch plural allomorphy would fall under the category of foot-conditioned suppletive 

allomorphy, to be described in chapter 4. I have rejected this option because the Dutch 

plural suffixes seem to be sensitive not only to the presence of a stem-final foot, but 

specifically to the stress pattern within the foot. In comparing kanón ‘gun’ with kánon 

‘canon’, I am assuming that each word is exhaustively parsed into a single, disyllabic 

foot, and that the difference is that kanón exceptionally has an iambic rather than a 

trochaic foot structure. Given this, it is not the alignment of the foot that differentiates the 

two words, but rather which syllable of the foot is stressed. 

I give subcategorization frames below showing the distribution of -en vs. -s. 

(18) Dutch plural construction A 

[[ σ́ #] stem en pl suffix ] pl word 

Dutch plural construction B (‘elsewhere’) 

[[ ] stem s pl suffix ] pl word 

In this analysis, -s is the ‘elsewhere’ allomorph. This proposal is corroborated by the fact 

that loanwords are inflected with -s rather than -en, though the choice of a default 

allomorph is not uncomplicated (see van Wijk 2002 for discussion). 

This is how the subcategorization approach handles tone- and stress-conditioned 

allomorphy. As was demonstrated, the approach straightforwardly captures the patterns 

of allomorphy described in this chapter. Although the P >> M model is also capable of 

138

modeling the same patterns, the subcategorization approach has an advantage in that, as 

has been pointed out throughout this chapter as well as chapter 2, the predictions of the 

subcategorization approach seem to fit better with the survey results being presented here 

than do the predictions of the P >> M approach. 


In this chapter, I have presented examples of suppletive allomorphy conditioned 

by tone or stress. It was noted that there are fewer such examples than we may have 

expected. Some reasons for this were discussed, including the fact that some cases 

potentially falling into the stress-conditioned category may overlap with the cases to be 

described in chapter 4, thereby artificially reducing the number of cases of stress- 

conditioned PCSA. Still, the lack of cases of tone-conditioned PCSA is surprising since 

phonological phenomena involving tonal alternations are quite common cross- 

linguistically; this remains an unresolved issue. 

The examples in this chapter were demonstrated to be straightforwardly 

analyzable using the subcategorization approach outlined in chapter 2. For the example of 

tone-conditioned allomorphy in Mixtepec Mixtec, it was argued that a subcategorization 

analysis is superior to an OT account based on homophony avoidance because the 

explanation provided by the OT account is made unnecessary by the historical 

explanation outlined above. In the following chapter, I discuss examples of allomorphy 

conditioned by elements of the prosodic hierarchy and demonstrate that the 

subcategorization approach is easily and fruitfully extended to those cases as well. 

139

Chapter 4: Prosodically conditioned suppletive allomorphy 

Probably the best-known cases of phonologically conditioned suppletive 

allomorphy (PCSA) are those that are prosodically conditioned. Kager (1996) devotes an 

article to the topic of syllable-counting allomorphy (SCA), which he claims can actually 

be reduced to foot-parsing (though, as will be seen in this chapter, not all cases of 

syllable-counting allomorphy can be handled in this way). In this chapter, I consider 

cases of prosodically conditioned suppletive allomorphy as follows. First, in §4.1, I 

present results of a cross-linguistic survey of the phenomenon. In §4.2, I discuss the 

historical development of a set of related examples in Pama-Nyungan languages, showing 

how a diachronic analysis of the phenomenon clears the way for a straightforward and 

unified approach to prosodically conditioned suppletive allomorphy that is consistent 

with the analysis of PCSA presented in chapters 2 and 3. In §4.3, I review the synchronic 

analysis of PCSA and demonstrate how it accounts for the prosodically conditioned 

examples presented in this chapter, contrasting it with the Output Optimization approach 

espoused by, e.g., Kager (1996). The chapter is concluded in §4.4. 


In this section, I present examples of prosodically conditioned suppletive 

allomorphy revealed by a cross-linguistic survey. I define ‘prosodically conditioned’ to 

refer to allomorphy that is sensitive to the presence of an element of the prosodic 

hierarchy, namely, a mora, syllable, or foot. The survey revealed 137 total cases of PCSA 

140

in 67 different languages, listed in the Appendix. Of these cases, 59 cases from 39 

different languages involve prosodic conditioning and will be discussed in this chapter. 1 

Under the P >> M approach, the same phonological constraints are responsible for 

both regular phonological processes and PCSA. Therefore, we expect that the same 

phonological constraints that drive foot parsing should also be manifested in PCSA with 

approximately the same frequency cross-linguistically. For example, constraints such as 

TROCHAIC (requiring feet to have a stressed-unstressed pattern), FOOTBINARITY 

(requiring feet to have exactly two moras or two syllables), and PARSESYLL (requiring 

every syllable to be part of a foot) are expected to drive PCSA in a considerable number 

of languages since they are commonly invoked in analyses of foot parsing not involving 

suppletive allomorphy. The effects of these constraints could be manifested in PCSA in a 

variety of ways. For example, in some language where suppletive allomorphs of an affix 

differ in whether they are inherently stressed or not, TROCHAIC could cause the unstressed 

allomorph to be used next to a stressed stem syllable and the stressed allomorph to be 

used next to an unstressed stem syllable. FOOTBINARITY combined with PARSESYLL 

could cause affixes with an odd mora/syllable count to occur with stems with an odd 

mora/syllable count (and likewise for affixes and stems with an even mora/syllable count) 

in order to ensure that words have an even mora count, allowing every syllable to be 

parsed into a binary foot. 

As we will see, some of these effects are exhibited in one or more languages, 

confirming a prediction of the P >> M approach and suggesting that it is on the right 

1 Note that the sum of the number of examples discussed in each chapter is greater than 

the total number of examples found in the survey; this is because a few of the examples 

are discussed in more than one chapter. 

141

track in accounting for prosodically conditioned PCSA. For example, as will be 

discussed, we do find cases where PCSA results in words with an even syllable or mora 

count. This was the impetus for Kager’s (1996) claim that ‘syllable-counting allomorphy’ 

reduces to optimal foot parsing. However, a perhaps surprising result of this survey is the 

finding that not all examples of PCSA conditioned by syllable or mora count involve 

some complementary relation between the stress pattern or the number of moras in the 

stem and in the affix. As will be seen in §4.1.1.3, there are examples in which suppletive 

allomorphs are distributed according to the syllable or mora count of the stem but where 

each allomorph contributes the same number of moras to the word. Thus, foot parsing 

constraints alone will not be able to handle all of the cases to be discussed here. In some 

cases, it must be stipulated that a certain allomorph occurs only with stems having a 

certain number of moras, syllables, or feet. This issue will be discussed again in §4.3. 

This section is organized as follows. First, I present the attested examples in 

§4.1.1. In §4.1.1.1, I discuss examples in which the distribution of suppletive allomorphs 

relates to foot parsing. In §4.1.1.2, I discuss examples that relate to word minimality. In 

§4.1.1.3, I discuss examples where the distribution of allomorphs seems to be arbitrary or 

non-optimizing, in that there is no clear relationship between the shape of the allomorphs 

and their distribution. The examples are summarized in §4.1.2. 


The examples to be presented in this section exhibit essentially the same 

properties as those presented in chapters 2 and 3. Some major generalizations that will 

emerge here, each of which has already been discussed in the preceding two chapters, are 

142

the following. First, the generalization that conditioning applies from the ‘inside-out’ is 

maintained here. PCSA in affixes is triggered by elements that are closer to the root, and 

not by elements that are farther away; we also find no instances of affix-conditioned stem 

allomorphy here. 

Second, we find here (as in chapters 2 and 3) some examples of opaque 

conditioning, where the element that conditions the occurrence of a particular allomorph 

is lost or changed in the surface form. This supports the claim that PCSA is sensitive to 

input elements, not surface elements. 

A third generalization made in chapters 2 and 3 is that PCSA is constrained by 

locality, so that conditions on PCSA always occur at the edge of the stem at which the 

affix attaches. Most of the examples in this chapter do not bear on this particular 

generalization, since they involve conditioning by general properties of the stem (such as 

syllable count) that cannot be said to occur at one edge or the other. However, there is at 

least one type of example that could contradict the locality generalization if it were 

attested. This would be a case in which allomorphy is conditioned by the weight of a 

particular syllable in the stem, but the affix undergoing allomorphy is at the opposite 

edge, and the weight generalization cannot be characterized in any more general way that 

would refer, e.g., to the overall mora count of the word. 2 This type of example would 

contradict the generalization, but no such cases are attested. 

A final generalization that has been made in previous chapters is that PCSA is not 

phonologically optimizing. In those chapters, numerous examples were presented in 

2 For example, imagine a case in which PCSA in a suffix is conditioned by whether the 

stem-initial syllable is light or heavy, and it does not matter how many moras or syllables 

intervene between the initial syllable and the right edge. 

143

which allomorphy did not seem to follow from any well-known or well-motivated 

phonological constraint. It was argued that even though many examples do appear on 

their surface to be optimizing, the existence of the non-optimizing examples is sufficient 

to contradict the predictions made by the P >> M model of PCSA because it is not able to 

handle non-optimizing cases without recourse to ad hoc, stipulative constraints that 

violate the spirit of the model. In this chapter, it seems that a higher percentage of the 

examples are optimizing than in chapters 2 and 3. For this reason, in §4.2 I explore the 

possibility of accounting for the optimization via a historical explanation, thereby freeing 

the synchronic model from having to account for it. 

I have given an overview of some generalizations that will emerge in the data 

presented in this chapter; I move now to discuss the examples. I begin in §4.1.1.1 with 

examples of PCSA that relate to foot parsing. 

4.1.1.1 Foot parsing 

This survey revealed a number of cases in which the outcome of PCSA can be 

viewed as maximizing the number of syllables that are parsed into feet. One could claim 

that PARSESYLL is the constraint responsible for these effects. However, this only works 

when PARSESYLL is combined with some other constraint such as FOOTBINARITY, since 

if a foot can consist of any number of syllables or moras, then the shape of an allomorph 

will have no bearing on whether or not all of the syllables in a word can be parsed into 

feet. In this section, I discuss several examples where such an analysis is possible; the 

ordering of examples is roughly by continent or geographical region. 

144

Among the languages surveyed, Latin provides the clearest examples of this type 

of PCSA. Mester (1994) discusses three cases of allomorphy in Latin, which he argues to 

result from the drive to parse all syllables into feet that are exactly bimoraic. The first of 

these involves the ‘theme vowel’, which has two allomorphs, -ī- and -ĭ-. This is a vowel 

that occurs in the so-called io-verbs and is sometimes referred to as the ‘stem vowel’ but 

is explicitly assumed here to be a distinct morpheme from the root. According to Mester 

(1994: 24), the form of the theme vowel is predictable ‘to a large extent’ based on the 

shape of the root: -ĭ- occurs when preceded by a single light syllable (as in capĭmus 

‘catch’), while -ī- occurs when preceded by a heavy syllable or two light syllables (as in 

audīmus ‘hear’ and aperīmus ‘open’, respectively). Mester does not make a case for why 

this allomorphy must be assumed to be suppletive, except to say that ‘[g]iven its lexically 

restricted nature, there is no motivation for positing a general phonological rule’ (1994: 

26). Thus, Mester’s implicit criterion for whether or not a particular instance of 

allomorphy is suppletive seems to be that rule-derived allomorphy can only result from 

general phonological rules. Considering how phonologically similar the two allomorphs 

are in this particular example, the Latin theme vowel alternation could be considered to 

be rule-derived rather than suppletive. However, I discuss this example here since, even if 

the theme vowel alternation is not suppletive, Mester discusses two other cases from 

Latin that are unambiguously suppletive and are argued to follow from the same 

principles as the theme vowel alternation. Furthermore, as was discussed in chapter 2, 

given that the P >> M model accounts for suppletive and non-suppletive allomorphy in 

the same way, it is not absolutely crucial that we exclude all non-suppletive examples 

from the survey. 

145

The proposed explanation for the theme vowel alternation is that it serves to avoid 

the ‘trapping’ configuration where a light syllable is located between a well-formed 

bimoraic foot and either the beginning or end of the word, or another well-formed 

bimoraic foot. Such configurations ‘trap’ the light syllable and prevent it from being 

parsed into a well-formed foot. In the environments in which the -ī- theme vowel occurs 

(after a heavy syllable or two light syllables), the root can be parsed into a single 

bimoraic foot. If the light syllable (monomoraic) theme vowel -ĭ- occurred after such a 

root, it would not be parsable into a bimoraic foot, and would therefore be trapped. In 

these environments, the -ī- theme vowel occurs instead, and since it contributes a heavy 

syllable, this theme vowel can be parsed into its own bimoraic foot, with the result that all 

syllables of the word are parsed into well-formed feet. 

Another Latin example described by Mester that is more likely to be suppletive 

involves suffixes that form abstract nouns (1994: 43), with the shapes -ia and -iēs. 

According to Mester, many roots can take either allomorph (for example, māter-ia and 

māter-iēs ‘matter’ apparently occur in free variation). However, some roots can only take 

one or the other, and there is a prosodic principle behind their selection. The –iēs 

allomorph is avoided after heavy syllables (Mester 1994: 43), which Mester attributes to 

their inherent trapping configuration: if -iēs were added to a root ending in a heavy 

syllable, the coda of the root would syllabify with the /i/ of the suffix, forming a light 

syllable, which would then be trapped between two heavy syllables since the ēs of the 

suffix would form a second heavy syllable. In this situation, -ia is used instead (as in 

grāt-ia ‘grace’), avoiding the trapping configuration since the entire suffix is parsed into a 

bimoraic foot. 

146

A third and final example from Latin discussed by Mester involves the perfect 

stem of the second (ē) conjugation (1994: 44). According to Mester, the generalization is 

that perfect stems take the u-perfect form except in exactly those cases where the use of 

the u-perfect would result in a trapping configuration. The u-perfect consists of a verb 

root followed by -ŭ-, as in mon-ŭ-ī ‘warn (1sg perfect)’. This u-perfect is the predominant 

way to form a perfect stem, but there are two other ways of forming a perfect stem. The 

first is the s-perfect (Mester 1994: 45), which adds -s to the root, as in auk-s-ī ‘enlarge 

(1sg perfect)’. The second is through reduplication, which Mester describes as an archaic 

formation; an example is spopondī ‘vow’ (

ases select -cita/-cito, while stems with three or more syllables select -ita/-ito. Examples 

are shown below (Harris 1979: 291). 

(1) madre-cita *madr-ita dinosaur-ito *dinosaurie-cito 

saurie-cito *saur-ito comadr-ita *comadre-cita 

There are several interesting points to be made regarding this example. First, note that the 

two affix allomorphs themselves both would appear to contribute two syllables to the 

stem, but the initial vowel of the -ita/-ito suffix triggers deletion of the stem-final vowel, 

resulting in a net increase of one syllable to the word, as compared to two syllables when 

-cita/-cito is used. Thus, the apparent complementarity between the stem and suffix is not 

a property of the suffix itself, but is rather the result of a phonological rule that is 

triggered by one suffix allomorph but not the other. This type of ‘indirect’ optimization is 

predicted by P >> M since hiatus resolution and allomorph selection are manifested 

simultaneously and can therefore freely interact. 

Note also, however, that not all words are optimized by this distribution of 

allomorphs. In these dialects of Spanish, stems with four syllables take the -ita/-ito form 

according to Harris; this would result in words having an odd syllable count (five 

syllables). Thus, the pattern cannot be claimed to be optimal for every stem in the 

language. In a P >> M analysis, one would have to find some way to force four-syllable 

stems to take -ita/-ito even though this would be phonologically less optimal than using 

the -cita/-cito form. In a subcategorization analysis, this would be unproblematic since 

the -ita/-ito suffixes can subcategorize for a disyllabic stem with -cita/-cito as the 

‘elsewhere’ allomorphs. 

148

A final point to be made regarding the Spanish example is that, as pointed out by 

Harris, there are some surface counterexamples to the pattern, in which trisyllabic stems 

are treated as disyllabic for the purposes of diminutive allomorphy. These examples 

apparently all begin with es, so Harris explains them by proposing that diminutive 

allomorph selection occurs prior to the epenthesis of initial e before s in these words 

(Harris 1979: 291). This type of interaction with a phonological process is not predicted 

to occur under P >> M, since allomorph distribution is supposed to be determined 

simultaneously with the application of phonological processes (and using the same 

phonological constraints). The fact that a phonological epenthesis rule renders the PCSA 

conditioning opaque in this set of examples constitutes an argument against the P >> M 

approach. 3 

Another example where PCSA produces words with an even syllable count is 

found in Saami (Lappic, Norway; Dolbey 1997). In this language, several person marking 

suffixes have allomorphy determined by the syllable count of the stem (Dolbey 1997). 

Some examples are shown below (Dolbey 1997: 103, 105). 

(2) jearra- ‘to ask’ veahkehea- ‘to help’ ‘even’ ‘odd’ 

1du je:r.re.-Ø veah.ke.he:-t.ne Ø ~ -tne 

2du jear.ra.-beaht.ti veah.ke.hea-hp.pi -beahtti ~ -hppi 

2pl jear.ra.-beh.tet veah.ke.he:-h.pet -behtet ~ -hpet 

3pl pret je:r.re.-Ø veah.ke.he:-d.je Ø ~ -dje 

passive je:r.ro.-juv.vo veah.ke.hu-v.vo -juvvo ~ -vvo 

In every case, stems with an even syllable count take suffix allomorphs 

contributing an even number of syllables (including zero), while stems with an odd 

syllable count take allomorphs contributing an odd number of syllables. The result, as 

pointed out by Dolbey, is to avoid unfooted syllables. An interesting point made by 

3 Several other examples of opaque conditioning were discussed in chapter 2 (§2.1.2.6). 

149

Dolbey regarding this example is that it provides strong evidence for cyclic, rather than 

global, evaluation of stems in terms of even syllable count. This is seen where two of the 

suffixes listed above are combined with a root having an even syllable count, such as in 

the example je:r.ro-juv.vo-beaht.ti ‘ask-Passive-2du’. In such cases, the whole word 

would also have an even syllable count even if the monosyllabic allomorph of each suffix 

were used, as in *je:r.ru-v.vo-hp.pi. It is not enough for the whole word to have an even 

syllable count, because the even syllable count requirement is evaluated at each level of 

morphological constituency. Since the stem of affixation for the 2du suffix in the form 

*je:r.ru-v.vo-hp.pi is the ungrammatical (trisyllabic) stem *je:r.ru-v.vo-, the whole word 

is ungrammatical even when the addition of the monosyllabic passive suffix allomorph 

would correct the syllable count problem for the purposes of global evaluation. 

Thus, interestingly, although Saami does confirm the prediction of P >> M that 

PCSA should result in words with an even syllable count, it also provides an argument 

against the simplest and most common implementation of P >> M. Generally, P >> M 

analyses do not refer to morphological constituency, but rather show the stem and all 

affixes present in the input, implying that they have equal status and therefore equal 

potential to trigger or undergo a change in the output. The Saami example demonstrates 

that this type of model is too simplistic; in an OT analysis, the grammar needs to be able 

to evaluate well-formedness each time an affix is added to the stem. 

Another case that may involve foot parsing constraints is found in Finnish, though 

the analysis is not as straightforward as in the Saami example discussed above. The 

Finnish illative has three allomorphs: -Vn, -hVn, and -seen (Karlsson 1999: 112-113). The 

-seen allomorph occurs after polysyllabic stems ending in a long vowel. The -hVn 

150

allomorph occurs after monosyllabic stems and after plural stems where the root ends in a 

vowel (yielding a Vi sequence at the end of the root after the addition of the -i plural 

suffix). The -Vn allomorph occurs with other stems. In each of the two allomorphs 

containing ‘V’, this vowel surfaces with the same quality as the immediately preceding 

vowel. The distribution of the three allomorphs is illustrated in the examples below 

(Karlsson 1999: 112-113). 

(3) talo-on ‘into the house’ koulu-un ‘to school’ 

kaupunki-in ‘to the town’ lehte-en ‘into the newspaper’ 

kunta-an ‘into the commune’ korkea-an ‘into the high’ 

maa-han ‘into the country’ tie-hen ‘to the road’ 

työ-hön ‘to work’ suu-hun ‘into the mouth’ 

pullo-i-hin ‘into the bottles’ kalo-i-hin ‘into the fish (pl.)’ 

vapaa-seen ‘into the free’ harmaa-seen ‘into the grey’ 

perhee-seen ‘into the family’ tietee-seen ‘into science’ 

taivaa-seen ‘to heaven/into the sky’ rikkaa-seen ‘into the rich’ 

The use of the -hVn allomorph with monosyllabic stems produces words with two 

syllables, and its use after plural stems prevents VVV sequences from arising. Perhaps 

the distribution of -hVn vs. -Vn can be analyzed in terms of a requirement that a foot have 

two syllables and a constraint against VVV; in fact, one could write rules deriving one 

allomorph from the other, so that the relationship between these two allomorphs is not 

necessarily suppletive. The distribution of -seen, however, does not follow from these 

constraints. Perhaps there is also a separate constraint requiring a foot to have at least two 

moras, so that with a stem such as vapaa, if the stem has to be parsed into a single 

disyllabic foot, then the suffix form is -seen, which can form its own bimoraic foot. Thus, 

the Finnish example may involve constraints on foot structure, but the overall pattern of 

allomorphy in the illative suffixes is complicated and would require several separate 

constraints in a P >> M analysis. 

151

Three examples of syllable count-conditioned allomorphy are found in Estonian 

(Finno-Lappic, Estonia; Mürk 1997). These examples are discussed by Kager (1996) in 

support of his argument that syllable-counting allomorphy is driven by foot-parsing 

constraints. In Estonian, the genitive plural takes the form -te after stems with two or four 

syllables, and -tte with three syllables; the partitive plural takes the form -sit after stems 

with two or four syllables, and -it (or by a vowel mutation) after stems with three 

syllables (Kager 1996: 157). The partitive singular is -Ø or -t after ‘even-numbered’ 

stems (with the -t allomorph occurring only after monosyllables) and -tt after ‘odd- 

numbered’ stems (Kager 1996: 168). Examples given below with page numbers in 

parentheses are from Mürk 1997; the remaining examples are from Kager 1996: 157, 

168. 

(4) Genitive plural Partitive plural Partitive singular Gloss 

maa:-te (74) ma-it (74) maa:-t ‘land’ 

aas:ta-tte (149) aas:ta-it (149) aas:ta-tt ‘year’ 

häda-te (77) häda-sit (77) häda-Ø (77) ‘trouble’ 

paraja-tte paraja-it paraja-tt ‘suitable’ 

raamattu-tte raamaattu-it raamattu-tt ‘book’ 

atmirali-te atmirali-sit atmirali-Ø ‘admiral’ 

telefoni-te telefoni-sit telefoni-Ø ‘telephone’ 

Kager’s analysis is that ‘even-numbered’ stems are those that can be exhaustively 

parsed into disyllabic feet, while ‘odd-numbered’ stems cannot be exhaustively parsed 

into disyllabic feet (1996: 158). The suffix forms selected by even-numbered stems do 

not end up being parsed into feet. The suffix forms selected by odd-numbered stems, on 

the other hand, are parsed into a foot along with the final syllable of the stem. The fact 

that in these cases the suffix allomorph contributes a coda consonant to the stem-final 

syllable, making that syllable heavy, follows from the preference for disyllabic trochaic 

152

feet. Thus, Kager claims that this is not actually a case of syllable counting, but of foot 

parsing. 

Kager makes the stronger claim that syllable count-conditioned allomorphy in 

general reduces to foot parsing optimization. We will see that this claim does not hold up 

for all cases since, as will be demonstrated in §4.1.1.3, there are several cases of 

allomorphy based on syllable count where the distribution of allomorphs is arbitrary 

rather than optimizing. Nonetheless, the Estonian examples do appear to involve a 

‘conspiracy’ that optimizes foot parsing. However, note that in each of the sets of 

allomorphs discussed by Kager (partitive singular, partitive plural, and genitive plural), 

the two allomorphs are very similar to each other phonetically. This suggests that we 

might view these as examples of rule-derived rather than suppletive allomorphy (though 

the rules needed to produce the allomorphy would probably have to be specific to each 

morpheme). 

In Turkana (Nilotic, Kenya; Dimmendaal 1983), we find an example in which the 

apparent complementarity between the size of the stem and the size of an affix cannot be 

attributed to the regular application of phonological processes to a single underlying affix 

form. According to Dimmendaal (2000), Turkana exhibits syllable count-conditioned 

suppletive allomorphy in the affix used to form abstract nouns. As shown in the examples 

below (Dimmendaal 2000: 166), roots of shape CVC take the -ɪ̀sɪ́ allomorph ((5)a), while 

roots of shape CiVjCiVjC take the -ù allomorph ((5)b), and other roots are partially 

reduplicated ((5)c). 

153

(5) a. á-rɛ́ŋ-ɪ̀sɪ́ ‘redness’ (-rɛ́ŋ ‘be red’) 

á-jɔ́k-ɪ̀sɪ́ ‘kindness’ (-jɔ́k ‘be kind, good’) 

b. á-gógóŋ-ù ‘strength’ (-gógóŋ ‘be strong’) 

á-bábár-ù ‘saltiness’ (-bábár ‘be salty’) 

c. á-cʊ́kʊ́l-ʊ́l ‘depth’ (-cʊ́kʊ́l ‘be deep’) 

á-ŋáráb-áb ‘roughness’ (-ŋáráb ‘be rough’) 

Dimmendaal’s (2000) characterization of the distribution implies that the allomorphy 

may result from a four-syllable requirement for abstract nouns. However, the distribution 

is actually more complicated than represented here, as indicated by Dimmendaal (1983). 

In addition to the types mentioned above, there are other root types for which the 

selection of the abstract noun suffix is not predictable from the factors outlined above. 

For example, in addition to CVC roots, roots with ‘an epipatetic high front vowel’ take 

the -ɪ̀sɪ́ allomorph (Dimmendaal 1983: 270), while roots with ‘an epipatetic vowel /o/ or 

/a/’ (1983: 271) take the -ù allomorph, as do stems containing a habitual extension. Thus, 

some morphological and/or stem class conditioning must be involved here in addition to 

the phonological conditioning; though we do not have examples of all of these different 

types of stems, it seems that not every nominalized form ends up with four syllables. 

Still, it seems that there is a strong tendency for nominalized forms to have four syllables, 

suggesting that a constraint enforcing this would not be implausible. One could view this 

as being foot-based, since a four-syllable word is easily parsed into two disyllabic feet; it 

could also fall into the category of minimality-related conditioning since it seems that 

there is a preference for a specific number of syllables in nominalized words. 

In some examples found in the survey, both the shape and the placement of the 

allomorphs are relevant to syllable count and foot parsing. For example, in Nancowry 

154

(Mon-Khmer, Nicobar Islands; Radhakrishnan 1981), 4 the causative is marked by either a 

prefix or an infix, depending on whether the stem is monosyllabic or disyllabic, 

respectively (Radhakrishnan 1980: 54). When the stem is monosyllabic, the causative is 

marked by a ha- prefix, as in the examples below. 

(6) ha-káh ‘to cause to know’ ha-míh ‘to cause to rain’ 

ha-míɁ ‘to cause to be soaked’ ha-suáh ‘to cause to burn’ 

When the stem is disyllabic, the causative is marked by an -um- infix, which ‘overwrites’ 

the rhyme of the stem-initial syllable, as in the examples below. 

(7) p-um-Ɂɯ̃ y ‘to cause to have a bad smell’ paɁɯ̃ y ‘bad smell’ 

p-um-rẽ ‘to cause something to be flat’ pirẽ ‘flat’ 

p-um-lóɁ ‘to cause someone to lose s.t.’ palóɁ ‘lose’ 

The result of this allomorphy appears to be that causative forms always have two 

syllables. As with the Turkana example above, the result of the allomorphy could be 

analyzed in terms of both foot parsing (since causative forms end up with two syllables, 

which presumably form a single foot) and minimality constraints. 

Radhakrishnan (1980: 55) discusses some interesting ‘double causative’ 

examples, which are formed by prefixing ha- and infixing -um-. The double causative can 

only apply to monosyllabic roots. Examples are shown below (Radhakrishnan 1980: 55). 5 

4 Recent discussions of Nancowry phonology can be found in Steriade 1988, Alderete et 

al 1999, Hendricks 1999, and Inkelas and Zoll (2005: 223-224); these deal primarily with 

reduplication. 

5 The meaning of the double causative is not clear since the examples are given out of 

context, but based on the glosses, it does not necessarily involve the introduction of two 

new causative agents as might be expected. 

155

(8) ha-káh h-um-káh ‘to cause to know’ 

ha-míh h-um-míh ‘to cause to rain’ 

ha-Ɂuáh h-um-Ɂuáh ‘to cause to cough’ 

ha-sɯ́ l h-um-sɯ́ l ‘to frighten’ 

On the surface, these examples might appear to provide an argument in favor of using a 

foot parsing or minimality/maximality constraint to analyze the Nancowry causatives, 

since the combination of two causatives still always results in a disyllabic word. 

However, the subcategorization approach also provides a straightforward explanation for 

the fact that one instance of each causative allomorph is used in the double causative 

construction. If the ha- allomorph subcategorizes for a monosyllabic stem, then this 

allomorph will apply first to the root; then, when a second causative is applied to the 

(now disyllabic) stem, the -um- allomorph will be used since ha- can only attach to a 

monosyllabic stem. Thus, the double causative does not provide an argument in favor of 

one analysis or the other. However, it does constitute an example in which allomorph 

selection is sensitive to a phonological property contributed by another affix (in this case, 

a different allomorph marking the same morphological category), since the use of -um- in 

the double causative is conditioned by the extra syllable contributed by ha-. Other 

examples involving this type of interaction between affixes have already been discussed 

(for example, the Saami examples discussed earlier in this section, and the Turkish 

‘stacked’ causatives discussed in chapter 2), but this is worth pointing out nonetheless. 

One example that has been claimed to relate to foot parsing is found in 

Martuthunira (Pama-Nyungan, Australia; Dench 1995), where the ‘full laden’ suffix takes 

the form -warlaya following trimoraic and larger stems, and -warla following bimoraic 

stems (Dench 1995: 65). Dench claims that this pattern is ‘based on a preference for an 

156

even pattern of stressed-unstressed syllables in the word’ (1995: 65). 6 Some examples are 

shown below (from Dench 1995: 94, except where noted). 

(9) jinyji-warla ‘fat, plump, obese’ marrari-warlaya (no English given) 

fat-FULL word-FULL (Dench 1995: 38) 

ngungku-warla ‘strong’ kunkuwarra-warlaya ‘full to 

weight-FULL honey-FULL bursting with 

honey’ 

Dench appears to claim that the -warla form occurs with bimoraic roots so that the 

resulting form will have an even number of moras, which can be parsed into exactly two 

feet. However, since the -warlaya allomorph occurs with stems with an even mora count 

of more than three moras (as in kunkuwarra-warlaya), Dench’s explanation does not 

extend to all cases since the distribution of allomorphs does not always result in an even- 

numbered syllable count. Therefore, the allomorphy seems not to refer to feet, but 

specifically to mora count, and in general it cannot be considered to optimize footing. 

Furthermore, note that this pattern cannot be analyzed in terms of either minimality or 

maximality, since the pattern here is the exact opposite of the type of complementarity 

that we have seen in some previous examples: the smaller suffix attaches to the smallest 

stems, while the larger suffix attaches to larger stems. This example lends itself to a 

subcategorization approach, since we can easily characterize the -warla suffix as 

subcategorizing for a stem with two moras (or, to avoid allowing the grammar to ‘count’, 

6 Since Dench’s claim refers specifically to stress patterns, it might have made sense to 

discuss this example along with the other stress-conditioned examples in chapter 3. 

However, the notion of alternating stress does relate directly to foot parsing, and 

furthermore, I argue against Dench’s explanation here. Therefore, I have opted to include 

the example in this section along with other similar examples where allomorph 

distribution is conditioned by syllable/mora count. 

157

we could assume that the bimoraic stem is parsed into a foot before the addition of the 

suffix, so that the suffix attaches to a stem consisting of a single foot). 

A final example of apparent foot-based allomorphy is found in Shipibo (Panoan, 

Peru; Elías-Ulloa 2004) . In Shipibo, the ergative suffix has two allomorphs, -n and -nin 

(Elías-Ulloa 2004: 1). The -n allomorph occurs when the noun stem has an even syllable 

count, while the -nin allomorph occurs with noun stems with an odd syllable count. Some 

examples are shown below (Elías-Ulloa 2004: 3). 

(10) bakɨ-n ‘child-ERG’ atapa-nin ‘hen-ERG’ 

The distribution of ergative allomorphs results in words with an even syllable count, as in 

the Saami case discussed earlier in this section. Thus, this is another example that could 

be analyzed in terms of foot parsing, since the distribution of allomorphs results in a 

configuration of syllables that is optimal for foot formation. 

In this section, we have seen several examples of allomorphy conditioned by 

syllable or mora count (17 examples from nine different languages) whose net result is 

that every syllable can be parsed into a binary foot. This is an important finding since it is 

the first type of allomorphy that we have seen so far in which a prediction of the P >> M 

model is borne out in a significant way. Since this is one of the more well-known types of 

PCSA, it should not be surprising that such cases exist. In fact, we might have expected 

to find more examples, since the effects of foot parsing constraints are evident in the 

phonological systems of so many of the world’s languages. However, one point should be 

kept in mind, which is that although the P >> M model predicts and explains examples of 

this type, the subcategorization approach is also able to account for these examples. The 

difference is that in a subcategorization analysis, we would not have a built-in 

158

explanation for the apparent drive towards exhaustive parsing into binary feet; the foot- 

parsing effect would be (from a synchronic perspective) coincidental. This missed 

generalization might seem to be a serious disadvantage of the subcategorization 

approach. However, as will become clear when we discuss the non-optimizing examples 

in §4.1.1.3, there is a good reason to avoid using foot-parsing constraints to analyze 

syllable-counting allomorphy: many examples of syllable-counting allomorphy seem to 

be arbitrary and make no difference to foot parsing, and yet in terms of their behavior and 

the factors that condition allomorph distribution, they look exactly like the examples seen 

in this section. The P >> M model fares poorly when we attempt a unified account of all 

syllable-counting allomorphy. 7 This issue will be taken up in §4.3. 

4.1.1.2 Minimality 

Another set of constraints expected to play a role in prosodically conditioned 

PCSA is minimality constraints. These constraints stipulate that a word must have some 

minimum number of moras, syllables, or feet in order to be a well-formed word. 

An example of allomorphy that could be analyzed as resulting from a minimality 

constraint is found in Spanish, where the nominalizing suffixes -ez and -eza are 

distributed according to the syllable count of the stem. As seen in the examples below, 

the form -ez is used when the citation form of the corresponding adjective has three or 

7 It has been suggested (Booij 1998) that the P >> M approach could be used for the 

optimizing type of PCSA, while another mechanism could be used to handle the nonoptimizing 

types. I argue against this approach in §4.3 and again in §6.4 on the grounds 

that it is unsatisfying to the linguist looking for a unified solution to the problem of 

syllable-counting allomorphy, and also that it is scientifically unsound. 

159

more syllables, while the suffix -eza is used when the adjective has one or two syllables 

(Aranovich et al 2005). 

(11) rigido rigid-ez ‘rigid’ 

estupido estupid-ez ‘stupid’ 

vil vil-eza ‘vile’ 

franco franqu-eza ‘truthful’ 

gentil gentil-eza ‘gentle’ 

real real-eza ‘royal, regal’ 

One generalization is that nominalized forms have no fewer than three syllables (e.g., 

*vil-ez is unacceptable). Thus, in a P >> M account, the distrubution of -eza could be 

attributed in part to a minimality constraint (3 syllable minimum for nominalized words) 

if we assume that -ez is the default or ‘elsewhere’ allomorph. However, this would not 

explain why disyllabic consonant-final stems such as gentil take -eza instead of -ez. The 

problem for a P >> M account is that the number of input stem syllables is obscured by 

the deletion of the stem-final vowel that is triggered by the addition of the vowel-initial 

suffix. 8 In a subcategorization analysis, we can state that the -ez suffix subcategorizes for 

a stem with at least three syllables, while -eza is the ‘elsewhere’ allomorph. 9 Thus, 

although we can still say that minimality may be playing a role here, the surface patterns 

cannot all be explained in this way, and the example does not lend itself to a standard, 

surface-based P >> M analysis involving minimality. 

In Qafar (Cushitic, Ethiopia; Parker and Hayward 1985, Hayward 1998), the 

indefinite genitive suffix exhibits PCSA that may be driven in part by word minimality 

8 

Aranovich et al (2005) argue that the distribution of allomorphs is driven by a constraint 

on inputs (stem + suffix) requiring them to consist of two binary feet. A problem with 

this type of approach is that it requires the phonology to build feet out of the stem and 

suffix prior to the evaluation of the input by the phonology, which should logically take 

place before any phonological processes (such as foot formation) apply to the input. 

9 

Alternatively, -eza could subcategorize for stems of one or two syllables, with -ez as the 

elsewhere allomorph. 

160

considerations. As seen in the examples below (Hayward 1998: 630), among masculine 

stems, 10 those that are monosyllabic and consonant-final take the -ti suffix ((12)a). Those 

that are polysyllabic and consonant-final undergo no segmental changes, though stems 

that bear lexical accent are de-accented in the indefinite genitive ((12)b). Those that are 

polysyllabic and vowel-final undergo de-accentuation and have their final vowel replaced 

by [i] ((12)c). 

(12) a. /ħan/ ‘milk’ ħan-tí dala ‘(a) milk gourd’ 

b. /áʕan/ ‘frog’ aʕán iba ‘(a) frog’s leg(s)’ 

/danan/ ‘donkey’ danán iba ‘(a) donkey’s leg(s)’ 

c. /kúta/ ‘dog’ kut-í ɖagor ‘(a) dog’s fur’ 

One result of the allomorphy seen here is that no possessor noun has fewer than two 

syllables. This could account for why monosyllabic possessor stems take the -ti suffix 

instead of -Ø. However, this still does not explain why polysyllabic vowel-final stems 

behave differently from their consonant-final counterparts in taking -i rather than -Ø, 

since they would still be disyllabic without the -i suffix. This aspect of the allomorphy 

does not follow from minimality. A possible way of accounting for the distribution of -i 

vs. -Ø is to propose a two-syllable maximality constraint as well, so that possessor nouns 

must have both a minimum and a maximum of two syllables. Suppose that the constraint 

MORPHEXPR (Rose 2000a) requires every morphological category to be marked by some 

affix, and suppose that the stress shift that applies to the forms in (12)b and (12)c is 

secondary and does not count as an expression of the category. Then, the distribution of -i 

vs. -Ø can be explained because while -i can fuse with a stem-final vowel, yielding 

10 Feminine nouns take a -C suffix, with the quality of the consonant determined by the 

initial consonant of the possessed stem that follows, unless this stem is vowel-initial, in 

which case the suffix consonant surfaces as [h] by default (Hayward 1998: 630). 

161

disyllabic words in (12)c, using -i with the consonant-final stems in (12)b would add a 

third syllable, which would violate the disyllabic maximality constraint. 

An important point to note regarding this example is that it requires us to posit 

both a minimality and a maximality constraint, plus MORPHEXPR, to account for the 

single fact that all possessor nouns have exactly two syllables. Thus, in a P >> M 

analysis, the drive towards disyllabic words here is the combined effect of multiple 

constraints; this is not a very straightforward way of getting at the syllable count 

generalization. 11 

One example that has been argued to result from word maximality is found in 

Miya (Chadic, Nigeria). According to Schuh (1998), the maximal native verb is 

trimoraic, accounting for the fact that trimoraic roots do not undergo reduplication in the 

pluractional form as other roots do. As shown below, roots with the shape /Ca/ undergo 

reduplication, surfacing as CəCa ((13)a). Roots with the shapes /C1əC2(ə)/ and /C1əC2a/ 

also undergo reduplication, surfacing as C1aC1əC2a ((13)b). Roots of the shape /C1aC2/ 

undergo vowel lengthening, surfacing as C1aaC2a ((13)c). However, roots with the form 

/C1VC2(ə)C3(ə)/ undergo a vowel change but no reduplication or vowel lengthening, 

surfacing as C1aC2(ə)C3a ((13)d). Examples are from Schuh 1998: 175-176. 

11 Note also that we cannot use foot parsing constraints to account for the pattern, since 

there do exist trisyllabic stems in the language which, according to Hayward’s (1998) 

generalization, pattern with disyllabic stems, rather than with monosyllabic stems as 

would be the case if foot parsing constraints were responsible. 

162

(13) a. za ‘enter’ à zə-za-ya sáy ‘he entered’ 

pa ‘collect’ à pə-pa sáy ‘he collected’ 

b. tsər ‘stop’ à tsa-tsə́r-a-yà sáy ‘he stopped’ 

pər ‘cut’ à pa-pə́r-à sáy ‘he cut’ 

zəna ‘spread to dry’ à za-zən-a sáy ‘he spread to dry’ 

c. tlakə ‘scrape’ à tláak-à sáy ‘he scraped’ 

dzar ‘disperse’ sə̀bə dzaar-a-ya sáy ‘the people dispersed’ 

d. tsəryə ‘step on’ à tsary-á say ‘he stepped on’ 

kərmə ‘scoop up’ à karm-a sáy ‘he scooped up’ 

Schuh’s explanation for the failure of the roots in ((13)d) to reduplicate is compelling but 

does not explain why the C1aC2 roots also do not reduplicate. If the generalization is that 

the pluractional involves reduplication except in the case that the three-mora restriction 

prevents reduplication from applying, then we cannot account for the behavior of C1aC2 

roots. One possible conclusion is that the form of the pluractional for C1aC2 and 

C1VC2(ə)C3(ə) is lexically conditioned (despite the fact that the generalization is stated in 

terms of the phonological shape of the roots), while the allomorphy that distinguishes CV 

roots from CVC/CVCV roots (excluding those with a lexically specified pluractional 

form) is conditioned by mora count and may relate, at least historically, to Schuh’s 

proposed trimoraic constraint on verbs. 

New Zealand Maori (Polynesian, New Zealand; Biggs 1961) provides another 

example of stem-affix complementarity that could be driven by word minimality. In New 

Zealand Maori, the form of the inceptive prefix is kaa- before monosyllabic stems and 

disyllabic stems where both syllables are short (i.e., before bimoraic stems), and ka- 

elsewhere. Examples of the kaa- allomorph are shown below (Biggs 1961: 29); Biggs 

does not provide examples with ka-. 

(14) kaa-pái ‘becomes good’ kaa-ríro ‘is carried away’ 

163

This distribution can be restated as referring to mora count: bimoraic stems take kaa-, 

while larger stems take ka-. A possible way of accounting for this would be to posit a 

constraint requiring inceptive forms to have a minimum of four moras. Note, however, 

that we can use such a constraint in an OT analysis of the allomorphy without P >> M, 

since the allomorphs are not necessarily in a suppletive relationship. Since they are so 

phonetically similar to each other, these allomorphs could be related to a single 

underlying prefix, /ka-/, whose vowel undergoes lengthening when it occurs before a 

stem with two moras. Therefore, this example may result from minimality, but it is not an 

ideal candidate for a P >> M analysis. 12 

In this section, we have seen five examples in which allomorphy appears to be 

driven by word minimality or maximality. However, some of these examples do not 

necessarily involve suppletive allomorphy, and as was discussed, some cannot easily be 

accounted for using the P >> M model. In §4.1.1.3, I discuss several more examples that 

defy analysis using P >> M since they involve allomorphy that is not optimizing with 

respect to any well-known P constraints. 

12 Note also that the distribution of allomorphs cannot be analyzed in terms of foot 

parsing. This is because there do exist stems in the language that are polysyllabic and 

have an initial long vowel. Some examples (from Harlow 1996, with page numbers in 

parentheses) include we:ra ‘whale’ (13), te:pu ‘table’ (19), pu:hi ‘shoot’ (47), wa:hi 

‘place’ (49), ta:ne ‘man’ (51), and ma:ku: ‘wet’ (18). According to Biggs’ (1961) 

generalization, stems with these shapes (CV:CV(:)) take the ka- prefix; this is 

problematic for foot parsing since the single light syllable contributed by ka- is prevented 

from being parsed into a well-formed foot when the stem-initial syllable is heavy. 

164


In the preceding sections, we have seen examples in which the distribution of 

allomorphs appears to have some motivation based on foot parsing or word 

minimality/maximality. In this section, we will consider some examples of PCSA in 

which the motivation for allomorphy is unclear, suggesting that the distribution of 

allomorphs is arbitrary rather than optimizing. This is not to say that one could never 

come up with some possible motivation for any of these examples. However, unlike some 

of the examples presented above in §§4.1.1.1-2, there is no immediately obvious 

phonological constraint that can be identified as driving the allomorphy. 

One example of PCSA in which the motivation for allomorph distribution is 

unclear is found in Turkana (Dimmendaal 1983). In Turkana, plural is marked on nouns 

by a prefix ŋa- or ŋI- and one of three suffix allomorphs, -a, -In, or -I. 13 As seen in the 

examples below (Dimmendaal 1983: 226, Dimmendaal 2000: 44, 235), the -a allomorph 

occurs with stems of shape CV(C)VC ((15)a), the -In allomorph occurs with stems of 

shape CV(C)V ((15)b), and the -I allomorph occurs with stems of shape C(C)VC ((15)c). 

(15) a. ŋa-kìɲàŋ-a ‘crocodiles’ ŋa-ŋàjɛ̀p-a ‘tongues’ 

ŋa-ŋàsɛ̀p-a ‘placentas’ ŋɪ-kààl-a ‘camels’ 

b. ŋa-kwap-ɪnˋ ‘lands’ ŋi-rot-inˋ ‘roads’ 

ŋa-rʊk-ɪn ‘humps’ ŋa-kʊŋ-ɪn ‘knees’ 

c. ŋi-wɔrʊ-ɪˋ ‘cloths’ ŋi-suro-iˋ ‘dik-diks’ 

ŋa-pɔɔ-ɪˋ ‘hares’ 

If coda consonants are moraic, then the use of -In with monosyllabic stems could be seen 

as the result of a requirement that plural forms have a minimum of four moras. However, 

13 I is a high, front vowel whose value for ±ATR is determined by the stem vowels. 

165

this still does not explain the distribution of -a vs. -I. This could have something to do 

with sequences of vowels that are permitted to be adjacent to each other; perhaps [a] is 

dispreferred after another vowel, so -i is used in that environment instead. We can 

therefore possibly explain the distribution of allomorphs in these examples using 

common constraint types, but such an explanation would be contingent on some 

unsubstantiated assumptions. 

Another example in which allomorph distribution is not obviously optimizing is 

found in Kimatuumbi (Odden 1996). As was also discussed in chapter 2, the perfective in 

Kimatuumbi has four allomorphs: -i̹te, -i̹i̹le, -i̹i̹ye, and the infix -i̹-. The allomorphs are 

distributed as follows (Odden 1996: 51-53). Monosyllabic verbs (of shape (C)VC or 

(C)VVC), as well as stems whose final vowel is long, take -i̹te. Examples are shown 

below (Odden 1996: 51). 

(16) ni̹-chól-i̹te ‘draw’ ni̹-tí̹n-i̹te ‘chop’ 

ni̹-bálaang-i̹te ‘count’ ni̹-chéleew-i̹te ‘be late’ 

Polysyllabic stems (except those with a final long vowel) take the -i̹- infix, which is 

inserted into the stem-final syllable, as in the examples below (Odden 1996: 51). These 

forms also select the final vowel e rather than a. 

(17) ni̹-áandi̹-i̹-ke ‘write’ ni̹-béli̹-i̹-ke ‘bear’ 

ni̹-chíili̹-i̹-ye ‘be late’ ni̹-sábi̹-i̹-te ‘beat’ 

As was discussed in chapter 2, glide-final roots take -i̹i̹le, while stems ending in yaan or 

waan take -i̹i̹ye (Odden 1996: 53). 

The distribution could be construed as resulting from a requirement that 

perfective forms, including their prefixes, have four syllables. However, the four-syllable 

166

generalization does not apply in all cases, since disyllabic stems whose final vowel is 

long take -i̹te, thereby ending up with five syllables. Also, all roots ending in yaan or 

waan take the -i̹i̹ye suffix, resulting in words with more than four syllables. Furthermore, 

if the four-syllable requirement were invoked to explain the use of -i̹te with monosyllabic 

stems, then we would also have to explain why this allomorph is used after stems with a 

final long vowel. Therefore, much of the pattern seen here cannot be attributed to a 

minimality constraint, and appears to be arbitrary. 

In Nancowry (Radhakrishnan 1980), there is another example of syllable count- 

based PCSA in which the distribution of allomorphs appears to be arbitrary. In 

Nancowry, there the instrumental is marked by either -an- or -in-. Monosyllabic stems 

select -an-, while disyllabic stems (including those built from a monosyllabic root plus 

causative prefix) select -in-, which overwrites the rime of the initial syllable 

(Radhakrishnan 1980: 60-63). Some examples are shown below (Radhakrishnan 1980: 

61, 62-63). 

(18) káp ‘to bite’ k,an,áp ‘tooth’ 

tián ‘to file’ t,an,ián ‘a file’ 

léʔ ‘to catch’ l,an,éʔ ‘an object to catch with’ 

rɯ́ k ‘to arrive’ r,an,ɯ́ k ‘a vehicle’ 

kaʔáp ‘to close’ k,in,ʔáp ‘a trap’ 

tikóʔ ‘to prick’ t,in,kóʔ ‘pin, needle’ 

sahuáŋ ‘cool’ s,in,huáŋ ‘something that cools’ 

ha-kiãk ‘to inflate’ h,in,kiãk ‘a pump’ 

In this example, there appears to be no functional motivation for the distribution of the 

allomorphs. As with the causative examples given above in §4.1.1.1, instrumental forms 

appear always to have two syllables, so one might argue that the distribution of 

allomorphs is ‘optimizing’ with respect to syllable count. However, once we distinguish 

167

etween affix shape and affix placement, the distribution of allomorphs seems arbitrary 

as in many other examples described in this section. The -in- allomorph overwrites the 

first vowel of a disyllabic stem, yielding a disyllabic word, while the -an- allomorph is 

inserted after the first consonant of a monosyllabic stem without overwriting the vowel. 

However, the infixes themselves could just as easily pattern in the opposite way; their 

actual behavior is no more natural than if the -in- suffix were inserted after the first 

consonant of a monosyllabic stem and the -an- suffix were to overwrite the first vowel of 

a disyllabic stem. Thus, the distribution of the allomorphs in terms of their shape does not 

appear to be functionally motivated. 

Another example involving apparently arbitrary allomorph distribution is found in 

Martuthunira, where there are two different forms of the collective suffix (Dench 1995: 

§6.1.5, §6.3.2) on ‘L-conjugation’ verbs that are distributed according to the size of the 

verb stem (1995: 38). Bimoraic stems take -yarri, while larger stems take -lwarri. Some 

examples are provided below (Dench 1995: 38). 

(19) karta-yarri stab-COLL thuulwa-lwarri pull out-COLL 

thani-yarri hit-COLL kartatha-lwarri chop-COLL 

This is an interesting case because not only is there not a complementary relationship 

between the size of the stem and affix, but if anything, the situation is the exact reverse: 

larger stems take the larger allomorph. 14 There does not seem to be any way to predict 

the distribution of allomorphs based on their shape. 

14 Of course, if coda consonants are not moraic, then the notion of the ‘larger’ allomorph 

is not very significant (it would simply refer to the allomorph with more segments). 

However, it is worth pointing out this generalization, because it underscores the fact that 

this example does not show the kind of complementarity between the stem and affix that 

was seen in some examples in §§4.1.1.1-2. 

168

Another interesting example of syllable count-conditioned allomorphy in which 

the allomorphs contribute the same number of syllables to the stem is found in Kashaya 

(Oswalt 1960, Buckley 1994). As discussed in chapter 2, the durative in Kashaya exhibits 

PCSA conditioned both by syllable count and by the stem-final consonant. According to 

Buckley (1994: 328-329), the allomorphs are distributed as follows. The first split 

(described in chapter 2) is between consonant- and vowel-final stems. Among vowel- 

final stems, there is another split based on syllable count: monosyllabic stems take -cin’, 

while polysyllabic stems take -men’. Finally, in general, consonant-final stems take -an. 15 

Examples of vowel-final roots exhibiting syllable count-based allomorphy are provided 

below (Oswalt 1960: 212-213). 16 

(20) sime-sime-ciˑd-u ‘to keep on sprinkling’ 

buwi-ciˑd-u ‘to keep on stringing’ 

duk’ilci-meˑd-u ‘one to keep pointing once’ 

mohqa-méˑd-u ‘one to drive’ 

The stems characterized as monosyllabic appear in these examples to have more than one 

syllable; this leads Oswalt to characterize the distribution of durative allomorphs 

somewhat differently from Buckley (Oswalt states that -cin’ is used ‘after a vowel... 

when that preceding vowel is in the second syllable of the verb’ (Oswalt 1960: 212), 

while -men’ ‘...is used after a vowel... when that preceding vowel is in the third or later 

syllable of the word’ (Oswalt 1960: 212). However, vowel epenthesis may be argued to 

apply to stems only after the attachment of -cin’, so that the stem for attachment of this 

allomorph is actually monosyllabic as characterized by Buckely. This is an important 

15 

Among consonant-final stems, there also exists an elaborate subpattern of allomorphy, 

discussed in chapter 2, which seems to reduce to morphological conditioning. 

16 

Note that what Buckley refers to as -cin’ is generally transcribed as [ciˑd], while 

Buckley’s -men’ is transcribed as [meˑd]. 

169

point, because if -cin’ attaches to monosyllabic stems which later gain an extra syllable 

through epenthesis, then the condition for the attachment of -cin’ is rendered opaque. 

This then is another example of the type of opaque conditioning of PCSA that would be 

difficult to capture in a P >> M analysis; other examples of this type include the Spanish 

example discussed in §4.1.1.1, and several examples discussed in chapter 2 (§2.1.2.6). 

Zuni (isolate, New Mexico; Newman 1965) exhibits another example of possibly 

arbitrary PCSA. The form of the singular suffix that attaches to nouns in Zuni depends in 

part upon the number of syllables in the stem, as follows (Newman 1965: 23-24). Nouns 

of class 2 (which are all monosyllabic) take the suffix -mmeʔ ((21)a). Nouns of class 3 

(which are all polysyllabic) take -ʔe ((21)b). Among class 1 nouns, the distribution of 

allomorphs depends on syllable count. Those with one syllable take -ʔleʔ ((21)c). Those 

with two syllables take the allomorph -nne ((21)d). Examples are shown below (Newman 

1965: 24). 

(21) a. kʔe-mmeʔ ‘stalk’ 

b. sap-ʔe ‘box of dishes’ (< sapa) 

c. ɬi-ʔleʔ ‘sinew’ 

d. homa-nne ‘juniper leaf’ 

Though part of the distribution refers to lexical classes, the allomorphy within class 1 is 

phonological since it is determined by syllable count. However, note that the distribution 

of allomorphs does not affect the surface syllable count. Each suffix allomorph 

contributes exactly one syllable to the word, so their distribution does not complement 

the syllable count of the stem. It is possible that the generalization could have to do with 

mora count instead, since in the examples seen above, both of the class 1 singular forms 

may have four moras, if it is assumed that coda consonants contribute a mora. However, 

170

not all singular nouns fit this pattern (e.g., the example sap-ʔe in (21)b has only three 

moras). 

Another apparently non-optimizing example of PCSA is found in Tzeltal (Mayan, 

Mexico). According to Walsh Dickey (1999), perfective suffix allomorphs in Tzeltal are 

selected based on syllable count, but the shape of the affixes themselves does not seem to 

relate to syllable count: the variants are -oh (occurring with monosyllabic stems) and -ɛh 

(occurring with polysyllabic stems). Examples of the -oh variant are provided below 

(Walsh Dickey 1999: 328-329). 

(22) s-mah-oh ‘he has hit something’ j-il-oh ‘he has seen something’ 

s-pas-oh ‘he has made something’ s-kut͡ʃ-oh ‘she has carried it’ 

s-nut͡s-oh ‘he has chased something’ j-al-oh ‘he has told something’ 

s-tsak-oh ‘he has taken something’ s-jom-oh ‘he has gathered it’ 

s-nɛt’-oh ‘he has squashed something’ 

When the root is polysyllabic, the perfective suffix is -Eh, as shown below (Walsh 

Dickey 1999: 328). 

(23) s-majlij-ɛh ‘he has waited for someone’ s-mak’lin-ɛh ‘he has fed someone’ 

s-maklij-ɛh ‘he has listened to something’ s-tikun-ɛh ‘he has sent something’ 

The -ɛh variant is also used with a monosyllabic root when another suffix intervenes, 

resulting in a polysyllabic stem, as shown in the examples below (Walsh Dickey 1999: 

329, no interlinear glosses provided). 

(24) s-hol-intaj-ɛh ‘he has thought about it’ 

h-pak’-antaj-ɛh ‘I have patched it’ 

s-kut͡ʃ-laj-ɛh ‘she was carrying it repeatedly’ 

This constitutes another example in which the shape of the allomorphs does not relate to 

the phonological condition on their distribution. 

171

A final example of apparently non-optimizing PCSA is found in Axininca Campa 

(Arawakan, Peru; Payne 1981). In Axininca Campa, there is a genitive suffix that takes 

one of two forms, -ni or -ti, depending on whether the stem (not including the person 

marking prefix) has two vocalic moras or more than two, respectively. Examples are 

shown below (Payne 1981: 244-246). 

(25) no-sari-ni ‘my macaw’ a-toniro-ti ‘our palm (aquaje)’ 

p-ana-ni ‘your black dye’ pi-wiiri-ti ‘your bat’ 

o-çaa-ni ‘her anteater’ o-itairiki-ti ‘her wild pig’ 

i-yimi-ni ‘his squash’ n-aawana-ti ‘my mahogany’ 

Once again, there does not appear to be any motivation for the distribution of the two 

allomorphs; the distribution seems to be arbitrary. 

The examples in this section are of interest because, unlike the more well-known 

examples in the literature, these examples do not seem to involve phonological 

optimization, at least not in any obvious way. Even if some motivation were proposed for 

some of the examples seen in this section, the existence of several examples in which 

stem syllable count determines allomorphs that contribute the same number of syllables 

to the word is sufficient to demonstrate that, contrary to Kager (1996), syllable-counting 

allomorphy does not always reduce to foot parsing constraints. 

4.1.2 Summary 

We have discussed numerous examples in which the distribution of suppletive 

allomorphs is conditioned by syllable or mora count. Some of them appeared to be driven 

by foot parsing (§4.1.1.1), and others by word minimality or maximality (§4.1.1.2); some 

did not appear to optimize words in any readily apparent way (§4.1.1.3). One 

generalization that holds for all three of these example types is that every example 

172

discussed in this chapter involves affix allomorphy, rather than allomorphy in stems or 

clitics. This follows from a model of morphology in which words are built from the 

inside out; it does not follow from the simplest version of the P >> M model, in which 

unordered stems and affixes appear together in the input with no bracketing or other 

indication of the structure of the word. 

As has been discussed throughout this section, some of the examples seen here 

have other characteristics that suggest that the P >> M model is not the best way to 

handle them. For instance, some of the examples discussed above involve opaque 

conditioning, in which the condition for the attachment of an affix is rendered opaque by 

a phonological process. This shows that PCSA is not necessarily a process that optimizes 

surface forms, and it suggests that the selection of suppletive allomorphs occurs before 

the application of regular phonological processes. 17 Also, in several examples, 

particularly the examples of minimality/maximality-related allomorphy discussed in 

§4.1.1.2, we can only capture the phonological conditions using language-specific 

constraints. Thus, the patterns are only ‘optimizing’ if we define optimization to mean 

that some constraint can be written to account for them. For example, in the Miya 

example discussed in §4.1.1.2, the differential application of reduplication and vowel 

lengthening to stems with different mora counts was claimed to result from a constraint 

requiring verbs to have no more than three moras. However, PCSA driven by such a 

constraint can only be ‘optimizing’ in a language that prefers verbs to have no more than 

three moras; there is nothing inherently optimal about trimoraic verbs. Thus, although it 

17 In using the word ‘before’ here, I mean that phonological rules/processes apply to 

whatever is the outcome of allomorph selection; this does not have to be understood as a 

serially ordered sequence of events. 

173

is possible to provide a P >> M analysis for the examples in this chapter, many of the 

analyses would involve stipulative language-specific constraints. 

I have argued against using the P >> M model to account for these examples, and 

yet, as has been acknowledged throughout this chapter, there are many examples of 

prosodically conditioned suppletive allomorphy that truly do seem to optimize words in 

some way, particularly in the cases relating to foot structure discussed in §4.1.1.1. Since 

the subcategorization approach does not incorporate the notion of optimization, there is 

some sense in which the burden is on advocates of subcategorization to explain why these 

examples appear to optimize words; to say that this is a coincidence would be 

unsatisfying. Explaining the apparent optimization in prosodically conditioned PCSA is 

therefore somewhat of a challenge for the subcategorization approach. In the following 

section, I respond to this challenge by providing a historical explanation for a set of 

related examples of PCSA from the Pama-Nyungan languages of Australia. I will 

demonstrate that, despite the fact that several of the examples appear to be phonologically 

optimizing, many are not, and all have a common historical source involving individual 

and independently motivated changes not driven by prosodic optimization. 

4.2 Historical development of prosodically conditioned suppletive allomorphy: The 

case of Pama-Nyungan ergative suffix allomorphy 

In this section I examine the historical origins of the syllable- or mora-counting 

allomorphy exhibited by the ergative suffix in many Pama-Nyungan (henceforth PN) 

languages, which will be described here. This is an important case study because it can 

help shed light on an important question raised by the subcategorization approach to 

PCSA. One potential objection to the subcategorization approach is that it does not 

174

explain why many cases of PCSA, perhaps more than expected to occur by chance, seem 

to be phonologically optimizing. For example, as has been seen in this chapter, many 

cases of syllable- and mora-counting allomorphy pattern in such a way that the 

distribution of allomorphs results in words with an even syllable/mora count, which 

facilitates exhaustive parsing into binary feet. The subcategorization approach can easily 

capture these patterns, but does not explain the apparent foot-parsing optimization that 

seems to drive them. 

One may therefore wish to argue that although subcategorization is needed to 

handle the apparently arbitrary (non-optimizing) cases of PCSA, an output optimization 

approach should still be used for the apparently optimizing cases so as not to lose any 

phonological generalizations (such as the foot-parsing optimization mentioned above). 

However, the lack of explanatory power is not problematic for subcategorization if there 

is an external explanation for the apparent optimization. For example, if we can explain a 

case of apparent optimizing PCSA diachronically, then there is no need to incorporate 

this into the synchronic model of PCSA at the expense of a unitary account of the 

phenomenon. In this section I discuss the origins of PN ergative allomorphy, arguing that 

the apparent optimizing pattern exhibited in the ergative allomorphy in some modern PN 

languages arose by accident and not for the purpose of phonological optimization. 18 

18 I do not claim here that there is no bias in language acquisition towards finding patterns 

that are optimizing; this may be the case. However, as I will show, none of the specific 

changes that led to the modern patterns of PN ergative allomorphy were necessarily 

optimizing. As will be seen, in some modern languages, the result of these accumulated 

changes has the appearance of being optimizing; in other languages, however, the result 

of these changes looks arbitrary. 

175


In this section, I describe some examples of ergative and/or locative allomorphy 

in several PN languages. I include the locative suffixes here because in many languages, 

they pattern just like the ergative (in most instances, the only difference is that the 

locative suffixes have the vowel a where the ergative suffixes have u). 19 A fuller list of 

examples is provided by Sands 1996 (for ergative only); here I focus on a smaller set of 

languages that exemplify the range of patterns found throughout the continent. I have 

attempted to balance the survey by providing broad geographical coverage of the Pama- 

Nyungan area; below I give a map showing the locations of the languages to be discussed 

in this section. 

19 In fact, it may be the case that in at least some of the languages to be discussed here, 

the initial consonant(s) of the ergative and locative suffixes is actually a stem ending, and 

that the suffixes themselves consist only of vowels. This would mean that in some 

languages, the apparent parallel allomorphy between ergative and locative reduces to a 

single pattern of allomorphy in the stem ending that co-occurs with the ergative and 

locative suffixes. The resolution of this issue is not crucial to my analysis. 

176

Figure 1: Map showing location of examples of ergative allomorphy 

Map outline © 2006 Commonwealth of Australia (Geoscience Australia). 

Used with permission. 

Below I present a number of examples of ergative and locative allomorphy in languages 

represented on the map. The examples are organized according to the condition that 

determines the distribution of allomorphs. I begin with examples conditioned by syllable 

count. For readers not wishing to read through descriptions of all of the examples, I 

provide tables summarizing the patterns of ergative and locative allomorphy; these tables 

are given at the end of all of the descriptions. 

One example of syllable count-conditioned suppletion that looks on its surface to 

be driven by a minimality constraint (but probably is not) is the ergative marker in 

177

Dyirbal (Dixon 1972). In Dyirbal, the ergative case is marked by -ŋgu or -gu, as shown 

below. 

(26) yara-ŋgu ‘man-ERG’ yamani-gu ‘rainbow-ERG’ 

The phonological generalization is that -ŋgu occurs with disyllabic stems, while -gu 

occurs with larger stems. According to McCarthy and Prince (1990), although it has been 

proposed that this (and similar phenomena in other Australian languages) be analyzed as 

fulfilling a requirement that the ergative construction has four moras, such an analysis is 

problematic because consonants generally are not mora-bearing in Pama-Nyungan 

languages. 20 Thus, this putative example of allomorph selection based on minimality 

probably does not go through. This is a simple case of arbitrary suppletive allomorphy 

conditioned by a phonological factor (syllable count) not related to the shape of the 

allomorphs. 

Another syllable count-related example is found in Kaititj (Koch 1980), where the 

ergative/ instrumental/ locative suffix has three surface realizations (probably relating to 

two underlying forms). The -ŋ form occurs only after disyllabic stems, while the -l form 

is selected after stems with three or more syllables (and exceptionally after a few specific 

disyllabic stems) (Koch 1980: 265-266). In addition to this split, the -l suffix also has two 

realizations: [ɭ] following an apical consonant and [l] elsewhere (Koch note 9). I assume 

that these latter allomorphs correspond to a single underlying form, /-l/. The distribution 

of the ergative/instrumental/locative forms is shown below (N, M = prestopped apical 

and labial nasals, respectively; examples are from Koch 1980: 264-266). 

20 This is made explicit in analyses of, e.g., Yindjibarndi (Wordick 1982). 

178

(27) aki-ŋ ‘head-ERG’ ilt y i-ŋ ‘hand-ERG’ 

aNmi-ŋ ‘red ochre-ERG’ aynpi-ŋ ‘pouch-ERG’ 

aliki-l ‘dog-ERG’ aʈuyi-l ‘man-ERG’ 

aɣirki-l ‘sun-ERG’ awiyawi-l-amiɳ ‘dead person (pl.)-ERG’ 

aɣiri-ɭ ‘kangaroo-ERG’ ɭuNpiri-ɭ ‘forehead-ERG’ 

at̪iri-ɭ ‘two-ERG’ aMu-ŋi-t̪iri-ɭ ‘snake-ERG’ 

The -N and -l allomorphs could be related to a single underlying form by using some 

arbitrary rule referring to the number of syllables in the root, but this seems undesirable. 

Such a rule would not be motivated by any principle of the language or by any functional 

considerations. Furthermore, the conditioning environment has no relation to the shape of 

the allomorphs. Therefore, this example appears to be non-optimizing in nature. 

Diyari (Austin 1981) marks ergative case with the suffixes -ndu, -li, and -yali. The 

distribution of the allomorphs is determined both semantically and phonologically (based 

on syllable count and the stem-final segment) as follows (Austin 1981: 48-49). Female 

personal names take -ndu. Male personal names, non-singular common nouns with final u 

becoming [a], and singular common nouns with final u or i becoming [a] take -li. 

Singular common nouns with two, four, or five syllables with final u or i take -yali, and 

singular common nouns with three syllables with final u or i can take either -yali or -li. 

Examples are shown below (Austin 1981: 49). 

(28) t̹iɽimiri -ndu ‘(woman’s name)-ERG’ 

waŋamiri-li ‘(man’s name)-ERG’ 

t̪aɽiwuɭa-li ‘youth.DUAL-ERG’ 

t̪aɽiwaɽa-li ‘youth.PL-ERG’ 

kanku-yali ‘boy-ERG’ 

wadukat̪i-yali ‘emu-ERG’ 

yapakaɲt̹in̪t̪u-yali ‘fear-EXCESS-PROP-ERG’ 

kan̪ina-li ~ kan̪ini-yali ‘mother’s mother-ERG’ 

A number of examples also refer to mora count rather than syllable count. For 

example, in Warlpiri (Nash 1986), the ergative is marked by -ngku after bimoraic stems, 

179

and by -rlu after longer stems. Examples are shown below (page numbers from Nash 

1986 are given next to each example). 

(29) ngurrpa-ngku ‘unknowing-ERG’ (34) nguurrpa-rlu ‘throat-ERG’ (34) 

ngapa-ngku ‘water-ERG’ (41) nyanungu-rlu ‘he-ERG’ (235) 

ngarrka-ngku ‘man-ERG’ (234) Jakamarra-rlu ‘(name)-ERG’ (217) 

kurdu-ngku ‘child-ERG’ (227) kurdu-jarra-rlu ‘child-DUAL-ERG’ (231) 

The locative in Warlpiri exhibits the same pattern as the ergative, except that the 

locative suffixes have a where the ergative suffixes have u. Examples are shown below 

(page numbers from Nash 1986 are shown next to each example). 

(30) ngulya-ngka ‘hole-LOC’ (30) karti-ngka ‘cards-LOC’ (203) 

karru-ngka ‘creek-LOC’ (203) pirli-ngka ‘rock-LOC’ (176) 

yali-rla ‘that-LOC’ (176) turaki-rla ‘vehicle-LOC’ (145) 

Yalikarangu-rla ‘(placename)-LOC’ (132) 

manangkarra-rla ‘spinifex plain-LOC’ (102) 

In Nhanda (Blevins 2001), ergative and instrumental suffix allomorphy is 

conditioned by the minimal vs. non-mininal word stem distinction, again in terms of 

moras (Blevins 2001: 48, 50). Ergative is marked by -(ng)gu with bimoraic stems and -lu 

with longer stems, as shown below (ng is deleted following a homorganic NC cluster 

(Blevins 2001: 48-49)). 

(31) nyarlu-nggu ‘woman-ERG’ nguutu-lu ‘horse-ERG’ 

uthu-nggu ‘dog-ERG’ arnmanu-lu ‘man-ERG’ 

munda-gu ‘poor-ERG’ kurndi-waa-lu ‘club-COM-ERG’ 

Instrumental exhibits the exact same allomorphs and pattern of allomorphy (Blevins 

2001: 50-51). Examples are shown below. 

(32) mara-nggu ‘hand-INST’ yadiwa-lu ‘axe-INST’ 

wana-nggu ‘digging stick-INST’ yawarda-lu ‘kangaroo-INST’ 

mambu-gu ‘bone-INST’ arnmanu-lu ‘man-ERG’ 

The locative in Nhanda is marked by -(ng)gu; there is no -lu allomorph (Blevins 2001: 

52). 

180

(33) malu-nggu ‘shade-LOC’ mambu-gu ‘bone-LOC’ 

uthudu-nggu ‘ground-LOC’ marnda-gu-tha ‘buttocks-LOC-1sgOBL’ 

In some cases, allomorph distribution refers both to mora count and to another 

phonological factor. For example, the locative suffix in Martuthunira (Dench 1995) 

exhibits allomorphy based on both the root-final segment and mora count. Bimoraic 

stems with a final vowel take the suffix -ngka, stems with three or more morae with a 

final vowel take -la, and there is further allomorphy among consonant-final stems 

depending on the consonant (Dench 1995: 64). The effector suffix has the same shape as 

the locative in all of these contexts, except that in each case, the effector suffix has /u/ 

rather than /a/. Some examples are provided below (Dench 1995: 38). This example is 

discussed further in chapter 2 because allomorph selection is sensitive to consonant 

features and the C/V distinction in addition to mora count. 

(34) nguu-ngka ‘face-LOC’ kaara-la ‘hip bone-LOC’ 

nharnu-ngka ‘sand-LOC’ malarnu-la ‘shade-LOC’ 

muyi-ngku ‘dog-EFF’ muyira-lu ‘dingo-EFF’ 

tharnta-ngku ‘euro-EFF’ mirntirimarta-lu ‘goanna-EFF’ 

Since codas are not moraic, the distribution of allomorphs has no effect on the number of 

moras in the surface forms. 

A similar example is found in Yindjibarndi (Wordick 1982), where the locative is 

marked on common nouns by -ŋka or -la (Wordick 1982: 56). The -ŋka allomorph occurs 

with monosyllabic roots ending in a vowel and with disyllabic roots ending in a vowel if 

each syllable contains only a single short vowel. The -la allomorph occurs with roots 

having three or more morae and with consonant-final roots. Some examples are given 

below (examples are from Wordick 1982: 63-65 and have been adapted into phonetic 

notation). 

181

(35) parlu-ŋka ‘cliff-LOC’ 

warkamu-la-ŋu ‘work-LOC-ABL’ maatha-la ‘boss-LOC’ 

kuɲciri-la-mpa ‘one-LOC-TOP’ 

Coda consonants are not considered to be moraic (Wordick 1982: 40), so although the 

allomorphy is based partly on mora count, the choice of the locative allomorph does not 

alter the mora count of the word: each contributes one mora. In terms of mora count, 

then, the allomorphy does not contribute to well-formedness. However, the V/C sensitive 

aspect of the distribution is optimizing with respect to syllable structure, since the 

selection of -la after consonant-final roots prevents the creation of three-consonant 

clusters (which would occur if -ŋka were suffixed to a consonant-final root, assuming that 

/ŋk/ is not a single segment but two separate segments). Three-consonant clusters do not 

exist in the language (Wordick 1982: 14). 

Yet another very similar example is found in Yingkarta (Dench 1998), where the 

ergative and locative markers exhibit allomorphy based on mora count and the final 

segment (Dench 1998: 15-16). Vowel final stems with two moras mark ergative with the 

suffix -ngku, and locative with -ngka. Vowel-final stems wtih three or more moras mark 

ergative with -lu and locative with -la. Stems with a final consonant take the ergative 

suffix -Su and locative -Sa (where S is a stop homorganic with the preceding consonant; 

ergative and locative examples with stem-final consonants are limited to apical nasals and 

laterals in Dench’s corpus, so it is not known which allomorph is selected by rr-final 

stems). Examples are shown below (Dench 1998: 19-21, except where noted). 

(36) Ergative 

ngampu-ngku ‘stick-ERG’ mayu-ngku ‘child-ERG’ 

kutharra-lu ‘two-ERG’ partirri-lu ‘returning-ERG’ 

majun-tu ‘turtle-ERG’ 

182

Locative 

jurla-ngka ‘scrub-LOC’ kuru-ngka ‘eye-LOC’ 

parrumpa-la ‘wattle-LOC’ paapa-la ‘guts-LOC’ 

kurtan-ta ‘bag-LOC’ 

(Dench 1998: 29) 

In several cases, the distribution of ergative or locative allomorphs refers solely to 

the final segment of the stem, and in particular, to whether the stem is consonant- or 

vowel-final. For instance, as discussed in chapter 2, the ergative in Warrgamay (Dixon 

1980: 266) is marked by the suffix -ŋgu after a vowel-final stem and by -du after a 

consonant-final stem, as shown below. 

(37) ŋulmburu-ŋgu ‘woman-ERG’ wurrbi-ŋgu ‘big-ERG’ 

wurrbi-bajun-du ‘very big-ERG’ 

This distribution results in avoidance of three-consonant clusters. 

A similar pattern is found in Yidiɲ (Dixon 1977). Ergative is marked by -ŋgu after 

a vowel-final stem, or -du after a consonant-final stem (d assimilates to a preceding stem- 

final nasal (Dixon 1977: 126)). Examples are given below (Dixon 1977: 126-127). 

(38) wagud̹a-ŋgu ‘man-ERG’ 

warabal-du ‘flying squirrel-ERG’ d̹ud̹u:m-bu ‘father’s sister-ERG’ 

The locative and allative allomorphs in Yidiɲ are selected based on whether the 

stem is consonant-final or vowel final (Dixon 1980: 296), though in this case (unlike in 

the ergative), the allomorphy cannot be claimed to optimize syllable structure in any way 

since both allomorphs have the shape -CV. Their distribution is as follows: -la occurs 

after vowel-final stems; -da occurs after consonant-final stems (d assimilates to the place 

of a preceding stem-final nasal (Dixon 1977: 129)). Examples are shown below. 

183

(39) digarra-la gabud̹u-la 

‘beach-LOC’ (Dixon 1980: 296) ‘white clay-LOC’ (Dixon 1977: 128) 

ward̹a:n-da mud̹a:m-ba 

‘boat-LOC’ (Dixon 1977: 129) ‘mother-LOC’ (Dixon 1977: 129) 

Although these allomorphs are not related through a general rule of the language, one 

might still assume that they correspond to a single underlying form, /-da/, and that /d/ 

surfaces as [l] after a vowel due to a lenition rule specific to the locative and allative 

suffixes. This example is therefore not necessarily suppletive, though the rule relating the 

allomorphs would apply only to the locative and allative. 

Another example of C/V-conditioned allomorphy is found in Biri (Terrill 1998), 

where once again the ergative/instrumental and locative suffixes exhibit allomorphy 

determined by the final segment of the stem. Vowel-final stems select the ergative/ 

locative form -ŋgu, while consonant-final stems select -du, as seen below (page numbers 

from Terrill 1998 are given next to each example). 

(40) gunhami-ŋgu ‘that-ERG’ (15) bunbun-du ‘pheasant-ERG’ (36) 

gayu-ŋgu ‘woman-ERG’ (15) dhagany-dyu ‘crocodile-ERG’ (35) 

mala-ŋgu ‘hand-INST’ (16) waŋgarany-dyu ‘all-ERG’ (50) 

balgu-ŋgu ‘axe-INST’ (16) 

The locative suffix in Biri shows the same pattern except that it has a where the 

ergative/instrumental has u, as shown below (Terrill 1998). 

(41) dhula-ŋga ‘tree-LOC’ (19) bidhal-da ‘Woodhouse Station-LOC’ (54) 

yamba-ŋga ‘camp-LOC’ (19) waŋal-da ‘boomerang-LOC’ (20) 

buri-ŋga ‘fire-LOC’ (19) 

ŋugunda-ŋga ‘night-LOC’ (19) 

In Nyangumarta (Sharp 2004: 117), ergative is marked by -ju with consonant- 

final stems and by -lu with vowel-final stems. 

184

(42) partany-ju ‘child-ERG’ ngagu-lu ‘1sg-ERG’ 

wangal-ju ‘wind-ERG’ mirtawa-lu ‘woman-ERG’ 

Minyjun-ju ‘(name)-ERG’ pirirri-lu ‘man-ERG’ 

parirr-ju ‘hand-ERG’ Yinyjana-lu ‘(name)-ERG’ 

Locative is marked in Nyangumarta by -jV with consonant-final stems and -ngV 

with vowel-final stems, where the vowel of the suffix assimilates to the final vowel of the 

stem (Sharp 2004: 118). 

(43) parirr-ji ‘hand-LOC’ 

tili-ngi ‘flame-LOC’ mirtawa-nga ‘woman-LOC’ 

karru-ngu ‘creek-LOC’ 

In Kuuku Yaʔu (Thompson 1976), ergative and instrumental are marked by -lu 

with consonant-final stems, and by -Vlu or -ʔV with vowel-final stems, where V is a copy 

of the stem-final vowel (the two suffix variants for vowel-final stems vary freely) (1976: 

329). Examples are shown below (Thompson 1976: 329-331). 

(44) John-lu ‘John-ERG’ 

piipi-ilu ~ piipi-ʔi ‘father-ERG’ mukana-ʔa ‘big-ERG’ 

yuku-ulu ~ yuku-ʔu ‘tree-ERG’ 

In Ngiyambaa (Donaldson 1980), allomorphy is conditioned not only by whether 

the stem ends in a consonant or a vowel, but also by more specific features of the stem- 

final segment. Ngiyambaa has three primary ergative/instrumental allomorphs, -gu, -u, 

and -DHu (DH is a laminal stop archiphoneme), though Donaldson (1980: 83) reduces 

these to two underlying forms, /-gu/ and /-DHu/. The -gu allomorph occurs with vowel- 

final stems (including stems ending in VN). The -u allomorph occurs with stems ending 

in l or r. The -DHu allomorph occurs with other consonant-final stems. Examples are 

shown below (Donaldson 1980: 82). 

185

(45) mura-gu ‘spear-ERG’ miri-gu ‘dog-ERG’ 

dhuruŋ-gu ‘snake-ERG’ dhuli:ŋ-gu ‘sand goanna-ERG’ 

gaɽul-u ‘stone-ERG’ mugar-u ‘prickle-ERG’ 

gamugin-du ‘mosquito-ERG’ 

bura:-dhu ‘child-ERG’ ŋurunh-dhu ‘emu-ERG’ 

( < bura:y-) ( < ŋuruyN-) 

Donaldson formulates a morphophonological rule (1980: 50) deleting dh of a case marker 

after l or r, which allows the -u allomorph to be derived from /-DHu/. The motivation for 

making this a general rule of the language is that two other cases, locative and 

circumstantive, exhibit similar patterns. Locative forms are identical to ergative forms 

except that they have a instead of u, as seen below (Donaldson 1980: 82). 

(46) a. mura-ga ‘spear-LOC’ miri-ga ‘dog-LOC’ 

dhuruŋ-ga ‘snake-LOC’ dhuli:ŋ-ga ‘sand goanna-LOC’ 

b. gaɽul-a ‘stone-ERG’ mugar-a ‘prickle-LOC’ 

c. gamugin-da ‘mosquito-LOC’ 

bura:-dha ‘child-LOC’ ŋurunh-dha ‘emu-LOC’ 

( < bura:y-) ( < NuruyN-) 

An interesting point to be made regarding this example is that in the forms bura:-dhu and 

bura:-dha in (45)c and (46)c, respectively, the stem-final /y/ that conditions the use of the 

-DHu suffix is absent in the surface form. Thus, this is an example in which a condition 

on the distribution of allomorphs is rendered opaque by a phonological process. 

Muruwari (Oates 1988) has three basic ergative/instrumental allomorphs (Oates 

1988: 56). The -ngku allomorph occurs with vowel/‘semivowel’-final stems. The -u 

allomorph occurs with stems ending in l or r. Finally, -tu/-thu/-tju occurs with stems 

ending in other consonants (the place of articulation of the coronal is determined by the 

stem-final consonant). Examples are given below (Oates 1988: 56, 58). 21 

21 Bidjara (Breen 1976a) exhibits a pattern of allomorphy that is similar to Muruwari in 

the locative and in the ‘operative’, which marks agents and instruments and corresponds 

in shape to PN ergative. According to Breen (1976a: 339), the operative is marked by -ŋu 

186

(47) kuthara-ngku ‘child-ERG’ kamay(i)-ngku ‘yam-INST’ 

kula-ngku ‘kangaroo-ERG’ kuliya-ngku ‘spear-INST’ 

kuntarl-u ‘dog-ERG’ kurlur-u ‘widow-ERG’ 

muwarn-tu ‘younger brother-ERG’ kaan-tu ‘snake-ERG’ 

wilanh-thu ‘sp. of cloud-ERG’ mayinj-tju ‘man-ERG’ 

Locative exhibits the same pattern, except that the vowel of the suffix is a rather than u. 

Examples of the locative are shown below (Oates 1988: 59-63). 

(48) thuri-ngka ‘sun-LOC’ nhuu-ngka ‘this-LOC’ 

partala-ngka ‘tomorrow-LOC’ paru-ngka ‘down-LOC’ 

kuntarl-a ‘dog-LOC’ thinkal-a ‘knee-LOC’ 

nhurran-ta ‘that-LOC’ karan-ta ‘across-LOC’ 

kirin-tja ‘husband-LOC’ yurrin-tja ‘night-LOC’ 

A slightly more complicated pattern is found in DuuNidjawu (Kite and Wurm 

2004), where the ergative/instrumental is marked by one of -ndu, -ru, or -yu when the 

stem has a final short vowel ((49) a), by -wu when the stem has a final long vowel 

((49)b), by -Du when the stem ends in a nasal ((49)c), and by -u when the stem has any 

other consonant as its final segment ((49)d). Examples below are from Kite and Wurm 

2004: 28-29 (ergative examples) and 31 (instrumental examples). 

(49) a. badja-ru ‘other.one-ERG’ bala-ru ‘jewfish-ERG’ 

bebere-yu ‘uncle-ERG’ djuŋa-ndu ‘vine-INSTR’ 

b. mama:-wu ‘uncle-ERG’ 

c. baran-du ‘boomerang-ERG’ ŋuwim-bu ‘sun-ERG’ 

d. guraŋgur-u ‘spear-INSTR’ gurruy-u ‘rain-INSTR’ 

The locative allomorphs in DuuNidjawu are identical to the ergative except that 

the locative suffixes have a where ergative has u (Kite and Wurm 2004: 32). Some 

examples are given below (Kite and Wurm 2004: 32). 

with vowel-final stems, -d̪u with final ny (which becomes n5), and -u with other 

consonant-final stems. The locative is marked by -ŋa with vowel-final stems, -d̪a with 

final ny (which becomes n̪), and -a with other consonant-final stems. No examples are 

provided. 

187

(50) doyi-ya ‘rock-LOC’ 

djuŋa-nda ‘vine-LOC’ 

bayer-djin-da ‘mountain-PL-LOC’ 

guyum-ba ‘camp-LOC’ 

The ergative in Dja:bugay (Hale 1976a) has five allomorphs distributed according 

to the stem-final segment (1976: 321). The form -ŋgu is used with vowel-final stems, -du 

is used with n-final stems, -ndu is used with l-final stems, -u is used with r-final stems, 

and -njdju is used with y-final stems (which then lose their final y). Stems with surface 

final m pattern with vowel-final stems. Examples are shown below (Hale 1976a: 321). 

(51) gura:-ŋgu ‘dog-ERG’ gurina-ŋgu ‘porcupine-ERG’ 

guyuru-ŋgu ‘wind-ERG’ 

bigumu-ŋgu ‘fingernail-ERG’ (< bigum) 

djulbin-du ‘tree-ERG’ yaraman-du ‘horse-ERG’ 

njiwul-ndu ‘one-ERG’ dayal-ndu ‘boy-ERG’ 

dugir-u ‘live-ERG’ 

bibu:-njdju ‘small-ERG’ (< bibu:y) 

djaru-njdju ‘bird-ERG’ (< djaruy) 

The instrumental and locative in Dja:bugay are marked by one of three 

allomorphs (Hale 1976a: 322). Bimoraic vowel-final stems have their final vowel 

lengthened. Vowel-final stems with three or more moras take -la. Consonant-final stems 

have instrumental/locative suffixes that are identical to the ergative except that the vowel 

is a: -da is used with n-final stems, -nda is used with l-final stems, -a is used with r-final 

stems, and -njdja is used with y-final stems (which then lose their final y). As with the 

ergative, m-final stems pattern with vowel-final stems. Examples are shown below (Hale 

1976a: 322). 

188

(52) a. bulmba-: ‘country-LOC’ bana-: ‘water-LOC’ 

b. djina:-la ‘foot-LOC’ djumburu-la ‘road-LOC’ 

c. djulbin-da ‘tree-LOC’ burŋan-da ‘ground-LOC’ 

d. baNgal-nda ‘big-LOC’ waNal-nda ‘boomerang-LOC’ 

e. dugir-a ‘live-LOC’ 

f. mula-njdja ‘hole-LOC’ (< mulay) 

gimu-njdja ‘Cairns-LOC’ (< gimuy) 

As in the Ngiyambaa example discussed earlier in this section, Dja:bugay exhibits some 

opaque conditioning in words with a /y/-final stem. As seen in the examples seen above 

in ((51)e) and ((52)f), stems with final /y/ take the ergative and locative suffixes -njdju 

and -njdja, respectively, but the stem-final /y/ is lost in the surface form. In a surface- 

based account of allomorphy (such as in a P >> M account), we would predict that these 

stems should pattern with vowel-final stems in terms of ergative and locative suffix 

allomorph selection, but as seen in these examples, this is not the case. 

In Wangkumara (Breen 1976b), allomorphy is conditioned by both phonological 

and morphological factors. In this language, the ergative (or ‘operative’) has four primary 

allomorphs (1976b: 336-337) distributed based on both morphological and phonological 

conditions (syllable count as well as the stem-final segment). The suffix -ulu is used with 

all masculine singular nouns, -anrru is used with non-masculine-singular nouns having 

two syllables or final i or u, -nrru is used with non-masculine-singular nouns of more 

than two syllables ending in a, and -ŋu is used with dual nouns. Examples are shown 

below (Breen 1976b: 337). 

(53) t̪it̪i-ulu ‘dog-ERG’ kat̪ikat̪i-ulu ‘snake-ERG’ 

kugad̪ari-anrru ‘wind-ERG’ muku-anrru ‘bone-ERG’ 

makurra-nrru ‘tree-ERG’ ŋamadya-nrru ‘mother-ERG’ 

t̪it̪i-ula-ŋu ‘dog-DUAL-ERG’ 

Locative is marked in Wangkumara by two allomorphs whose distribution is 

morphologically conditioned (Breen 1976b: 337). The suffix -ŋala is used with dual 

189

nouns, and -laŋa is used with all other nouns. Examples are given below (Breen 1976b: 

337-338). 

(54) mirrulu-ŋala ‘Naryilco-LOC’ naka-laŋa ‘water-LOC’ 

kukad̪iula-ŋala ‘btwn. the two hills-LOC’ murrkadya-laŋa 

‘younger brother- LOC’ 

In some cases, allomorphy is semantically conditioned. For example, in Kuku 

Yalanji (Patz 2002), the distribution of ergative allomorphs is both semantically and 

phonologically conditioned. The suffixes -nkV and -VnkV are used with ‘potent’ stems 

that are vowel- or consonant-final, respectively ((55)a-b) (Patz 2002: 46). With ‘neutral’ 

stems, -bu is used after vowel-final stems ((55)c), -Vbu after r-final stems ((55)d), -njV 

after y-final stems ((55)e), and -dV elsewhere ((55)f). 

(55) a. ngurrku-ngku ‘mopoke-ERG’ 

kaya-ngka ‘dog-ERG’ 

kami-ngka ‘father’s father / mother’s mother-ERG’ 

b. ngangkin-angka ‘porcupine-ERG’ 

dingkar-angka ‘man-ERG’ 

walkarr-angka ‘black goanna-ERG’ 

dubuy-ungku ‘small brown kingfisher (messenger bird)-ERG’ 

wabul-ungku ‘Torres Strait Pigeon-ERG’ 

c. jna-bu ‘foot-ERG’ 

kulji-bu ‘stone-ERG’ 

juku-bu ‘tree-ERG’ 

d. wungar-abu ‘sun-ERG’ 

badur-ubu ‘fishing line-ERG’ 

e. balbay-nja ‘light, lightning-ERG’ 

diliy-nja ‘corkwood pine-ERG’ 

junjuy-nja ‘something-ERG’ 

f. jalun-du ‘sea-ERG’ 

jalbany-da ‘taboo food-ERG’ 

bayil-da ‘freshwater perch-ERG’ 

balarr-da ‘scabies-ERG’ 

The locative in Kuku Yalanji also exhibits both phonologically and semantically 

conditioned allomorphy (Patz 2002: 48). Locative is marked by -ndV and -VndV with 

190

potent stems that are vowel- or consonant-final, respectively ((56)a-b). Among neutral 

stems, vowel-final stems take -ngV ((56)c), r-final stems take -V ((56)d), y-final stems 

take -mbV ((56)e), and all other stems take -bV ((56)f). 

(56) a. ngiwa-nda ‘salt-water eel-ERG’ 

bulki-nda ‘cattle-ERG’ 

dunyu-ndu ‘husband-ERG’ 

b. mukay-anda ‘mother’s older sister,/ younger sister’s child-ERG’ 

mukirr-anda ‘freshwater oyster-ERG’ 

wuyngkul-undu ‘spirit of dying person-ERG’ 

kukur-undu ‘rat-ERG’ 

c. ngara-nga ‘roots-ERG’ 

ngarri-nga ‘leg-ERG’ 

kiju-ngu ‘mud crab-ERG’ 

d. bibar-a ‘shin-ERG’ 

burrir-a ‘island-ERG’ 

bujur-u ‘feather-ERG’ 

e. buray-mba ‘spring (water)-ERG’ 

dajaliy-mba ‘deep water-ERG’ 

wakuy-mbu ‘upper arm-ERG’ 

f. nyabil-ba ‘tongue-ERG’ 

dalkan-ba ‘casuarina pine-ERG’ 

diburr-bu ‘egg’ 

The locative suffix allomorphs in Kuuku Yaʔu (Thompson 1976) are also 

semantically distributed. The allomorph -lu (which can reduce to -l) is used with a 

‘demonstrative reference’, while -ŋuna ~ -Vla are used elsewhere (-ŋuna also has surface 

variants -ŋun and -ŋu; -Vla can reduce to -Vl). Examples are given below (Thompson 

1976: 329-331). 

(57) dinghy-lu ‘dinghy-LOC’ airstrip-ŋun ‘airstrip-LOC’ 

kat̪aa-lu ‘dead-LOC’ kulamu-ŋun ‘road-LOC’ 

There also exist languages in which the ergative and locative do not exhibit any 

allomorphy. For example, in Warluwara (Breen 1976c), ergative is always marked by the 

suffix -gu, and locative is always marked by -ga (1976c: 331-332). Examples are given 

below (1976c: 332-333). 

191

(58) bumad̪a-gu ‘sun-ERG’ wada-ga ‘stone-LOC’ 

yaŋad̪a-ra-gu ‘see-SUBJ-ERG’ wuŋaramba-ga ‘Carandotta-LOC’ 

Similarly, in Gooniyandi (McGregor 1990), the ergative is marked invariably by 

the suffix -ngga (McGregor 1990: 177-178). This suffix can have both an ergative and an 

instrumental function, though McGregor explicitly rejects splitting it into two separate 

suffixes. Examples are shown below with page numbers from McGregor 1990. 

(59) yoowarni-ngga ‘one-ERG’ (178) gambayi-ngga ‘boy-ERG’ (506) 

ngirndaji-ngga ‘this-ERG’ (588) niyi-ngga ‘that-ERG’ (586) 

The tables below summarize the cases described above. Table 1 shows languages 

exhibiting ergative allomorphy, and Table 2 shows languages exhibiting locative 

allomorphy (as seen above, some languages have both and are therefore listed in both 

tables). 

192

Table 1: Examples of ergative allomorphy 

Language Form Environment Form Environment 

Bidjara -d̪u ny-final stem -u other C-final stem 

-ŋu V-final stem 

Biri -ngu V-final stem 

-du C-final stem 

Diyari -ndu female personal name -li male personal name, 

a-final noun 

-yali other common 

noun 

Dja:bugay -ŋgu V-final stem -u r-final stem 

-du n-final stem 

-ndu l-final stem 

-njdju y-final stem 

Dyirbal -ŋgu disyllabic stem 

-gu trisyllabic+ stem 

Duuŋidjawu -ndu/-ru/-yu stem w/final short V -wu stem w/final long V 

-Du nasal-final stem -u stem w/other final C 

Gooniyandi -ŋgga all environments 

Kaititj -ŋ disyllabic stem -l trisyllabic+ stem 

Kuku Yalanji -ŋkV potent referent 

-dV neutral referent 

Kuuku Yaʔu -Vlu/-/V V-final stem -lu C-final stem 

Muruwari -ngku V-final stem -u l- or r-final stem 

-tu stem w/other final C 

Ngiyambaa -DHu V-final stem -u l- or r-final stem 

-gu stem w/other final C 

Nhanda -(ng)gu bimoraic stem -lu stem with 3+ moras 

Nyangumarta -ju C-final stem -lu V-final stem 

Wangkumara -ŋu dual stem -ulu masc. singular stem 

-anrru non-masc.-sg. w/2 

syllables or final 

i or u 

-nrru non-masc.-sg. w/3+ 

syllables and final a 

Warlpiri -ŋku noun stem with -rlu stem with 3+ moras 

2 moras or non-noun stem 

Warluwara -gu all environments 

Warrgamay -ŋgu V-final stem -du C-final stem 

Yidiɲ -ŋgu V-final stem -du C-final stem 

Yingkarta -ngku V-final stem w/2 moras -lu V-final stem w/3+ 

-Su C-final stem moras 

193

Table 2: Examples of locative allomorphy 

Language Form Environment Form Environment 

Bidjara -d̪a ny-final stem -a other C-final stem 

-ŋa V-final stem 

Biri -nga V-final stem 

-da C-final stem 

Dja:bugay -: V-final stem w/2 moras -la V-final stem w/3+ 

-da n-final stem moras or r-final 

-nda l-final stem stem 

-njdja y-final stem 

Duuŋidjawu -nda/-ra/-ya stemw/final long V -wa stem w/final long V 

-Da nasal-final stem -a stemw/other final C 

Kaititj -ŋ disyllabic stem -l trisyllabic+ stem 

Kuku Yalanji -ndV potent referent 

-ŋV neutral referent 

Kuuku Yaʔu -ŋunaʔ-Vla non-demonstrative -lu demonstrative 

Martuthunira -ngka V-final stem with -la V-final stem with 

2 moras 3+ moras 

Muruwari -ngka V-final stem -a l- or r-final stem 

-ta stem w/other final C 

Ngiyambaa -DHa V-final stem -a l- or r-final stem 

-ga stem w/other final C 

Nhanda -(ng)gu all environments 

Nyangumarta -ngV V-final stem 

-jV C-final stem 

Wangkumara -ŋala dual stem -laŋa any other stem 

Warlpiri -ŋka noun stem with -rla stem with 3+ moras 

2 moras or non-noun stem 

Warluwara -ga all environments 

Yidiɲ -da C-final stem -la V-final stem 

Yindjibarndi -ŋka V-final stem with -la C-final stem or stem 

Figure 2: Map showing distribution of conditions on ergative allomorphy 

<strong>Conditions</strong> determining allomorph distribution 

syllable count semantic mora count lexical 

V- vs. C-final final segment feature no allomorphy other/multiple 

Map outline © 2006 Commonwealth of Australia (Geoscience Australia). 

Used with permission. 

As can be seen from the map above, some generalizations can be made regarding 

the geographical ranges of the different types of conditions on allomorph distribution. For 

example, in this sample, mora count-based allomorphy occurs primarily on Australia’s 

west coast, although one example is found in central Australia. On the other hand, 

allomorphy based on features of the stem-final segment seems to be limited to eastern 

Australia. Semantically conditioned allomorphy occurs in both the south central region 

and in the northeast. Syllable count-based allomorphy occurs in the central and eastern 

195

egions. Finally, allomorphy determined by whether the stem is consonant- or vowel-final 

appears to be widely distributed, occurring in the eastern, central, and western regions. 

In the following section, I discuss some previous approaches to PN ergative 

allomorphy, then propose an account for the development of the phenomenon based on 

the earlier studies and consideration of the distribution of allomorphy patterns in modern 

PN languages. 

4.2.2 Previous discussions 

The historical development of Pama-Nyungan ergative allomorphy has been 

discussed in at least three previous studies. The first of these is Hale (1976b). Hale’s 

proposal for the development of ergative allomorphy can be summarized as follows. 

First, Hale narrows the discussion to the alternant set in (60)a and the ‘extremely 

common’ ergative/locative allomorph distribution pattern shown in (60)b (Hale 1976b: 

414; S = a stop homorganic to the preceding consonant). 

(60) a. ergative locative 

*-lu *-la 

*-ŋku *-ŋka 

*-mpu *-mpa 

*-njtju *-njtja 

b. (i) *-ŋku, *-ŋka attach to vowel-final disyllabic stems; 

(ii) *-lu, *-la attach to vowel-final polysyllabic stems; 

(iii) *-Su, *-Sa attach to consonant-final stems 

Hale assumes (1976b: 414) that the basic forms of the ergative and locative were *-lu and 

*-la, respectively. To collapse categories (ii) and (iii) in (1) above, Hale proposes an 

assimilation rule changing l into a stop with place features determined by an immediately 

preceding consonant across a morpheme boundary (1976b: 415). Hale’s analysis centers 

196

around the observation (1976b: 415) that there is a range of allowable stem-final 

consonants in Australian languages, such that some languages allow no final consonants, 

some allow apicals only, some allow coronals, and some allow both coronals and non- 

coronals. In a language that allowed stem-final coronals and non-coronals and had the l- 

assimilation rule described above, the ergative forms corresponding to a CVCVm and 

CVCVŋ stem would have been CVCVmpu and CVCVŋku, respectively. Now suppose 

that this language lost all of its non-coronal stem-final consonants at some later stage in 

its history. This would mean that both stems had the form CVCV, but their ergative 

forms would still have had the forms CVCVmpu and CVCVŋku (this leaves open the 

question of whether the same change applied to other suffixes in the language; the 

parallel patterning of locative allomorphy with ergative allomorphy in many PN 

languages suggests that the locative underwent the same changes as the ergative, though 

there are other possible sources for this similarity (e.g., the initial consonant of the 

ergative and locative could be a stem former rather than being part of the case suffixes)). 

Some reanalysis could have taken place such that the ergative stem was the same as the 

plain stem (CVCV) and the ergative suffix was reanalyzed as -mpu ~ -ŋku. This would 

yield a system with four ergative allomorphs, where coronal-final stems take -Su, and V- 

final stems take lexically determined allomorphs: -lu ~ -mpu ~ -ŋku. Some semantic or 

phonological condition could then develop to account for the distribution of these 

allomorphs. Languages that also exhibit *-njtju can be accounted for under this scenario 

if the rule deleting non-coronal consonants is expanded to delete all distributed 

consonants, which would then include tj. 

197

To account for languages that mark disyllabic stems with *-ŋku and stems with 

three or more syllables (‘polysyllabics’) with *-lu, Hale points out (1976b: 416) that 

many languages, including some Australian languages (Uradhi, Wik Me’nh, and Arandic 

Anmatjera) have a rule that Hale terms ‘velar-posthesis’, which inserts ŋ after a word- 

final vowel. There is no obvious phonetic motivation for such a rule, but it does recur in 

many different languages (Hale 1976b). In Arandic Anmatjera, this rule is limited to 

disyllabic words, and Hale proposes that this form of the rule was present in the mother 

language to the languages exhibiting the *-ŋku ~ *-lu allomorphy in disyllables vs. 

polysyllables. Hale proposes further that the stems undergoing ŋ-insertion were 

reanalyzed as having underlying final ŋ by analogy with other ŋ-final stems since these 

two types of stems would have had the same shape in their citation forms due to the 

application of velar posthesis; this would have resulted in a disproportionately large 

number of ŋ-final disyllabic stems. Dixon (1980: 211) reports that this has happened 

‘over a wide area in the south-east,’ citing Muk-Thang of Gippsland, where ‘almost all 

disyllabic roots end in a consonant, with the most common final segment being /ŋ/; 

amongst monosyllables and trisyllabics there are more final vowels’. Since vowel-final 

polysyllables would not have undergone velar posthesis, they would still have had -lu as 

their ergative allomorph, whereas disyllabic stems would all take -Su due to the rule 

changing l to a stop homorganic to a previous consonant. If this language then underwent 

deletion of final consonants (a relatively common process that may result from the loss of 

acoustic cues to the consonant when no vowel follows), the ergative suffix could have 

been reanalyzed as -ŋku for all of the stems that previously had final ŋ (either 

etymologically or via velar posthesis). Most of these stems would have been disyllabic, 

198

since all of the ones that got final ŋ through velar posthesis would be disyllabic, plus 

many of the stems with etymological final ŋ would already have been disyllabic as well. 

If a very high percentage of ŋ-final stems were disyllabic (in contrast to stems ending in 

other segments, which would have had a lower percentage of disyllabic stems), this 

pattern could have been interpreted such that all disyllabic stems took -ŋku, while 

polysyllabic stems took -lu. In Table 3 below, I summarize Hale’s proposed sequence of 

changes leading to syllable counting allomorphy involving -ŋku vs. -lu. 

Table 3: Hale’s (1976b) proposed history of PN ergative/locative allomorphy 

Stage I: Ergative/locative is /-lu/; l → S / C + __ (where C is any consonant, + is a 

morpheme boundary, and S is a stop at the same place of articulation as 

C); velar posthesis inserts ŋ after every V-final disyllabic stem. 

Stage II: CVCV stems are reanalyzed as CVCVŋ, yielding a disproportionately 

high number of disyllabic stems ending in ŋ; ergative forms of these stems 

change from CVCV-lu to CVCVN-ku (by l-assimilation). 

Stage III: Final consonants are lost; former CVCVC stems are now CVCV, but their 

ergative forms are still CVCVCSu, reanalyzed as CVCV-CSu. Multiple 

primary ergative allomorphs now exist: -lu for stems that were always Vfinal 

(non-disyllabic stems only) and multiple different -SCu forms (-ŋgu, 

-mpu, -njtju) for stems formerly ending in consonants (the largest class of 

these likely being ŋ-final since this class would include stems with 

etymological final ŋ as well as final ŋ from velar posthesis). 

Stage IV: Since all stems taking -lu are non-disyllabic and most stems taking -ŋku 

are disyllabic, the ergative allomorphy is reanalyzed as follows: disyllabic 

stems take -ŋku; other stems take -lu. 

A second source that discusses the development of ergative/locative/instrumental 

allomorphy is Dixon (1980). 22 Dixon rejects Pama-Nyungan as a genetic group, but 

reconstructs Proto-Australian ergative allomorphs that could also be interpreted as Pama- 

22 See also Dixon 2002, which includes further discussion of some issues presented here 

but leaves out some data that were included in Dixon 1980. For that reason I cite Dixon 

1980 throughout the remainder of this section. 

199

Nyungan. In fact, as Sands (1996) points out, Dixon’s proposed Proto-Australian ergative 

marker is based on evidence from Pama-Nyungan. Dixon reconstructs the following 

allomorphs for Proto-Australian (S is a stop homorganic to the preceding stop; DH is a 

laminal stop). 

(61) -Su after nasals 

-DHu after y 

-du after l, rr 

-lu after vowels when root has 3+ syllables 

-ŋgu after vowels when root has 2 syllables 

Dixon accepts as plausible Hale’s (1976b) proposal that all of these allomorphs arose 

from *-lu via the series of changes described above in the discussion of Hale’s analysis. 

Dixon’s proposal is therefore basically the same as Hale’s proposal, except that Dixon 

claims that his reconstruction holds for the ancestor of all of the Australian languages 

rather than just the Pama-Nyungan languages. 

Sands (1996) argues for a different reconstruction of the ergative in Pama- 

Nyungan. Sands claims that Pama-Nyungan had three ergative allomorphs as follows 

(1996: 8). 

(62) *-lu / V_ nominals which are not common nouns 

*-ŋgu / V_ 2 syllables 

*-DHu / elsewhere 

Sands claims further that an ergative suffix *-DHu can be reconstructed for Proto- 

Australian, though the evidence for this claim is weaker than that for the proposed Pama- 

Nyungan ergative reconstruction. Hale’s (1976b) scenario for the development of *-ŋgu 

from *-lu is not disputed, but Sands argues (1996: 11) that *-ŋgu must already have been 

restricted to occuring with disyllabic stems in Pama-Nyungan since the syllable count- 

based distribution occurs across a wide geographical area in Sands’ survey. Although 

200

Sands includes more languages than the present survey, it should be pointed out that at 

least four of the ten western Australian languages that Sands cites (1996: 9) as having 

syllable count-based allomorphy (Martuthunira, Nhanda, Yindjibarndi, and Yingkarta, 

each discussed above) are more accurately described as having mora count-based 

allomorphy. Therefore, Sands’ proposal may need to be modified to state that PN had 

ergative allomorphy based on mora count rather than syllable count. The argument for the 

*-DHu allomorph is based both on the common occurrence of laminal-initial allomorphs 

across the Pama-Nyungan area and the fact that *-DHu is also plausible for some non- 

Pama-Nyungan languages. Note that the latter argument applies only if one accepts the 

claim that the Pama-Nyungan languages form a larger Australian group with the non- 

Pama-Nyungan languages. This is a controversial claim that has not been substantiated to 

the satisfaction of researchers in the field, with the notable exception of R.M.W. Dixon, 

who has argued for the existence of an Australian group and for the non-existence of 

Pama-Nyungan. Despite the question of whether it is desirable to reconstruct a Pama- 

Nyungan ergative suffix by taking the non-Pama-Nyungan languages into consideration, 

I nonetheless do not find evidence in the present survey to contradict *-DHu as opposed 

to an alveolar-initial suffix. 

4.2.3 A history of PN ergative allomorphy 

Each of the previous studies of the history of PN ergative allomorphy assumes 

that both *-lu and *-ŋgu were present in the common ancestral language of modern PN 

languages (though as discussed above, Dixon (1980) denies the existence of PN and 

instead reconstructs these allomorphs for Australian). Under Hale’s (1976b) and Dixon’s 

201

(1980) proposals, this allomorphy could have been derived by rule from a single 

underlying form, /-lu/. Under Sands’ (1996) account, on the other hand, the allomorphs 

were already in a suppletive relationship in Proto-PN. Sands’ justification for this 

assumption (1996: 7-8) is that there are several modern PN languages where -lu occurs 

with proper nouns (Sands 1996: 14-15): Yankunytjatjara, Badimaya, Yindjibarndi, 

Atynyamathanha, and Watjarri. However, a problem with this assumption is that 

according to Ethnologue all five of these languages are classified as South-West Pama- 

Nyungan languages (Gordon 2005). 23 Therefore, the use of -lu with proper nouns may be 

better analyzed as an innovation in South-West Pama-Nyungan than as a feature of Proto- 

PN. I will assume, following Hale (1976b) and Dixon (1980) that the Proto-PN ergative 

suffix had a single underlying form. 

As mentioned above, Sands (1996) makes a good case for reconstructing a 

laminal as the initial segment of the ergative suffix (*-DHu), though she does not analyze 

-lu as being a reflex of this suffix. I propose that the Proto-PN ergative suffix allomorphs 

were derived from one underlying form and two (synchronic) phonological rules, as 

follows: 

23 More specifically, according to Ethnologue (Gordon 2005), Yankunytjatjara is Wati 

language, Atynyamathanha is a Yura language, Yindjibarndi is a Coastal Ngayarda 

language, Badimaya and Watjarri are Watjarri langauges; all of these are grouped 

together in the Ethnologue as South-West Pama-Nyungan. But it should be 

acknowledged that according to Claire Bowern (p.c.), this classification is disputed 

among some researchers, publicly if not in print. 

202

(63) Underlying form: /-DHu/ 

Place Assimilation: C + C V 

erg 

= 

Place Place 

[Laminal] 

Lenition: V + C V 

erg 

= = 

Place 

[-cont] 

[Laminal] 

Though the Lenition rule may cause concern since it involves, in effect, two separate 

delinking processes (in addition to the later default insertion of [Alveolar] and [+cont], 

which I assume), this is nonetheless not an unnatural rule. It may be thought of as a 

general process of reduction, which happens to delete both the place and manner features 

of the consonant between two vowels. 

The proposed underlying form and phonological rules can account for Dixon’s 

(1980) allomorphs as follows: 

(64) Dixon (1980) allomorph Rule(s) 

-Su after nasals Place Assimilation 

-DHu after y Place Assimilation 

-du after l, rr Place Assimilation 

-lu after vowels when root has 3+ syllables Lenition; see below 

-ŋgu after vowels when root has 2 syllables Place assimilation; see below 

A few further comments are necessary to explain how this proposal accounts for the 

allomorphs identified by Dixon (1980). Although Dixon reconstructs the -lu allomorph as 

occurring after vowels when the root has 3+ syllables and -ŋgu after vowels when the 

root has two syllables, we can assume given Hale’s explanation of the origin of -ŋgu that 

Proto-PN had only -lu after vowels, and that -ŋgu arose later in some languages where 

203

final ŋ was lost, according to Hale’s (1976b) scenario. 24 Given what was said above 

regarding the wide distribution of -ŋgu vs. -lu for bimoraic vs. larger stems, I hypothesize 

that this was the basis for distribution in the first language/group to develop -ŋgu, and 

that this underwent later changes in some languages to distributions based on syllable 

count, consonant- vs. vowel-final stem, semantic distinctions, etc., as are found in 

modern PN languages. I assume, therefore, that Hale’s (1976b) proposed rule of velar 

posthesis actually applied to bimoraic stems, rather than to disyllabic stems. 

If this scenario is correct, then it need not be assumed that PN ergative 

allomorphy exists for the purpose of phonological optimization. Rather, as can be seen in 

the above discussion, the modern patterns of PN ergative allomorphy arose via a complex 

series of phonological changes, each of which was independently motivated, but not 

necessarily for reasons of optimization 25 (in fact, in the case of velar posthesis, it is 

difficult to imagine a phonetic or phonological motivation, 26 though Hale (1976b) and 

Dixon (1980) cite enough examples of the rule to justify the assumption that it is 

plausible and could have existed in Proto-PN). The purpose of this section has been to 

show that PN ergative allomorphy could have developed for other reasons than 

phonological optimization. Having done this, I now move to an analysis of the general 

phenomenon of prosodically conditioned suppletive allomorphy. 

24 This has the advantage of accounting for languages that never have -ŋgu postvocalically. 

Sands (1996: 13-14) lists 15 such languages. 

25 It could be argued that analogy is optimizing, since its overall effect can be to simplify 

the grammar. However, this would be a different kind of ‘optimization’ from the 

phonological optimization of words, which is the type of optimization that I am talking 

about here and the type that the P >> M model is meant to capture. 

26 Of course there must have been some motivation for the process, but it was probably 

something external to the phonological system itself (e.g., analogy, borrowing, or perhaps 

some acoustic or articulatory factor not yet identified). 

204

4.3 Analysis of prosodically conditioned suppletive allomorphy 

The apparent optimizing nature of Pama-Nyungan ergative allomorphy has been 

explained based on its historical origins in §4.2, and the majority of the remaining cases 

of syllable and mora count-based allomorphy are not phonologically optimizing. In this 

section, I contrast two approaches to the phenomenon as they apply to prosodically 

conditioned suppletive allomorphy. The first is the Output Optimization approach, which 

was applied specifically to cases of ‘syllable-counting allomorphy’ (SCA) by Kager 

(1996). The second approach, which I argue to be superior, is a subcategorization 

approach. As I will show, subcategorization is capable of handling both optimizing and 

non-optimizing allomorphy, and it does a much better job of handling the non-optimizing 

cases. 

4.3.1 The Output Optimization approach 

Kager (1996: 170) claims that ‘[s]yllable-counting allomorphy is an output- 

oriented phenomenon,’ resulting from the Emergence of the Unmarked (TETU; 

McCarthy and Prince 1994). In this approach, the morphology supplies multiple 

suppletive allomorphs of an affix in candidates in an OT tableau, and phonological foot 

structure and parsing constraints select the appropriate output. This approach falls under 

McCarthy and Prince’s (1993a,b) ‘P >> M’ ranking schema, where phonological effects 

in morphology are modeled by ranking phonological (P) over morphological (M) 

constraints. In phonologically conditioned suppletive allomorphy (PCSA), the relevant M 

constraint is one enforcing uniform marking of morphological categories. One prediction 

205

of this approach is that syllable-counting allomorphy should serve to optimize words with 

respect to phonological constraints that are independently motivated elsewhere in 

Universal Grammar. 

Kager’s argument is based on Estonian, where the Genitive plural, Partitive 

singular, and Partitive plural exhibit SCA, as seen below (examples below with page 

numbers in parentheses are from Mürk 1997, and the remaining examples are from Kager 

1996: 157, 168; this is reproduced from (4) above). 

(65) Genitive plural Partitive plural Partitive singular Gloss 

maa:-te (74) ma-it (74) maa:-t ‘land’ 

aas:ta-tte (149) aas:ta-it (149) aas:ta-tt ‘year’ 

häda-te (77) häda-sit (77) häda-Ø (77) ‘trouble’ 

paraja-tte paraja-it paraja-tt ‘suitable’ 

raamattu-tte raamaattu-it raamattu-tt ‘book’ 

atmirali-te atmirali-sit atmirali-Ø ‘admiral’ 

telefoni-te telefoni-sit telefoni-Ø ‘telephone’ 

Kager claims that these apparent syllable counting effects in Estonian result from 

foot parsing considerations rather than from true ‘counting’. Kager proposes (1996: 162- 

163) the following constraints for Estonian: 

(66) FT-BIN: Feet are binary under syllabic or moraic analysis. 

PARSE-2: One of two adjacent stress units (μ, σ) must be parsed by a foot. 

ALIGN-HD-L: Align (PrWd, L, Head(PrWd), L) 

ONSET: Syllables must have onsets. 

ALIGN-ST-R: Align (Stem, R, Foot, R) (Each right stem edge coincides with a 

right foot edge) 

PK-PROM: Peak(x) > Peak(y) if |x| > |y|. 

The correct genitive plural allomorphs for even-syllable (67)a and odd-syllable (67)b 

stems are selected via the following constraint rankings (Kager 1996: 164-167). 

(67) Genitive plural 

a. /visa, {-te, -tte}/ FT-BIN PARSE-2 ALIGN-HD-L ONSET ALIGN-ST-R PK-PROM 

[(ví.sa)-te] L 

[(ví.sa-t).te] *! L 

[vi.(sá-t.te)] *! * H 

[(ví).(sà-t.te)] *! * L, H 

206

. /paraja, {-te, -tte}/ FT-BIN PARSE-2 ALIGN-HD-L ONSET ALIGN-ST-R PK-PROM 

[(pá.ra).(jà-t.te)] * L, H 

[(pá.ra).(jà-te)] * L, L! 

[pa.(rá.ja)-te] *! L 

[(pá.ra).ja-te)] *! * L 

[(pá.ra).(jà)-te] *! L, L 

[(pá.ra.ja)-te)] *! L 

Similarly, the partitive plural allomorphs for even-syllable (68)a and odd-syllable (68)b 

stems are selected as shown below (Kager 1996: 164-167). 

(68) Partitive plural 

a. /visa, {-sit, -it}/ FT-BIN PARSE-2 ALIGN-HD-L ONSET ALIGN-ST-R PK-PROM 

[(ví.sa)-sit] L 

[(ví.sa-i)t] *! L 

[(ví.sa).-it] *! L 

[vi.(sá-i)t] *! * H 

[(ví).(sà-i)t] *! * L, H 

b. /paraja, {-sit, -it}/ FT-BIN PARSE-2 ALIGN-HD-L ONSET ALIGN-ST-R PK-PROM 

[(pá.ra).(jà-i)t] * L, H 

[(pá.ra).(jà-si)t] * L, L! 

[(pá.ra).(jà-.i)t] *! * L, L 

[pa.(rá.ja)-sit] *! L 

[(pá.ra).ja-sit)] *! * L 

[(pá.ra).ja-it)] *! * L 

[(pá).(rà.ja)-sit] *! L, L 

[(pá.ra).(jà)-sit] *! L, L 

This type of analysis works for Estonian, but we are left with the question of whether 

SCA is always reducible to general markedness and faithfulness, as Kager claims. In the 

following section, I address this question with reference to the other cases of syllable 

count-conditioned suppletive allomorphy described in §4.1 and §4.2. 

4.3.2 Survey results revisited 

As described above, this survey uncovered many cases of SCA in which the 

distribution of allomorphs is not predictable from their shape and must be stipulated; this 

in itself is an important result since it contradicts Kager’s claim that SCA is optimizing. 

207

In some cases of SCA, the distribution of allomorphs does relate to their shape. However, 

in some of these cases, the phonological constraints that would be needed to model the 

distribution are very specific constraints that cannot be assumed to be part of UG, and 

therefore these cases still cannot be analyzed as TETU effects. 

For example, recall that in Kimatuumbi (Odden 1996), monosyllabic verbs mark 

perfective with -i̹te, while polysyllabic stems take an -i̹- infix (Odden 1996: 51-53). The 

distribution could be construed as responding to a requirement that perfective forms have 

four syllables, though this is not surface-true due to some systematic exceptions. But note 

that if this were captured using a phonological constraint, it would be language-specific, 

not part of UG. Therefore, Kimatuumbi perfective allomorphy is not a TETU effect. 

4.3.3 Examples of non-optimizing SCA 

As mentioned above, this survey revealed several examples of non-optimizing 

SCA. An example is found in Tzeltal (Walsh Dickey 1999), described above in §4.1, 

where the perfective is marked by -oh with monosyllabic stems, and with -ɛh elsewhere. 

[o] and [ɛ] do not alternate elsewhere (Kaufman 1971: 28), so the allomorphy is probably 

truly suppletive. Stress in Tzeltal is word-final (Walsh Dickey 1999: 327), so the 

allomorphy is not stress-conditioned. A constraint banning [ɛ] in the second syllable has 

not been proposed for UG, so this is not a TETU effect. Such a constraint in Tzeltal 

would be stipulative and used only for this phenomenon, so this appears to be a case 

where we would not want to describe the distribution of allomorphs as phonologically 

optimizing in any way. 

208

Another such case is seen in Kaititj (Koch 1980), discussed above in §4.2. In this 

language, the -ŋ form of the ergative/instrumental/locative suffix occurs with disyllabic 

stems, while -l occurs with larger stems (examples are from Koch 1980: 264-266). Note 

that in these examples, /l/ → [ŋ] when the preceding consonant is apical (Koch 1980: 

274), and N is a prestopped apical nasal. 

(8) aˈki-ŋ ‘head-ERG’ aˈliki-l ‘dog-ERG’ 

ilˈt y i-ŋ ‘hand-ERG’ aˈʈuyi-l ‘man-ERG’ 

aNˈmi-ŋ ‘red ochre-ERG’ aˈɣirki-l ‘sun-ERG’ 

aynˈpi-ŋ ‘pouch-ERG’ ˈɭuNpiri-ɭ ‘forehead-ERG’ 

There does not seem to be anything ‘better’ about /ŋ/ rather than /l/ as a stressed syllable 

or second syllable coda, so this seems once again to be a case of non-optimizing SCA. 

A third example is found in Zuni (Newman 1965). As discussed above, in Zuni, 

singular nouns are marked with -ʔleʔ for monosyllabic stems, and with -nne for 

polysyllabic stems. Again, there is nothing to suggest that the distribution of allomorphs 

is phonologically optimizing. 

Finally, a fourth example of non-optimizing SCA is seen in Dyirbal. As discussed 

in §4.2, in vowel-final stems in Dyirbal, disyllabic stems mark ergative with -ŋgu, while 

longer stems take -gu. McCarthy and Prince (1990) characterize this as a ‘compensatory 

relationship’ in that the ‘shorter’ suffix goes with the ‘longer’ stem and vice versa. 

McCarthy and Prince acknowledge that coda consonants are not generally considered to 

be moraic in Dyirbal, but one would have to assume that codas are moraic, at least in this 

context, to give substance to the claim that the allomorphy is motivated by compensation. 

However, if the notion of a closed syllable does have some status in the language, then if 

209

anything, we would expect the exact opposite distribution of allomorphs from what 

actually occurs in Dyirbal. The reason is as follows: 

The stress pattern of Dyirbal is initial and alternating, and final syllables are 

unstressed (Dixon 1972: 274-276). This means that when we compare disyllabic stems 

with trisyllabic stems, the distribution of allomorphs causes words with -ŋgu to have 

closed unstressed syllables, while words with the -gu suffix have open stressed syllables. 

As seen in (69)a with a two-syllable stem, the ŋ of -ŋgu causes the penultimate syllable, 

which is unstressed, to be closed. And as seen in (69)b, the use of -gu rather than -ŋgu 

with the three-syllable stem results in the stressed penultimate syllable being left open. 

(69) a. ˈya.raŋ.gu ‘man-ERG’ *ˈya.ra.gu 

b. ˈya.ma.ˈni.gu ‘rainbow-ERG’ *ˈya.ma.ˈniŋ.gu 

If ‘syllable-counting allomorphy’ results from TETU as claimed by Kager (1996), then 

we would expect the distribution of allomorphs in Dyirbal to be reversed, since the 

Stress-to-Weight Principle should enforce a correspondence between heavy syllables (to 

the extent that closed syllables can be called ‘heavy’ in this language) and stress. 

The observation that coda consonants are not moraic in Dyirbal may in fact render 

the Stress-to-Weight argument moot, and a further complication is that in at least some 

Pama-Nyungan languages, NC sequences can be syllabified as onsets and may even be 

single segments (i.e., prenasalized stops). However, if the Stress-to-Weight argument is 

rejected on these grounds, one must also reject McCarthy and Prince’s original comment 

on this example. 27 Without the notion of ‘compensation’, there is no clear way in which 

27 It should be noted that in later work, McCarthy and Prince (1993b: 117-120) abandon 

reference to a ‘compensatory relationship’ between the root and affix in Dyirbal ergative 

210

the distribution of ergative allomorphs is optimal in Dyirbal, and therefore this example is 

neutral at best. Even if the example is only neutral and not truly ‘perverse’, it is 

problematic for the P >> M approach because the relevant P constraint in such an 

analysis would have to be language-specific, arbitrary, and not externally motivated. If an 

advantage of P >> M is supposed to be the reduction of PCSA to phonological principles 

that are independently identified in the regular phonologies of the world’s languages, 

then proposing arbitrary, item-specific P constraints for individual languages completely 

undermines the spirit of the model. 

4.3.4 Examples of SCA possibly resulting from TETU 

The survey did reveal some cases of SCA aside from Estonian that may be analyzed 

as TETU effects. For example, as described above in §4.1, the repetitive in Shipibo 

(Elías-Ulloa 2004) is marked by -ribi with even-syllabled stems, and by -riba elsewhere. 

Elías-Ulloa (2004) claims that this allomorphy is driven by sonority, so that /a/ is a good 

foot head, while /i/ is not (foot-initial syllables are assumed to be heads). The main stress 

is on the second syllable if closed; otherwise, the first syllable. There is no secondary 

stress. Elías-Ulloa uses the constraints *i/Head and *a/NonHead to select the correct 

allomorphs in a TETU analysis. 

allomorphy. Instead, they propose a P >> M account of this pattern in which the relevant 

P constraint is AFX-TO-FT, which requires the -ŋgu suffix to attach to a base that consists 

of a foot. In this analysis, failure to satisfy AFX-TO-FT results in a null parse, in which 

case the -gu allomorph is used by ‘default’. Wolf and McCarthy (to appear: §6.1) discuss 

this aspect of the analysis further, proposing that there is a ‘stipulated priority 

relationship’ between the two suffix allomorphs such that the -ŋgu allomorph is ‘tried 

first’, and if this yields a null parse, then -gu is used since -gu is ‘not indexed to AFX-TO- 

FT’. 

211

Note, however, that we do not need to assume that the allomorphy is suppletive. 

We could say simply that the repetitive suffix is /-ribV/ with an unspecified vowel, and 

then *i/Head and *a/NonHead could determine the vowel quality. This would not then be 

an instance of the P >> M ranking schema, though it could still be considered a TETU 

effect. 

Another case of SCA that can be analyzed as a TETU effect is found in Turkana 

(Dimmendaal 2000). Recall from §4.1 that the suffix used to form abstract nouns occurs 

as -ɪ̀sɪ́ with CVC roots and -ù with CiVjCiVjC roots. Since each noun has an á- feminine 

gender prefix, the resulting forms all have four syllables; this could be driven by foot 

structure/parsing constraints and could therefore result from TETU. 

4.3.5 The Subcategorization approach 

The existence of non-optimizing SCA demonstrates the need for a mechanism 

other than output optimization to handle SCA. A subcategorization approach captures the 

fact that there are cases of SCA where phonological well-formedness considerations have 

no bearing on the choice among allomorphs in a given environment. 

An example of an SCA pattern that is particularly well-suited to the 

subcategorization approach is found in Nakanai (Austronesian, New Britain; Johnston 

1980). In Nakanai, the -il- form of the nominalizing affix occurs when it can be in the 

first syllable and adjacent to main stress; -la occurs elsewhere (examples are from 

Johnston 1980: 177-178). 

212

(70) au ‘steer’ vi-gile-muli ‘tell a story’ 

il-au ‘steering’ vigilemulimuli-la ‘story’ 

peho ‘die’ vi-kue ‘fight (v.)’ 

p-il-eho ‘death’ vikue-la ‘fight (n.)’ 

loso ‘dive’ go-ilo ‘go in’ 

il-oso ‘diving’ goilo-la ‘entrance’ 

Stress is on the penult (Johnston 1980: 256). McCarthy (2003: 101-102) claimed that in 

Nakanai, -il- is attracted to the main stress of the word but is also a ‘formal prefix’, and 

the pattern is driven by dispreference for a stress shift with respect to the base. In 

McCarthy’s analysis, OO-PK-MAX penalizes the stress shift caused by -la, while the 

constraints AFX-TO-HD(-il-) and PREFIX/σ(-il-) limit -il- to occurring with disyllabic and 

smaller stems. Each of these constraints presumably outranks UNIFORMEXPONENCE, 

which requires each morphological category to be marked by a single affix across all 

forms (though McCarthy does not explicitly propose such a constraint), making this a 

case of P >> M. Since OO-PK-MAX is not active elsewhere in the language (McCarthy 

2003: 102), this is a case of ‘the emergence of the faithful’, not TETU. Thus, even a case 

of suppletive allomorphy that is apparently phonologically optimizing still contradicts 

Kager’s (1996) claim that SCA results from TETU. 

A subcategorization account for the Nakanai allomorphy is characterized as 

follows. The -il- affix subcategorizes for the first vowel and main stress (under a cyclic 

account of stress). -il- can then attach only to disyllabic and smaller stems (see Yu 2003 

on using subcategorization for infixation), and -la will attach to all other stems by virtue 

of its less restrictive subcategorization frame. The Nakanai example is thus exactly the 

type of case predicted and accounted for by subcategorization. 

213

4.3.6 Using Subcategorization to model SCA 

In this section, I show how subcategorization works using some examples seen 

above. First, to account for Tzeltal perfective allomorphy, we can propose the 

subcategorization frames shown below. 

(71) Perfective construction A Perfective construction B (‘elsewhere’) 

[[#σ#] verb stem oh perf suffix ] perf verb [[ ] verb stem εh perf suffix ] perf verb 

In this analysis, the -oh suffix left-subcategorizes for a verb stem with only a single 

syllable, while the -εh suffix left-subcategorizes for a verb stem with no phonological 

requirements. This is how -εh is selected as the ‘elsewhere’ allomorph. 

In Nakanai, we can characterize the nominalizing allomorphy seen above as 

subcategorization for a vowel and a stressed syllable. Below are proposed 

subcategorization frames for the two nominalizing allomorphs in Nakanai. 

(72) Nominal construction A Nominal construction B (‘elsewhere’) 

[il nominalizing prefix [V, σ́] verb stem ] noun [[ ] verb stem la nominalizing suffix ] noun 

Here, the il- allomorph right-subcategorizes for the leftmost vowel and stressed syllable 

of a verb stem. In the event that il- cannot be adjacent to both the leftmost vowel and the 

stressed syllable, then the elsewhere allomorph, -la, is selected. This allomorph left- 

subcategorizes for a verb stem with no phonological requirements. 

4.3.7 Estonian revisited 

Given the existence of the subcategorization mechanism discussed above, there 

are two possible reanalyses for the Estonian data that were the impetus for Kager’s claim 

that SCA is output optimization. The first is the subcategorization account outlined 

below. Here, I propose subcategorization frames to account for the allomorphy found in 

214

the genitive plural and partitive plural. Each of these is characterized as having the more 

specific allomorph subcategorize for a stem that ends in a complete foot. 

(73) a. Estonian genitive plural const. A 

[[Ft#] stem te gen pl suffix ] gen pl word 

Estonian genitive plural const. B (‘elsewhere’) 

[[ ] stem tte gen pl suffix ] gen pl word 

b. Estonian partitive plural const. A 

[[Ft#] stem sit part pl suffix ] part pl word 

Estonian partitive plural const. B (‘elsewhere’) 

[[ ] stem it part pl suffix ] part pl word 

A second reanalysis is possible for the Estonian genitive plural and partitive 

plural: a non-suppletive account involving a single underlying form of the affix for each 

morphological category and phonological rules specific to each category: a rule of tt 

degemination for the genitive plural, and a rule of s insertion for the partitive plural. 

These rules are schematized below. 

(74) a. Genitive plural tt degemination b. Partitive plural s insertion 

(σ σ) Ft ( σ σ) Ft 

μ 

stem 

= 

stem s i t 

t t e 

gen pl word 

Ø 

part pl word 

This analysis in terms of phonological rules is possible because the allomorphs in each 

category are phonologically similar to each other, allowing us to propose a single 

underlying form (/-tte/ for the genitive plural and /-it/ for the partitive plural). 

Interestingly, this is true of some other cases of optimizing SCA as well, so that among 

those cases of SCA found in this survey that appear to be optimizing, some may not 

actually involve suppletive allomorphy. For example, in the Shipibo repetitive 

215

allomorphy discussed above, it is possible to propose a single underlying form, /-ribV/, 

that corresponds to the two allomorphs. This is important because not only have we 

demonstrated that SCA is not always optimizing, but it may even be the case that the 

majority of cases of SCA are non-optimizing once we factor out cases of optimizing SCA 

that do not actually involve suppletive allomorphy. 

4.3.8 Summary of analysis 

Given the existence of SCA that output optimization cannot handle, and the 

existence of subcategorization, which can handle these cases as well as those modeled by 

output optimization, we have 2 options for our theory. Option A is to model the 

optimizing cases using output optimization and the non-optimizing cases using 

subcategorization; this is similar to a proposal advanced by Booij (1998). Option B, the 

option advocated here, is to model all cases of PCSA (including SCA) using 

subcategorization. 

Option B avoids the problem of having multiple theoretical mechanisms to model 

a single phenomenon. Option B also captures the fact that SCA appears to be a unitary 

phenomenon, rather than two distinct phenomena that can be easily differentiated, which 

is what Option A implies. As shown below, cases of SCA are distributed along a 

continuum of optimization ranging from optimizing, TETU effects (as in Estonian) to 

anti-optimizing or ‘perverse’ effects (as in Dyirbal). 

(75) A continuum of optimization in SCA 

Optimizing; possible Non-arbitrary distribution; not Arbitrary distribution; ‘Perverse’ 

TETU effects necessarily optimizing non-optimizing 

e.g. Estonian, Turkana Kimatuumbi, Nakanai Tzeltal, Kaititj Dyirbal 

216

Furthermore, Option A represents a poor way of doing science. If we use the P >> M 

model only to account for the type of examples that it handles best, then the model can 

have no predictive power and is completely untestable. Since we are dealing with an 

apparently unitary phenomenon in a single domain of grammar, there is no principled 

reason to use two completely different models to account for different instances of the 

phenomenon. 1 


In this chapter we have seen a number of cases of suppletive allomorphy 

conditioned by elements of the prosodic hierarchy (mora, syllable, or foot). Examples 

from a wide variety of languages were presented in §4.1. One large class of cases 

revealed by the present survey involves the ergative/locative in the Pama-Nyungan 

languages of Australia. On their surface, the PN examples are problematic for my claim 

that PCSA is not inherently optimizing and should be modeled using subcategorization 

rather than Output Optimization because many of the patterns do appear to be optimizing. 

However, a look at the history of the PN ergative/locative allomorphy (§4.2) has revealed 

1 

This reflects a more general issue that has parallels in other areas of research, notably in 

the study of fixed segment reduplication. In a seminal article on the topic, Alderete et al 

(1999) propose two separate analyses, a phonological analysis and a morphological 

analysis, for what they view as two different types of fixed segment reduplication. In 

effect, the examples where the fixed segmentism seems to follow from phonological 

principles get a phonological analysis, while examples where the identity of the fixed 

segment is arbitrary get a morphological analysis. One could criticize this approach along 

the same lines as I have criticized the approach advocated by Booij (1998). However, an 

important difference between the use of a dual approach for fixed segment reduplication 

vs. PCSA is that Alderete et al identify several independent properties of the 

morphological type of fixed segmentism that differ from the phonological type (1999: 

355-356). No such independent properties differentiate the optimizing vs. non-optimizing 

types of PCSA, and for this reason, a dual analysis of PCSA is unmotivated and 

unprincipled. See §6.4 for further discussion of this point. 

217

a source for the apparent effect of optimization that does not involve morphological 

change for the sake of phonological optimization. This eliminates any problem that the 

PN examples might have posed for a subcategorization approach to PCSA. The 

subcategorization approach to prosodic PCSA is spelled out in §4.3 and contrasted with 

Kager’s (1996) proposed use of Output Optimization to model the same phenomenon. 

Subcategorization is argued to be the superior account because it offers a unified analysis 

of PCSA and is able to handle non-optimizing allomorphy. The approach to prosodic 

PCSA described in §4.3 is the same subcategorization approach used in chapters 2 and 3 

to model cases of PCSA conditioned by segments and by stress/tone, respectively. In 

chapter 5, I make explicit some predictions for other types of phonology-morphology 

interactions that are made by this approach and contrast them with the predictions of 

Output Optimization. I will argue that subcategorization offers a superior account of 

these other phenomena in addition to PCSA. 

218

Chapter 5: Predictions of the P >> M model 

In the preceding three chapters we have seen examples of phonologically 

conditioned suppletive allomorphy (PCSA) followed by demonstrations of how these 

cases can be modeled using the subcategorization approach to morphology. In most of 

these cases, the so-called ‘P >> M’ approach espoused by McCarthy and Prince 

(1993a,b), discussed in chapter 1, can handle the examples equally well. In some cases of 

apparently non-optimizing PCSA, the P >> M analysis is inferior to the subcategorization 

approach because the P >> M analysis requires some highly stipulative and/or poorly 

motivated new constraints to be proposed, as in the Tzeltal, Kaititj, Zuni, and Dyirbal 

examples discussed in chapter 4. However, some may argue that this disadvantage of the 

P >> M model is outweighed by its explanatory power, which exceeds that of the 

subcategorization approach in that it is able to make sense of the optimizing nature of 

many of the examples of PCSA observed in the survey; under the subcategorization 

approach, this optimizing character is coincidental (in the synchronic realm). It was 

argued that this failure to capture the apparent optimization is not a fatal shortcoming of 

the subcategorization approach because many examples of PCSA, as demonstrated in 

chapters 2-4, do not appear to involve optimization. 

In this chapter I discuss phenomena other than PCSA where phonological 

considerations have some bearing on affixation. I will demonstrate that, based on what 

we know about those other phenomena, the P >> M model does not match very well with 

the range of phenomena found in the world’s languages. A consideration of other 

phenomena at the phonology-morphology interface thereby lends further support to my 

claim based on PCSA that the subcategorization model is the superior one. 

219

<strong>Phonological</strong> conditions on affixation can in principle be manifested in several 

different ways involving the occurrence, shape, selection, ordering, and placement of 

affixes. In this chapter, I consider each of these possible effects, discussing the 

predictions made by P >> M for each type of effect and then comparing these predictions 

with the range of attested examples of each type among the world’s languages (in 

instances where relevant survey data are available). 

In terms of morpheme occurrence, phonology is expected to determine whether 

or not a particular morpheme will appear in surface forms. On the one hand, there should 

be instances in which a morpheme is inserted for the purpose of phonological well- 

formedness; this corresponds to the phenomenon of the empty morph, discussed in §5.1. 

The other logical possibility is that phonological considerations could prevent a 

morpheme from surfacing, resulting in a phonologically induced morphological gap 

(§5.2). Relating to this, phonology could also in principle determine whether a process 

such as reduplication is allowed to apply; this is covered in §5.3. 

The shape of the surface form of a particular morpheme can be influenced by 

phonology whenever phonological processes end up applying to a morpheme in some 

words but not others depending on the phonological context. This is regular, 

phonologically conditioned (non-suppletive) allomorphy. I will not be discussing this 

type of allomorphy any further here; I consider this to be within the domain of pure 

phonology since it does not have any effect on the actual process of affixation per se. 

A related effect is the phonologically determined selection of multiple different 

underlying forms of an affix. This is the phenomenon of phonologically conditioned 

suppletive allomorphy, which has been the primary subject of this dissertation. The 

220

eason why PCSA is considered as a phonological effect in morphology (while non- 

suppletive allomorphy is not) is that PCSA involves the direct influence of phonological 

factors on the process of affixation itself. PCSA will not be discussed further in this 

chapter since the predictions of P >> M and the cross-linguistic results have already been 

presented for this phenomenon in chapters 2-4. 

A final possible phonological effect on affixation involves the location of an affix 

in the word, in terms of its linear ordering with respect to other morphemes in the word, 

or (in the case of infixation) its placement within the stem itself. <strong>Phonological</strong>ly 

conditioned affix order and infix placement are discussed in §5.4 and §5.5, respectively. 

As will be seen, those phenomena for which survey data are available tend to 

support the subcategorization approach while providing arguments against the P >> M 

approach. This will be discussed further in the conclusion in §5.6. 

5.1 Empty morphs 

The P >> M model predicts that empty morphs (defined by Bauer (1988: 242) as 

‘a recurrent form in a language that does not appear to be related to any element of 

meaning’) should occur based on phonological considerations. In principle, any 

phonological constraint should be able to force the insertion of an empty morph. The 

phonological constraint in question would have to outrank a morphological constraint 

that I will call DEPMORPH, which bans insertion of morphemes not present in the input 

(note that this is a different constraint from the phonological constraint DEP, which 

penalizes insertion of phonological elements). 

221

Some specific types of phonological requirements that are predicted to be met via 

the use of empty morphs are: dissimilation (similar phonological elements are separated 

from one another by insertion of an empty morph), word minimality (subminimal words 

are augmented by an empty morph that contributes enough moras or syllables to allow 

the word to satisfy minimality), licensing (an affix containing a particular marked 

segment or feature is only allowed to attach to the stem if an empty morph containing a 

segment or feature that licenses the marked segment/feature is added to the root), 

alignment (an empty morph that contains a segment that is eligible to link to a particular 

feature is added to the edge of a word so that the feature in question can spread to that 

edge, satisfying alignment), stress clash avoidance (an empty morph consisting of an 

unstressed syllable intervenes between two stressed morphemes to avoid a stress clash), 

anti-homophony (an empty morph occurs in some member of a paradigm to avoid 

confusion with an otherwise homophonous member of the paradigm), and syllable 

structure constraints (an empty morph consisting of a single C or V is inserted between 

V+V or C+C, respectively, to avoid a vowel hiatus or consonant cluster). 

To my knowledge, no cross-linguistic survey has been undertaken to determine 

which of these scenarios actually occurs in any of the world’s languages. 

Impressionistically, some of these possibilities, such as the use of empty morphs to 

satisfy word minimality (attested in Ndebele (Sibanda 2004)) and for stress clash 

avoidance, seem plausible, while others, such as alignment, seem less likely to be 

attested. This is an empirical question for future study. 

222

5.2 <strong>Phonological</strong>ly induced morphological gaps 

Prince and Smolensky (1993) first pointed out the use of OT to account for 

phonologically induced morphological gaps. In the traditional OT approach there is some 

winning candidate corresponding to every input, even if the winning candidate is not very 

‘good’ in that incurs many violations of even highly-ranked constraints. In standard OT, 

phonologically induced morphological gaps should therefore not be allowed. Prince and 

Smolensky (1993) found a way around this problem involving the so-called ‘null parse’ 

and a constraint called MPARSE. If “Ø” is considered as a candidate output alongside the 

faithful candidate and other candidates that are minimally different from the input, then a 

gap could result from the selection of “Ø” as the winning candidate. Prince and 

Smolensky assumed that the null parse satisfies all constraints except MPARSE, and that 

no other candidate violates MPARSE. Thus, any candidate violating a constraint ranked 

above MPARSE is ungrammatical, and if every non-null candidate is ungrammatical, then 

the null parse is selected. 

Under this model, the P >> M ranking that creates gaps is the ranking of 

phonological constraints ahead of the (presumably morphological) constraint MPARSE. In 

principle, any phonological constraint could give rise to gaps. As pointed out by Orgun 

and Sprouse (1999), Prince and Smolensky’s model predicts that any phonological 

constraint that causes a morphological gap will not be violated anywhere else in the 

language. Thus, for example, a language that has morphological gaps due to stress clash 

avoidance cannot have any words exhibiting stress clash. 

Orgun and Sprouse (1999) demonstrate that this prediction is incorrect. The 

analysis of some languages (Turkish, Tagalog, and Tiene are discussed) requires 

223

violation of certain constraints to be allowed in some contexts and to result in 

ungrammaticality in other contexts. This gives rise to a ranking paradox in Prince and 

Smolensky’s model: in some languages a constraint must be ranked above MPARSE for 

some purposes but below MPARSE for other purposes. Prince and Smolensky’s model 

undergenerates, since it cannot account for these three languages. To solve this problem, 

Orgun and Sprouse (1999) propose an alternative approach that involves adding a new 

layer to OT. Their proposal is that EVAL, the familiar OT candidate evaluation module, 

will always select a non-null winning candidate. This candidate must then be submitted to 

a higher level that Orgun and Sprouse term CONTROL. Constraints in CONTROL are 

absolutely inviolable, such that if the winning candidate from EVAL violates any 

constraint in CONTROL, then it is ungrammatical and there will be no output. 

In principle, the revised OT model with CONTROL is compatible with the P >> M 

ranking schema. However, if gaps are created in CONTROL, then the P-M constraint 

interaction that actually creates gaps would be a phonological constraint in CONTROL 

taking precedence over a morphological constraint in EVAL (namely, MORPHEXPR, 

shown below) (Rose 2000a). 1 

(1) MORPHEXPR: An input morphological category is expressed in the output. 

Orgun and Sprouse’s model predicts that every phonological constraint should be 

able to cause a morphological gap. However, unlike in Prince and Smolensky’s model, it 

1 

Since phonological constraints in CONTROL can ‘override’ morphological constraints in 

EVAL, the CONTROL-based model gives rise to a perhaps more restricted set of effects 

similar to those predicted by the EVAL ranking schema P >> M. Orgun and Sprouse 

(1999) do not put any limits on which constraints or constraint types can be part of 

CONTROL, but presumably if the CONTROL proposal were found to overgenerate in the 

same way as P >> M, then this problem could be addressed by putting some principled 

restrictions on which constraints can be in CONTROL. 

224

is not the case that constraints that cause gaps must be inviolable everywhere in the 

language. Given that some languages have gaps in some paradigms caused by constraints 

that are violated elsewhere, Orgun and Sprouse’s model is superior to Prince and 

Smolensky’s in accounting for the kinds of phonologically induced morphological gaps 

that are found in the world’s languages. 

5.3 Reduplication 

Under the Base-Reduplicant (BR) Correspondence model of reduplication 

(McCarthy and Prince 1995), the reduplicant (or RED) is a morpheme that copies 

phonological material from the BASE. In this model, the production of RED is driven by 

the morphological constraint MORPHEXPR (discussed above), and by the phonological 

constraint (or constraint family) BR-CORRESPONDENCE, which enforces similarity 

between RED and the BASE. This model allows for effects of P >> M just like those 

described in §5.2 giving rise to morphological gaps, since certain phonological 

constraints if ranked ahead of MORPHEXPR could prevent the reduplicant from being 

realized. For example, a constraint against stress clash could prevent the reduplicant from 

surfacing if it would have a stressed syllable that would surface next to a stressed syllable 

of the base. A constraint banning sequences of (tier-) adjacent like consonants could 

prevent the reduplicant from being realized if its realization next to the base would have 

resulted in such a sequence. Constraints requiring foot binarity and exhaustive foot 

parsing could prevent a reduplicant with an odd syllable count from surfacing if the base 

has an even syllable count. In general, any possible phonological constraint can in theory 

prevent the realization of a reduplicant. The P >> M model therefore predicts the 

225

existence of cases where the semantic content of a reduplicative morpheme is expressed 

but the phonological form of the word is identical to the base and does not contain any 

reduplicated material. 

Under an alternative approach, Morphological Doubling Theory (MDT; Inkelas 

and Zoll 2005), the same effects of P >> M are predicted. In MDT, reduplication is a 

morphological rather than a phonological process. There is no ‘base’ to be contrasted 

with a ‘reduplicant’, but instead two underlyingly identical instances of the stem, which 

are governed by separate sets of ranked phonological constraints (cophonologies). In this 

model, presumably the expression of each copy of the stem is driven by MORPHEXPR, 

which could be ranked differently between the two cophonologies corresponding to the 

two copies of the stem. If in either cophonology, some phonological constraint outranked 

MORPHEXPR under the P >> M schema, then the P constraint in question could prevent 

that copy of the stem from being realized. However, note that it is probably more difficult 

to generate such an effect under MDT than under BR Correspondence. This is because in 

order to account for why the root (rather than the affix) is always expressed when a 

particular affix is incompatible with it, there must be some very highly-ranked (or 

perhaps inviolable) constraint requiring expression of the root (ROOTEXPR). In the BR 

Correspondence model, RED is an affix, so its expression is driven by the more general 

MORPHEXPR, which is not inviolable or as highly ranked as ROOTEXPR. This is what 

allows the possibility of a phonological constraint blocking the expression of the 

reduplicant. But in MDT, the two copies of the stem have equal status, and neither is 

considered an affix, so both are subject to ROOTEXPR (unless the stem for reduplication 

includes some affixes, in which case those particular parts of each copy of the stem 

226

would not be subject to ROOTEXPR). This means that if ROOTEXPR is inviolable, then 

under MDT, the P >> M model cannot result in null expression of either copy of the 

stem. If ROOTEXPR is not inviolable but only very highly ranked, then a phonological 

constraint could still outrank it, resulting in null expression, but this is no more likely 

than the failure of any root (in a non-reduplicative context) to be expressed. 

5.4 <strong>Phonological</strong>ly conditioned affix order 

Though this has not to my knowledge been made explicit in the OT literature, the 

P >> M model makes broad predictions for phonologically conditioned affix order and 

could in principle be used to model that phenomenon just as it has been used for other 

effects of phonology in morphology (such as suppletive allomorphy, as we have seen; 

other effects will be discussed in subsequent sections of this chapter). In §5.4.1, I discuss 

important predictions made by P >> M for affix order. Then, in §5.4.2, I discuss results of 

a cross-linguistic search for cases of phonologically conditioned affix order, showing that 

the predictions of P >> M for affix order do not hold up in the face of cross-linguistic 

data. Finally, in §5.4.3, I discuss an affix ordering phenomenon in Pulaar (West Atlantic, 

West Africa) which, on its surface, appears to exemplify a type of system predicted by 

the P >> M model to exist. I show, however, that upon close inspection Pulaar turns out 

not to be an example of this, and that its affix order can be reduced almost exclusively to 

semantic scope. 

5.4.1 Predictions for affix order 

The P >> M model makes important predictions for affix order. <strong>Phonological</strong>ly 

227

driven affix order is easily generated by ranking phonological constraints over 

morphological constraints such as Condoravdi and Kiparsky’s (1998) SCOPE and 

Hyman’s (2003) MIRROR, constraints that relate affix order to semantic scope and the 

order of syntactic operations. If P constraints can outrank M constraints, as allowed in OT 

and made explicit by McCarthy and Prince (1993a,b), then phonological constraints can 

outrank these and other morphological constraints governing affix order. Therefore, P >> 

M predicts the existence of phonologically driven affix order. 2 

Canonical OT does not respect morphological constituency. The input to a 

tableau, at least in the standard version of OT, includes the root and all affixes making up 

the word, unordered with respect to each other. 3 Therefore, P >> M predicts that affixes 

can be ordered phonologically. More specifically, it also predicts global, across-the-board 

reordering of all of the morphemes in a word for purposes of phonological optimization. 

This means that in some language, there could be a series of several affixes whose 

relative order (with respect to each other and to the root) is phonologically determined, 

perhaps along a phonological scale. Such a case has never, to my knowledge, been 

discussed as such in the literature, though a putative case from Fula/Pulaar (West 

2 

In fact, McCarthy and Prince (1993a: 85) explicitly claim that ‘…infixation shows that 

phonological constraints can determine even the linear order of morphemes and 

morpheme parts’ [emphasis mine]. Strictly speaking, this is false. The existence of 

phonologically determined infix placement does not entail the existence of 

phonologically determined affix order. Infix placement is not the same as affix order; 

infix placement involves the order of phonological segments making up affixes, not the 

order of the morphemes themselves. Looking at the placement of an infix in a stem does 

not tell us the linear order of the infix morpheme with respect to the stem morpheme, 

unless one admits into the theory the rather strange concept of X being ordered inside Y. 

Therefore, it is still very much an open question whether phonological constraints can (or 

should be allowed to) determine the ‘linear order of morphemes’. 

3 

Though Kiparsky (2000) introduces Stratal Optimality Theory, an approach that does 

respect morphological constituency. 

228

Atlantic, West Africa) will be discussed at length in §5.4.3. In the following section, I 

will discuss all of the possible examples of phonological affix order that were uncovered 

in the course of the search. 

5.4.2 Cross-linguistic findings 

Only 5 possible cases of phonologically driven affix order were identified in a 

search of over 400 grammars of languages representing a wide range of language 

families. The search also included a survey of the theoretical linguistics literature on affix 

ordering. The existence of templatic affix order is well established (see Zwicky 1985, 

Simpson and Withgott 1986, Speas 1990, Stump 1992, 1993, Inkelas 1993, Hyman and 

Inkelas 1997, and Good 2003), so the lack of cases of phonologically conditioned affix 

order is striking and may in itself be considered a negative result for the P >> M 

prediction. 4 I summarize the examples below. 

First, Jacobsen (1973) claims that suffixes in Washo (Hokan, California/Nevada) 

are reordered ‘to insure an even distribution of stressed and unstressed syllables, and to 

draw most sequences of unstressed syllables to the end of the word’ (1973: 9). Affixes 

occur in a non-scope order in some examples. Under P >> M, we can analyze this by 

ranking footing constraints over SCOPE. However, this may also be analyzed via 

subcategorization without P >> M: stressed suffixes subcategorize for a foot to their left. 

4 

Of course it cannot be claimed that no other examples of phonological affix order exist 

in any language. However, the number of grammars examined in the course of this search 

(approximately 400 grammars) is close to the number examined in the survey of 

phonologically conditioned suppletive allomorphy described in chapters 2-4 

(approximately 600 grammars), so it is striking that only five putative examples were 

uncovered here, in contrast to the 137 examples of PCSA found in the other survey. This 

should increase our level of confidence in the claim that phonological affix order is 

uncommon. 

229

Descriptively, the verb in Awtuw (Ram, Papua New Guinea) has 13 affix slots 

(Feldman 1986: 53). The plural -m occurs in suffix position +3 or +6. -m cannot appear in 

+3 if -re Future or -rere Desiderative appears in +5, unless the -iy Imperfective or -kay 

Perfect is in +2. -m also cannot appear in +3 if -(k)ek Conditional is in +4 and nothing 

follows it (Feldman 1986: 71). Where -m cannot occur in +3, it can appear in +6. 

Wherever -m can occur in +3, it can also occur, doubly marked, in both +3 and +6. This 

complicated +3 ~ +6 alternation for plural marking may be phonologically motivated 

because the alternation seems to optimize syllable structure in many contexts. In each 

example where -m is disallowed in +3, the ungrammatical form would have had a more 

complex consonant sequence than the grammatical form. This suggests that the apparent 

morphological conditions on the position of -m are really phonological. Note, however, 

that two consonant sequences are allowed, even where a form with only one consonant in 

the same position is available. Therefore, the alternation is phonologically optimizing 

only for consonant-final roots. Because there is not a phonological explanation for the 

pattern across all root types, this may be best analyzed as morphological conditioning. 

Hargus and Tuttle (1997) use P >> M to account for the placement of the s- 

Negative prefix in Witsuwit’en (Athabaskan, British Columbia). In some examples, s- 

occurs inside the Tense/Aspect prefix (Hargus and Tuttle 1997: 207). With ‘inner’ 

subjects, the s- prefix occurs outside the Tense/Aspect prefix, which Hargus and Tuttle 

claim occurs in order to avoid a complex coda (1997: 207). Hargus and Tuttle’s analysis 

is that the normal order of the prefixes is Neg-T/A, but the order changes so that s- can be 

a coda, except where this would create a complex coda. Hargus and Tuttle (1997: 207) 

formulate this via P >> M using the constraints in (2). 

230

(2) *COMPLEX 

ALIGN-CODA-SNEG: SNEG should be a coda. 

TENSE-STEM: Align the R edge of the Tense prefix to the L edge of the verb stem. 

NEG-STEM: Align the R edge of the Neg prefix to the L edge of the verb stem. 

The ranking *COMPLEX >> ALIGN-CODA-SNEG >> TENSE-STEM >> NEG-STEM selects the 

observed orderings. Note, however, that the data are consistent with a phonological 

metathesis analysis that does not involve P >> M. Since the prefixes in question each 

consist of a single segment, it may be the segments and not the morphemes themselves 

whose order is phonological. Perhaps the regular order is T/A-Neg-, and after the prefixes 

are in place, phonological metathesis occurs to repair complex codas. The metathesis 

does not have to result from the regular constraints of the language; it could be specific to 

the s- prefix (as is the ALIGN-CODA-SNEG constraint proposed by Hargus and Tuttle). 

According to Wiering and Wiering (1994), a series of consonantal suffixes in 

Doyayo (Adamawa-Ubangi, Cameroon) is ordered by scope, except that the -m 

pluralizing suffix is first in any combination. It also occurs before the final consonant of a 

consonant-final verb root. We can model the pattern by ranking a constraint banning [m] 

as the second in a consonant cluster over a constraint aligning -m to the right edge of a 

stem. However, we do not need P >> M: the placement of [m] could result from simple 

phonological metathesis. The generalization that [m] comes first in consonant clusters is 

surface-true, not specific to -m (Wiering and Wiering 1994: 70). Therefore, this can be 

handled via phonological metathesis. We need not assume that the placement of the affix 

itself is phonological. 

We have seen that the four examples presented above do not provide strong 

evidence for the existence of phonological affix order, since each example was either 

weakly documented or consistent with a different analysis not involving phonological 

231

conditioning of affix placement. Furthermore, even if these did constitute examples of 

phonological affix order, none involved the kind of global phonological reordering of 

multiple morphemes that is predicted by the P >> M approach. The correctness of that 

particular prediction hinges on one final possible example, to be discussed in the 

following section. The example is from Fula/Pulaar, a language that appears on its 

surface to uphold the prediction of global phonological reordering. In fact, I will argue 

that despite this surface appearance, affix order in Fula/Pulaar is driven by other, non- 

phonological considerations. 

5.4.3 Pulaar 

In this section, I discuss affix order in Pulaar, a West Atlantic language spoken in 

a wide area of West Africa and comprising a large number of dialects. 5 I will use Pulaar 

as a case study here because, as will be discussed, it appears to exhibit global 

phonological affix reordering of the type predicted by P >> M. The discussion in this 

section is presented in more detail in Paster (2006) and especially Paster (2005b). 

5.4.3.1 Sonority-based affix order? 

Fuuta Tooro Pulaar, which I describe beginning in §5.4.3.1.3, is spoken in the 

Fuuta Tooro region along the border between Senegal and Mauritania. The consultant for 

this study is a 42 year-old speaker who moved to the US from a town near Matam, which 

5 The name ‘Fula’ is sometimes used for all Pulaar dialects plus other dialects known by 

names such as Fulfulde, Fulani, and Fulbe. However, ‘Fula’ usually does not include 

Pulaar, so there is no single cover term for all of the dialects. Since the focus of the 

discussion to follow is a Pulaar dialect, I use ‘Pulaar’ to refer to the entire language 

group. 

232

is in Senegal in the eastern part of the Fuuta Tooro region. Before turning to Fuuta Tooro, 

I discuss in §§5.4.3.1.1-2 the order of affixes in Gombe Fula (Arnott 1970), a dialect of 

northern Nigeria. Arnott (1970: 333, 366) reports that the order of affixes is largely fixed 

in Gombe Fula. In particular, according to Arnott, immediately after the verb stem are 

consonantal suffixes ordered according to the formula ‘TDNR’: all of the /-t/ suffixes 

precede the /-d/ suffixes, which precede the /-n/ suffix, which precedes the /-r/ suffixes 

(1970: 366). This generalization, if true, is interesting (among other reasons) because 

‘TDNR’ corresponds to increasing sonority on the sonority scale (see, e.g., Ladefoged 

1982), meaning that the fixed order of affixes may be phonological (Paster 2001). This 

can be modeled via P >> M and would constitute the first known example of the ordering 

of multiple affixes along a phonological scale, a phenomenon predicted by P >> M. As I 

show, however, there is a better, non-phonological analysis of Gombe Fula affix order. 

This is an important negative result because it means that P >> M overgenerates in 

predicting a phenomenon that apparently does not exist. 

5.4.3.1.1 Gombe Fula affix order 

Arnott described eleven verb suffixes in Pulaar whose basic shape is a single 

consonant. These are shown below (Arnott 1970: 334, 340-364). 6 

6 

In Arnott’s orthography, < ’ > represents glottal stop, is a palatal implosive, 

is a palatal fricative, and and are voiceless and voiced palatal affricates. I have 

normalized Arnott’s transcriptions to the official Senegalese orthography (Hartell 

1993:250) by using spaces where Arnott used hyphens between subject/object markers 

and verbs. 

233

(3) Shape Label Example 

-ɗ DENominative fur-ɗ-a ‘be grey’ 

-t REVersive taar-t-a ‘untie’ 

-t REPetitive soor-t-o ‘sell again’ 

-t REFlexive ndaar-t-o ‘look at oneself’ 

-t RETaliative jal-t-o ‘laugh at... in turn’ 

-t INTensive yan-t-a ‘fall heavily’ 

-d ASSociative nast-id-a ‘enter together’ 

-d COMprehensive janng-id-a ‘read, learn all...’ 

-n CAUsative woy-n-a ‘cause to cry’ 

-r MODal ɓe mah-ir-i Îi ‘they built them with’ 

-r LOCative ’o ’yiw-r-ii ‘he came from’ 

Arnott lists both -C and -VC forms of these suffixes (V is usually [i]). The -VC forms 

occurs somewhat inconsistently, so this may or may not be analyzable as epenthesis. 

The meanings of the suffixes as described by Arnott are summarized below. First, 

the Denominative -ɗ generally attaches to adjectives, converting them to verb stems. The 

Reversive -t suffix produces a verb that ‘undoes’ the result of the plain verb (e.g., the 

reversive corresponding to ‘close’ is ‘open’). The Repetitive -t denotes repetition of an 

action. The Reflexive -t reduces the number of arguments of the verb by one, so that the 

subject performs the action on him/herself. The Retaliative -t indicates that an action is 

done to someone else in retaliation. The Intensive -t indicates ‘completeness, severity, 

intensity, etc.,’ (Arnott 1970: 343). The Associative -d denotes either ‘joint action’ or 

‘action in association with some person or thing’ (1970: 344). The effect on the 

arguments of the verb is to require either a plural subject or else any subject plus a second 

actor introduced by a preposition. The Comprehensive -d indicates ‘totality or 

completeness’ of the subject or object (Arnott 1970: 345). The Causative -n suffix adds 

an object and contributes the meaning ‘cause to,’ ‘arrange for,’ or ‘make’ (Arnott 1970: 

234

346-347). The Modal -r introduces either an instrument or a manner in which an action is 

done. Finally, Locative -r introduces a location where an action is done. 

Arnott claims (1970: 366) that these suffixes have fixed order: ‘As far as [the 

consonantal] extensions are concerned... [the] normal order can be summarized by the 

formula T-D-N-R’. 7 Forms with the ‘TDNR’ order are given below with page numbers 

from Arnott (1970). 

(4) ’o maɓɓ-it-id-ii jolɗe fuu 

3sg close-REV-COM-past doors all 

‘he opened all the doors’ (367) 

’o yam-ɗ-it-in-ir-ii mo lekki gokki kesi 

3sgi healthy-DEN-REV-CAU-MOD-past 3sgj medicine other new 

‘hei cured himj with some new medicine’ (368) 

war-t-ir- 

come-REV-MOD- 

‘bring back’ (367) 

’o maɓɓ-it-ir-ii yolnde hakkiil 

3sg close-REV-MOD-past door slowly 

‘he opened the door slowly’ (367) 

no njooɗ-od-or-too mi ’e maɓɓe 

how sit-ASS-MOD-rel.fut 1sg with 3pl 

‘how shall I sit/live with them?’ (367) 

Arnott also cites several examples of combinations of consonantal suffixes that do 

not obey the ‘TDNR’ generalization. Some of these are shown below (Arnott 1970: 368). 

7 The Denominative -ɗ occurs immediately next to the root, but Arnott omits it from the 

‘TDNR’ formula. 

235

(5) mi wol-d-it-at-aa ’e maɓɓe 

1sg speak-COM-REP-fut-neg with 3pl 

‘I won’t speak with them again’ 

mi hul-n-it-oo mo 

1sg fear-CAU-RET-subjunctive 3sg 

‘I’ll frighten him in turn’ 

’o nyaam-n-id-ii ɗi 

3sg eat-CAU-COM-past 3pl 

‘he fed them all’ 

mi war-t-ir-id-an-te ɗi 

1sg come-REV-MOD-COM-DAT-future 3pl 

‘I’ll bring them all back to you’ 

Arnott claims that exceptions to the ‘TDNR’ generalization involve lexicalized 

combinations of the root and first extension: ‘Variation from the usual order seems to be 

confined to cases where the basic radical and first extension… frequently occur together 

as an extended radical…’ (1970: 370). Lexicalized forms often have idiomatic meanings 

not predictable from the meaning of their parts, and yet these forms have compositional 

meanings. 8 I show in §5.4.3.1.5 that a scope-based analysis avoids having to posit 

lexicalized stems; first, I present below a phonological account based on Arnott’s 

generalization. 

8 This may not be immediately apparent in the example mi war-t-ir-id-an-te ɗi ‘I’ll bring 

them all back to you’ in (5), since Arnott glosses the -t extension here as a Reversive. 

However, since Reversive and Repetitive are homophonous, the [t] here could be an 

instance of the Repetitive suffix instead. In that case, we would have a compositional 

meaning: come + repetitive = come again, i.e., come back. The remaining problem would 

be to account for the causative meaning, make come back, i.e., bring back. Perhaps 

causative meaning is attributed to the verb due to the presence of the 3pl object ɗi, since 

war- ‘come’ usually does not have a direct object. A similar phenomenon occurs in some 

pairs of transitive vs. intransitive verbs in English, such as ‘I walked’ (which is 

intransitive and non-causative) vs. ‘I walked the dog’ (which is transitive and can be 

interpreted as ‘I caused the dog to walk’). This also occurs in some varieties of American 

English in such constructions as to learn someone meaning ‘to teach someone’. 

236

5.4.3.1.2 A P >> M account of Gombe Fula affix order 

Assuming that, despite the exceptions shown above, ‘TDNR’ is the correct 

generalization for the order of consonantal suffixes, we can analyze this phonologically 

using P >> M. As mentioned above, the ‘TDNR’ generalization lends itself to a 

phonological analysis because the order corresponds to increasing sonority, as 

schematized below. 

(6) t d n r 

voiceless stops voiced stops nasals liquids 

sonority 

To model sonority-based affix order, we need a phonological (P) constraint 

enforcing increasing sonority between consonants in separate morphemes, as shown 

below. 

(7) *FALLINGSONORITY C+C: When a consonant C1 is followed by a consonant C2 

across a morpheme boundary, C2 may not be less sonorous than C1. 

The P constraint outranks a morphological (M) constraint, seen below, that requires affix 

order to correspond to semantic scope (Condoravdi and Kiparsky 1998). 

(8) SCOPE: Morphological constituency reflects scope. 

This constraint refers to the proposal (see, e.g., Baker 1985, Bybee 1985, Rice 2000) that 

the order of affixes corresponds to their semantic scope (broadly defined; see Rice 2000) 

so that an affix occurs further from the root than another affix over which it has scope. 

We will make use of this concept in the analysis of Fuuta Tooro Pulaar in §5.4.3.1.4. 

To ensure that the P constraint affects only the order of consonantal suffixes, we 

also need to assume some undominated constraints to prevent non-consonantal suffixes 

from being reordered and to prevent violations of *FALLINGSONORITY C+C from being 

237

epaired by consonant feature changes rather than reordering (I will not formulate these 

here). Under this analysis, the ranking P >> M (*FALLINGSONORITY C+C >> SCOPE) 

selects forms with the TDNR order, even when the order violates SCOPE, as shown 

below. 

(9) **’o irt-in-ir-ii kam supu ’o kuddu ‘he made me stir the soup with a spoon’ 

/irt, -r, -n/ *FALLINGSONORITY C+C SCOPE 

irt-ir-in- *! 

irt-in-ir * 

Based on scope, we expect the order -r-n because Causative has scope over Modal: the 

instrument is used by the causee, so Causative applies to a stem that already refers to an 

instrument. The -n-r order is selected because the sonority constraint outranks SCOPE. 

Crucially, however, the form **’o irt-in-ir-ii kam supu ’o kuddu is not attested by 

Arnott (1970) (indicated by two asterisks above). I constructed this example based on the 

TDNR generalization for the sake of the argument. In fact, not a single example cited by 

Arnott contradicts our scope-based expectation for affix order. I argue in §5.4.3.1.5 that 

this is not accidental. 

This analysis allows for the ‘exceptional’ non-TDNR orderings if we assume, 

following Arnott, that these have lexicalized stems. In these forms, the suffix attaches 

straightforwardly to a stem ending in the consonant formerly belonging to another suffix. 

Thus, we can account for Arnott’s (1970) data using P >> M. But since Arnott’s 

data are also consistent with scope, perhaps P >> M is not needed. Arnott provided no 

examples allowing us to distinguish the two analyses. I present data below from the Fuuta 

Tooro dialect including many examples that distinguish the two analyses in favor of the 

scope-based analysis. I argue in §5.4.3.1.5 that we can extend this analysis to Gombe 

Fula. 

238

5.4.3.1.3 Fuuta Tooro Pulaar (Northeastern Senegal) 

The consonantal suffixes of Fuuta Tooro are shown below. 9 

(10) Shape Label Example 

-ɗ DENominative mi dom-ɗ-ii ‘I became thirsty’ 

-t SEParative/Reversive mi udd-it-ii baafal ŋgal ‘I opened the door’ 

-t REPetitive ’o haal-t-ii ‘he spoke again’ 

-d COMprehensive/Associative mi udd-id-ii baafe ɗe ‘I closed all the doors’ 

-n CAUsative mi jaŋg-in-ii ‘I taught’ 

-r MODal/Instrumental/Locative mi dog-r-ii paɗe ‘I ran with shoes’ 

Note that there are fewer consonantal suffixes here than listed by Arnott (1970). This is 

because some suffixes of Gombe Fula are not used productively in Fuuta Tooro and 

because in some cases Arnott distinguished suffixes where the data in Fuuta Tooro 

suggest a single suffix (e.g., Modal/Instrumental/Locative). See Paster (2005b) for 

details. 

Several pairwise combinations of consonantal suffixes exhibit ordering 

alternations directly related to scope. For example, when Comprehensive has scope over 

Separative, the Comprehensive -d is ordered outside Separative -t, as in (11)a. The scope 

relation is shown by the fact that the comprehensive action takes place simultaneously, a 

meaning contributed by the Comprehensive (Paster 2005b). In (11)b, the order is 

reversed. This corresponds to Separative having scope over Comprehensive, evidenced 

by the ‘sequential’ meaning, which results from the fact that Separative has no 

simultaneous action meaning. The original action takes place simultaneously, but the 

undoing does not. 

9 

Each suffix also has a -VC form as in Gombe Fula. For Fuuta Tooro, I use the official 

Senegalese Pulaar orthography, which omits predictable word-initial glottal stops, uses 

for the palatal nasal, and uses a ‘hooked y’ for the palatal implosive, which I replace 

with . 

239

(11) a. o sok-t-id-ii baafe ɗe fof 

3sg lock-SEP-COM-past doors det all 

‘he unlocked all the doors (at once)’ 

b. o sok-d-it-ii baafe ɗe fof 

3sg lock-COM-SEP-past doors det all 

‘he unlocked all the doors (in sequence)’ 

The Repetitive -t is ordered after the Comprehensive -d when Repetitive has 

scope over Comprehensive, as in (12)a. The fact that Repetitive has scope over 

Comprehensive is evidenced by the repetitive meaning in each case applying not only to 

the verb, but also to the participants referred to by the Comprehensive. When 

Comprehensive has scope over Repetitive, as in (12)b, the Repetitive -t is ordered first. 

The evidence for Comprehensive having scope over Repetitive is that in this example, the 

same participants are not necessarily involved in both the original and repeated actions. 

The Repetitive applies only to the verb, and then the Comprehensive applies to the output 

of Repetitive affixation, which is a repeated action. This is consistent with the order of 

the affixes. 

(12) a. min cok-t-id-ii baafal ŋgal 

1pl lock-REP-COM-past door det 

‘we all locked the door again together’ (someone else locked it before) 

b. mi yaa-d-it-ii ’e makko 

1sg go-COM-REP-past with 3sg 

‘I went with her again’ (I went with her before) 

As shown in (13), the Causative -n is ordered after the Separative -t. This is 

consistent with scope, since the Causative refers not to the original action, but to the 

‘undoing’. Thus, Causative applies to a verb that already has the separative meaning, 

which is consistent with the ordering of the Causative -n outside the Separative -t. 

240

(13) o haɓɓ-it-in-ii kam ɓoggol ŋgol 

3sg tie-SEP-CAU-past 1sg rope det 

‘he made me untie the rope’ 

If order is scope-based, we predict that the opposite order should correspond to 

the opposite scope relation, as was seen above where Separative-Comprehensive and 

Repetitive-Comprehensive were combined. In the case of Causative-Separative, however, 

it is impossible to find an order alternation corresponding to a meaning change because it 

is apparently impossible for Separative to have scope over Causative. This is explained 

by the fact that Separative generally applies to a verb whose semantics involve putting 

things together. Thus, in order for Separative to apply to a Causative, the entire Causative 

verb would have to have a ‘putting together’ meaning. There are apparently no verbs 

corresponding to ‘make be together’ that use the Causative -n such that a Separative 

would be expected to attach to the Causativized stem (see Paster 2005b for details). 

When the Repetitive -t combines with the Causative -n, both orderings are 

acceptable, corresponding to scope. When Repetitive has scope over Causative, the 

Causative -n precedes the Repetitive -t, as in (14)a. The scope relation is indicated by the 

fact that the same agent causes both the original and repeated actions. Thus, Repetitive 

applies to a Causativized verb, corresponding to the ordering of the Repetitive suffix 

outside the Causative. As predicted by scope, the opposite order of the Causative and 

Repetitive suffixes corresponds to the opposite scope relation from (14)a. When 

Causative has scope over Repetitive (14)b, the Repetitive -t precedes the Causative -n. 

The scope relation is evidenced by the fact that the original action is understood to have 

been done voluntarily rather than being caused by the same agent who causes the 

repeated action. Thus, Repetitive applies to the bare verb, and Causative applies to the 

241

Repetitive verb, meaning that causation applies to the repeated action and not necessarily 

to the original action. 

(14) a. o sood-it-in-ii een deftere nde 

3sg buy-REP-CAU-past 1pl book det 

‘she made us buy the book again’ 

b. o sood-in-it-ii een deftere nde 

3sg buy-CAU-REP-past 1pl book det 

‘she made us buy the book again’ (we bought it before voluntarily) 

The relative order of the Separative -t and Modal -r corresponds to their scope. In 

(15), Modal has scope over Separative, as indicated by the fact that the instrument is used 

to undo the action and not necessarily to do the original action. Thus, the scope of the two 

suffixes in this example corresponds to the ordering of Modal -r outside Separative -t. 

(15) a sok-t-ir-ii baafal ŋgal coktirgal 

2sg lock-SEP-MOD-past door det key 

‘you (sg.) unlocked the door with a key’ 

It is apparently impossible to produce a single verb form where Separative has 

scope over Modal. When asked to produce such a form corresponding to, e.g., ‘we un- 

sewed the shirts with a needle,’ ([we un-[sewed the shirts with a needle]]) where the 

needle was used to do the sewing but not the unsewing, the speaker is unable to express 

this with a single verb. Therefore, we cannot test the prediction of the scope principle that 

Separative should occur after Modal when Separative has scope over Modal. 

When Modal has scope over Repetitive (16), -r is ordered after -t, as predicted. It 

is clear that Modal has scope over Repetitive here since it is specified that a different 

instrument is used in the original vs. repeated action. Thus, Repetitive applies to the verb 

first, and then Modal applies to the Repetitive stem. 

242

(16) mi irt-it-ir-ii supu o kuddu goɗɗo 

1sg stir-REP-MOD-past soup det spoon different 

‘I stirred the soup again with a different spoon’ 

When Repetitive has scope over Modal, the Modal -r suffix is ordered after the 

Repetitive -t suffix, as shown in (17). 

(17) mi udd-it-ir-ii baafal ŋgal sawru 

1sg close-REP-MOD-past door det stick 

‘I closed the door with a stick again’ (the same stick) 

This is the first example we have seen where the affix order does not correspond to 

scope. Based on scope, we expect Repetitive -t to be ordered after Modal -r here. We 

know that Repetitive has scope over Modal because it is understood that the same 

instrument is used for both the original and repeated actions. 10 This corresponds to the 

application of Modal to the verb root, and then application of Repetitive to the verb that 

already has an instrument so that the repeated action involves the use of the same 

instrument. Since this order has no apparent semantic explanation, I assume the -t-r order 

is part of a morphological template. We will account for this in the analysis in §5.4.3.1.4. 

The order of the Causative -n with the Comprehensive -d is scope-based. When 

Comprehensive has scope over Causative, Causative -n precedes the Comprehensive -d, 

as in (18)a. When Causative has scope over Comprehensive, the Causative -n is ordered 

after the Comprehensive -d as predicted. However, the opposite ordering, -n-d, is also 

compatible with this reading, as in (18)b. This may be due to the difficulty of 

10 

It is possible that the ‘same instrument’ reading is arrived at contextually, since this 

would likely be the default interpretation of ‘I closed the door with a stick again’ in, e.g., 

English. However, (17) contrasts crucially with the type of example seen in (16), where 

the consultant states specifically that a different instrument is used in the original vs. 

repeated action. Unfortunately I do not have any Modal-Repetitive sentences that differ 

minimally in this respect (using exactly the same verb and instrument), but there are 

parallel cases involving the Modal and Comprehensive ((19)a and b) which do show a 

minimal difference in the same vs. different instrument reading. 

243

constructing English stimuli where Causative has unambiguous scope over 

Comprehensive. When the speaker is given with English sentences like this, he may 

interpret them so that Comprehensive has over Causative, explaining why -n-d is 

accepted. The meaning difference is subtle, and this should be investigated in 

conversation or narratives. 

(18) a. ɓe njal-n-id-ii mo 

3pl laugh-CAU-COM-past 3sg 

‘we all made him laugh together’ 

b. mi woy-d-in-ii ɓe ~ mi woy-n-id-ii ɓe 

1sg cry-COM-CAU-past 3pl 

‘I made them cry together’ 

When the Comprehensive combines with the Modal so that Comprehensive has 

scope over Modal, Comprehensive -d is ordered after Modal -r, as in (19)a. The scope 

relation is indicated by the fact that a different instrument is used to perform the action on 

each object. When Modal has scope over Comprehensive, as in (19)b, Modal -r is 

ordered after Comprehensive -d. The fact that Modal has scope over Comprehensive here 

is clear since the same instrument is used to perform the action on each object. This is 

consistent with the application of Modal to a stem already having the comprehensive 

meaning so that the instrument applies to all of the objects referred to by the 

Comprehensive. 

(19) a. mi sok-r-id-ii baafe ɗe coktirgal goɗŋgal 

1sg lock-MOD-COM-past doors det key different 

‘I locked each of the doors with a different key’ 

b. mi sok-d-ir-ii baafe ɗe coktirgal 

1sg lock-COM-MOD-past doors det key 

‘I locked all of the doors with a key’ (the same key) 

244

The final pairwise combination of suffixes shows variation. When Causative has 

scope over Modal, we expect -n to come after -r. We do find this order corresponding to 

this scope reading, but the opposite order can also be used, as in (20)a; either order is 

allowed with no meaning difference. We find the same variation when Modal has scope 

over Causative, as in (20)b. Here, we expect Modal -r to be ordered after Causative -n, 

but the opposite order can also be used. Our analysis will need to account for this. 

(20) a. o irt-ir-in-ii kam supu o kuddu ~ 

3sg stir-MOD-CAU-past 1sg soup det spoon 

o irt-in-ir-ii kam supu o kuddu 

3sg stir-CAU-MOD-past 1sg soup det spoon 

‘he made me stir the soup with a spoon’ (I used a spoon) 

b. o irt-in-ir-ii kam supu o laɓi ~ 

3sg stir-CAU-MOD-past 1sg soup det knife 

o irt-ir-in-ii kam supu o laɓi 

3sg stir-MOD-CAU-past 1sg soup det knife 

‘he made me stir the soup with a knife’ (he used a knife) 

This exhausts the pairwise combinations of the consonantal suffixes. Based on 

these examples, we have three generalizations about the order of consonantal suffixes in 

Fuuta Tooro (21). In §5.4.3.1.4 I propose an analysis to account for these generalizations. 

(21) a. Repetitive -t precedes Modal -r regardless of their relative scope. 

b. Causative -n and Modal -r are freely ordered with each other regardless of 

scope. 

c. Otherwise, order is determined by scope. 

5.4.3.1.4 A scope-template account of Fuuta Tooro affix order 

Rice (2000) claims that scope determines the order of affixes whenever there is a 

scope relation between them. Otherwise, order is arbitrary. Templates are ‘emergent’; in 

OT terms, SCOPE always outranks TEMPLATE. However, a language with scope-based 

245

order can have elements of fixed order that override scope, as seen above. This follows 

from TEMPLATE >> SCOPE; a ranking I use in the analysis of Fuuta Tooro affix order 

below. 

In this analysis, I break TEMPLATE into three separate constraints, shown below. 

(22) TREP PRECEDES R: Repetitive -t precedes Modal -r. 

N PRECEDES R: Causative -n precedes Modal -r. 

R PRECEDES N: Modal -r precedes Causative -n. 

The observed affix order effects will be analyzed by ranking these templatic constraints 

above SCOPE so that scope determines the order of affixes as long as the template is not 

violated; when scope and the template conflict, the template ‘wins’. 

The ranking TREP PRECEDES R >> SCOPE successfully selects forms where the 

template and scope agree, as shown in the tableau in (23). 

(23) mi irt-it-ir-ii supu o kuddu goɗɗo ‘I stirred the soup again with a different spoon’ 

/irt, -t, -r/ TREP PRECEDES R SCOPE 

irt-it-ir- 

irt-ir-it- *! * 

Since the repeated action is not done with the same instrument as the original action, 

Modal has scope over Repetitive, and therefore we expect Modal -r to be ordered after 

Repetitive -t. This is the order that is selected since SCOPE agrees with the template. 

This ranking also selects forms where TEMPLATE forces a violation of SCOPE (24). 

(24) mi udd-it-ir-ii baafal ŋgal ‘I closed the door with a stick again’ (same stick) 

/udd, -t, -r/ TREP PRECEDES R SCOPE 

udd-it-ir- * 

udd-ir-it- *! 

The scope order is -r-t since Repetitive outscopes Modal (the same instrument is used in 

the original and repeated actions), but the -t-r order is selected by TREP PRECEDES R. 

246

The order of -n and -r is handled by variable ranking. In ranking #1, N PRECEDES 

R >> R PRECEDES N, SCOPE. This selects forms where -n precedes -r (25). 

(25) o irt-in-ir-ii kam supu o laɓi ‘he made me stir the soup with a knife’ (he used a knife) 

/irt, -n, -r/ N PRECEDES R R PRECEDES N SCOPE 

irt-in-ir- * 

irt-ir-in- *! * 

Here, the order predicted by scope is -n-r, since Modal has scope over Causative (the 

causer, not the causee, uses the instrument). Since this agrees with the high-ranked N 

PRECEDES R constraint, the winning candidate also satisfies SCOPE. 

Our ranking also selects forms where SCOPE is violated (26). 

(26) o irt-in-ir-ii kam supu o kuddu ‘he made me stir the soup with a spoon’ (I used a spoon) 

/irt, -r, -n/ N PRECEDES R R PRECEDES N SCOPE 

irt-in-ir- * * 

irt-ir-in- *! 

The scope order is -r-n since Causative has scope over Modal (the causee uses the 

instrument), but since N PRECEDES R outranks SCOPE, the non-scope -n-r order wins. 

In ranking #2, R PRECEDES N outranks N PRECEDES R and SCOPE. This selects 

forms with the Modal -r preceding the Causative -n regardless of scope. As seen in the 

tableau in (27), this ranking selects forms where the template and scope agree. 

(27) o irt-ir-in-ii kam supu ’o kuddu ‘he made me stir the soup with a spoon’ (I used a spoon) 

/irt, -r, -n/ R PRECEDES N N PRECEDES R SCOPE 

irt-in-ir- *! * 

irt-ir-in- * 

Here, the scope order is -r-n since Causative has scope over Modal (the causee uses the 

instrument). This agrees with R PRECEDES N, so the winner obeys SCOPE as well. 

This ranking also selects forms where scope and the template disagree, so that the 

winning candidate obeys the template at the expense of a scope violation (28). 

247

(28) o irt-ir-in-ii kam supu ’o laɓi ‘he made me stir the soup with a knife’ (he used a knife) 

/irt, -n, -r/ R PRECEDES N N PRECEDES R SCOPE 

irt-in-ir- *! 

irt-ir-in- * * 

Here, the scope-based order is -n-r since the causer uses the instrument. However, 

because R PRECEDES N outranks SCOPE, the winning candidate has the -r-n order. 

The two constraint rankings are summarized below. 

(29) Ranking #1: TREP PRECEDES R, N PRECEDES R >> R PRECEDES N, SCOPE 

Ranking #2: TREP PRECEDES R, R PRECEDES N >> N PRECEDES R, SCOPE 

By using template constraints and SCOPE, all M constraints, we have captured the order 

of consonantal suffixes in Fuuta Tooro without P >> M. We will now revisit Gombe 

Fula, and I will show that a scope-based analysis is possible for that dialect as well. 

5.4.3.1.5 A scope-based reanalysis of Gombe Fula 

As mentioned earlier, every example provided by Arnott (1970) is consistent with 

scope. Though most of Arnott’s examples obey the ‘TDNR’, they are also consistent with 

scope. For instance, -t-r in (30) is a ‘TDNR’ order. However, the order is also consistent 

with scope, since the adverb ‘slowly,’ introduced by Modal, applies to the Reversive 

action, not the original action. The Modal applies to a verb to which Reversive has 

already applied, corresponding to the ordering of the Modal -r outside the Reversive -t. 

(30) ’o maɓɓ-it-ir-ii yolnde hakkiilo 

3sg close-REV-MOD-past door slowly 

‘he opened the door slowly’ (Arnott 1970: 367) 

(31) is also consistent with both scope and the ‘TDNR’ generalization. Here, the order 

corresponds directly to the order of logical operations performed on the root. First, the 

Denominative -ɗ attaches to the adjective, converting it into a verb meaning ‘be healthy’. 

248

Then, the Repetitive -t applies to this stem, yielding a new stem meaning ‘be healthy 

again’ (=‘be cured’). Next, the Causative -n applies, resulting in a stem meaning ‘make 

be cured’ (=‘cure’). Finally, the Modal -r attaches to this stem, introducing an instrument, 

giving the meaning ‘cure with (some new medicine)’. The order of attachment of the 

affixes is reflected in the order of the suffixes: -t-n-r (Repetitive-Causative-Modal). 

(31) ’o yam-ɗ-it-in-ir-ii mo lekki gokki kesi 

3sgi healthy-DEN-[REP]-CAU-MOD-past 3sgj medicine other new 

‘hei cured himj with some new medicine’ (Arnott 1970: 368) 

Thus, the ordering generalization that Arnott (1970) accounts for using the ‘TDNR’ 

generalization can also be accounted for by scope. The two examples shown above are 

the clearest examples, but no example contradicts the scope generalization. 

Not only does scope account for Arnott’s (1970) examples obeying the ‘TDNR’ 

generalization, but it also accounts for the ‘exceptions’, which Arnott explained as having 

lexicalized stems. Several of Arnott’s exceptions can be explained straightforwardly 

based on scope. In (32), Repetitive has scope over Comitative. 

(32) mi wol-d-it-at-aa ’e maɓɓe 

1sg speak-COM-REP-future-negative with 3pl 

‘I won’t speak with them again’ (Arnott 1970: 368) 

Ignoring the Negative, the form mi-wol-d-it-ii ’e maɓɓe ‘I spoke with them again’ would 

mean that the subject had spoken with ‘them’ before and did so again. This can be 

schematized as [[speak with] again]. Thus, the order of the affixes corresponds to scope. 

Similarly, in (33), the Comitative -d has scope over Causative -n, since the word 

‘fed’ is used in the English translation. 

(33) ’o nyaam-n-id-ii ɗi 

3sg eat-CAU-COM-past 3pl 

‘he fed them all’ (Arnott 1970: 368) 

249

Once again, an apparent exception is explained straightforwardly based on scope. 

Finally, in (34), Retaliative has scope over the Causative since the term ‘frighten’ 

in the English gloss means ‘cause to fear’. The interpretation is [[cause to fear] in turn], 

which corresponds to the -n-t order, which disobeys the ‘TDNR’ generalization. 

(34) mi hul-n-it-oo mo 

1sg fear-CAU-RET-future 3sg 

‘I’ll frighten him in turn’ (Arnott 1970: 368) 

Since some of Arnott’s exceptional forms are explained by scope, and none of his 

examples contradict scope, we can say that Arnott provided no evidence for a non-scope 

principle in the order of consonantal suffixes. The scope-based analysis allows an 

explanation of the ‘exceptional’ forms, which Arnott ignored. It also avoids the problem 

that the meaning of Arnott’s lexicalized stems is derived straightforwardly from their 

component parts rather than being idiomatic as ‘frozen’ forms often are. 

Thus, we have seen that neither Fuuta Tooro Pulaar nor Gombe Fula exhibits 

phonological affix order. In each dialect, the order of consonantal suffixes is scope-based, 

with some exceptions in Fuuta Tooro that are not phonologically conditioned. 

Recall from §5.4.2 that only four other possible cases of phonological affix order 

were uncovered in this study, and no example was particularly compelling. Thus, with 

this alternative explanation for the Fula/Pulaar example, we now appear to have no 

examples of phonological affix order requiring P >> M. This significantly weakens the 

case for P >> M, since a major prediction of this model is not substantiated cross- 

linguistically. 

250

5.5 Infixation 

Yu (2003) critiques what he terms the Displacement Theory (DT) account of infix 

placement, exemplified by the Prosodic Optimization approach of McCarthy and Prince 

(1993b). The idea behind this approach is that, under the P >> M ranking schema, 

infixation is accounted for by ranking prosodic well-formedness constraints ahead of the 

morphological constraints that state whether a given affix is a prefix or suffix (under this 

model, no affix is an underlying infix). In the original incarnation of the Prosodic 

Optimization approach, alignment constraints were gradiently violable, so that a violation 

was incurred for each segment by which a prefix or suffix strayed into the root from its 

underlying, peripheral position. Subsequently, McCarthy (2003) has revised the model so 

that constraints can only be violated categorically. 

Yu (2003) argues that the Prosodic Optimization approach to infixation makes 

several inaccurate predictions. The first is that affixes should surface as infixes only if the 

result is prosodically better-formed than if the affix had surfaced in its underlying, 

peripheral position. Contrary to this prediction, Yu finds several cases in his survey of 

141 examples of infixation in 101 languages where certain affixes are obligatorily 

infixed, never surfacing in what McCarthy and Prince would claim to be their underlying 

prefixed or suffixed positions. Alabama, Archi, and Tagalog (in the pluractional affix) are 

cited as having such infixes. An analysis where an affix that invariably surfaces as an 

infix is treated as a formal prefix or suffix is counterintuitive and unnecessarily abstract 

in comparison to the alternative, that these are underlying infixes. This criticism holds not 

of the P >> M ranking schema in general, but of McCarthy and Prince’s use of it in a 

model specific to infixation. Nonetheless, since my overarching purpose is to contrast the 

251

subcategorization approach as a coherent model with the P >> M schema as part of a 

coherent model, I am focusing on predictions of P >> M as it has been employed in the 

mainstream literature, not on predictions that would be made under some different use of 

P >> M. Therefore, I take this problem with the Prosodic Optimization account of 

infixation to be a shortcoming of the P >> M schema as a whole. 

A second inaccurate prediction of Prosodic Optimization pointed out by Yu 

(2003) is that infixation should always result from considerations of prosodic well- 

formedness. Yu points out that prosodic optimization is not sufficient to account even for 

languages where an affix occurs only sometimes as an infix. In some such languages, a P 

>> M account for infix placement works only if the P constraint is an Edge Avoidance 

constraint. Edge Avoidance may be construed as optimizing (certainly it can be made into 

an OT constraint), but it cannot be said to increase phonological well-formedness. Thus, 

the Prosodic Optimization model at least has to be weakened enough to allow an Edge 

Avoidance constraint to be the P constraint in the relevant P >> M ranking. Here again, 

this incorrect prediction is not made by P >> M in general, but by the implementation of 

it in which the P constraint has to be a well-formedness constraint. 

In addition to these two ways in which the Prosodic Optimization account 

undergenerates, Yu argues that this approach also overgenerates. Namely, it predicts what 

Yu calls ‘hyperinfixation’, where an affix that underlyingly belongs at one edge of a root 

migrates any number of segments towards the opposite edge, even going so far as to 

surface on the opposite periphery. For example, a formal prefix might surface just one 

segment in from the right edge of the root or even as a suffix. But hyperinfixation is not 

attested in any of the languages in Yu’s (2003) broad cross-linguistic survey of languages 

252

exhibiting infixation. Yu points out that there are ways to avoid hyperinfixation using the 

proper constraints and rankings, but questions why we should need special mechanisms 

to rule out hyperinfixation given that it apparently never occurs in any language. 

A possible example of hyperinfixation not discussed by Yu (2003) is the 

phenomenon better known as mobile affixation. ‘Mobile’ affixes, which occur sometimes 

as prefixes and sometimes as suffixes depending on the phonological context, have been 

documented in Huave (Noyer 1994) and Afar (Fulmer 1991). Under the Prosodic 

Optimization approach, mobile affixation can receive the same treatment as infixation 

and could be construed as hyperinfixation. However, if this is to be counted as extreme 

hyperinfixation, then we are left with the puzzling question of why there are no ‘in- 

between’ cases. That is, if an underlying prefix can surface as a suffix or vice versa, then 

why do we not find any examples of an underlying prefix surfacing as an infix at the far 

right edge of the word? The lack of such examples suggests that Huave and Afar are 

better treated as cases of phonologically conditioned suppletive allomorphy, where one 

allomorph is a prefix and the other is a suffix. As cases of PCSA, Huave and Afar are 

unusual in that the separate allomorphs are identical in shape and differ only in their 

position with respect to the stem. However, they differ only minimally from Chimariko, 

which, as described by Conathan 2002, has a phenomenon similar to mobile affixation 

except that the prefix and suffix allomorphs are slightly phonologically different from 

each other, making it clear that this is a case of suppletive allomorphy rather than a single 

affix occurring in different positions. Thus, within what I consider to be PCSA, there are 

cases where two or more different phonologically selected allomorps occur in the same 

position (as in most of the examples described in chapters 2, 3, and 4), cases where two 

253

or more differently shaped allomorphs occur in different positions in the word (as in 

Chimariko), and cases where identical allomorphs occur in different positions in the word 

(Huave and Afar). Since mobile affixes in Huave and Afar never surface as infixes, they 

should not be counted as cases of hyperinfixation, and therefore the Prosodic 

Optimization model still overgenerates in predicting this phenomenon. 


In this chapter, I have discussed some predictions of the P >> M ranking schema 

for several different phenomena: affix order, infixation, morphological gaps, empty 

morphs, and reduplication. Those phenomena for which significant cross-linguistic data 

are available show that the P >> M model makes many predictions that are inaccurate, 

which offers further substantiation for the point made in chapter 1 that the P >> M model 

makes incorrect predictions for phonologically conditioned suppletive allomorphy 

(PCSA). We have seen that in each of the logically possible effects of phonology on 

affixation, the P >> M predicts a wider range of effects than are actually attested, and yet 

there are also instances in which P >> M fails to predict or account for certain types of 

effects (for example, opaquely conditioned PCSA (chapters 2-4), and infixes that always 

surface as infixes but never as peripheral affixes (§5.5)). Thus, the P >> M model both 

over- and undergenerates in the domain of several different phenomena at the phonology- 

morphology interface. 

254

Chapter 6: Conclusion 

In this dissertation, I have presented a survey of examples of phonologically 

conditioned suppletive allomorphy (PCSA). In chapters 2, 3, and 4, I discussed examples 

of PCSA conditioned by segments and features, tone/stress, and prosodic elements, 

respectively. In each of these chapters, typological generalizations were drawn for each 

type of PCSA, and these generalizations were compared with predictions made by two 

competing models of phonological conditions on affixation, namely, the P >> M 

approach and the subcategorization approach. It was argued that the survey results for 

PCSA are more consistent with subcategorization than with P >> M. In chapter 5, I laid 

out the predictions of P >> M for types of phonological conditions on affixation other 

than PCSA. It was shown that results for each of these other types of effects converge 

upon the same conclusion drawn here, that the P >> M model both over- and 

underpredicts with respect to the range of phonological conditions on affixation found in 

the world’s languages. The subcategorization approach, on the other hand, is compatible 

with the cross-linguistic findings and is therefore a superior model of phonological 

conditions on affixation. 

In this chapter, I summarize the findings and arguments presented throughout the 

dissertation. I begin in §6.1 with a summary of the predictions of the P >> M and 

subcategorization approaches. Then, in §6.2, I summarize the survey results presented in 

chapters 2-4. In §6.3, I discuss the theoretical implications of both the survey of PCSA 

and the findings for other types of phonological conditions on affixation. In §6.4, I 

consider and argue against a ‘hybrid model’ using both subcategorization and P >> M,. 

§6.5 provides arguments in favor of a particular way of eliminating the possibility of P 

255

M in an OT grammar. Finally, in §6.6, I conclude with some suggested directions for 

future research on phonological effects in morphology. 

6.1 Summary of predictions 

In chapter 1, I laid out the general predictions of the P >> M and 

subcategorization approaches. In this section, I reiterate those predictions before 

summarizing the cross-linguistic findings in §6.2. 

As discussed in chapter 1, the predictions of P >> M for PCSA can be 

summarized as in (1) below (repeated from chapter 1, §1.1.1.2). 

(1) a. PCSA is ‘optimizing’ and analyzable using preexisting P constraints 

b. PCSA is sensitive to phonological elements in surface forms, not underlying 

forms 

c. <strong>Phonological</strong> conditioning between stem and affix can be bidirectional 

d. <strong>Conditions</strong> on allomorph selection can be located anywhere in the word 

Prediction (1)a means that the P constraints used to model PCSA in a P >> M account 

should be motivated elsewhere, either in the language being analyzed or as a widely 

attested constraint in other languages. Prediction (1)b indicates that PCSA should result 

in surface-true generalizations about allomorph distribution, since allomorph selection is 

handled simultaneously with regular phonological processes. By prediction (1)c, the 

conditions determining allomorph distribution are expected to come from morphemes 

that are either closer to or farther from the root than the affix undergoing the allomorphy, 

since as discussed in chapter 1, standard P >> M analyses do not observe morphological 

bracketing. Finally, prediction (1)d states that conditions on PCSA are not limited to stem 

edges and can instead be located anywhere in the word. This means, for example, that 

prefix allomorphy could be conditioned by a stem-final segment, or suffix allomorphy 

256

could be triggered by a stem-initial segment, or allomorphy in a peripheral affix could be 

triggered by some element in the middle of the stem. 

The predictions of the subcategorization model for PCSA, also discussed in 

chapter 1, are summarized below. 

(2) a. PCSA is not always phonologically optimizing 

b. PCSA is sensitive to phonological elements in underlying/input forms, not surface 

forms 

c. <strong>Phonological</strong> conditions on PCSA can come only from the ‘inside’ 

d. Affix allomorphs occur adjacent to the phonological elements of stems that 

condition their distribution 

Prediction (2)a means that we should find some examples of PCSA in which the 

distribution of allomorphs does not optimize words. This is in contradiction to prediction 

(1)a, made by the P >> M approach, that PCSA can be accounted for by using established 

phonological well-formedness constraints. Prediction (2)b means that we should find 

instances in which PCSA is conditioned crucially by an input element that does not 

surface, rendering the conditions on allomorphy opaque. This can be contrasted with 

prediction (1)b, made by P >> M, that PCSA should be sensitive to surface elements. If 

prediction (2)c is correct, we should find that PCSA in an affix can be conditioned only 

by a property of the root or another affix closer to the root, and never to an affix farther 

from the root. On the other hand, as discussed above, the P >> M model predicts the 

existence of conditioning from both the ‘inside’ and the ‘outside’ (prediction (1)c). 

Finally, prediction (2)d states that affixes should be conditioned by the part of the stem to 

which they are immediately adjacent. This means that we expect prefixes to be 

conditioned only by elements at the left edge of the stem, and suffixes to be conditioned 

only by elements at the right edge of the stem. In contrast, as discussed above, P >> M 

257

predicts that we should find examples that do not obey this kind of locality restriction 

(prediction (2)d). 

This concludes the general summary of predictions of the P >> M and 

subcategorization approaches. As can be seen, the predictions outlined here constitute 

clear-cut differences between P >> M and subcategorization. The reader should keep 

these different predictions in mind when considering the survey results from chapters 2-4, 

which will be discussed in the following section. 

6.2 Summary of survey results 

In chapters 2-4, I presented the results of a cross-linguistic survey of examples of 

PCSA. One hundred thirty-seven examples were found in 67 different languages, in a 

survey of sources on 600 languages. Here, I summarize the findings of the survey. 

The examples in chapter 2 involve conditioning by segments or features. Several 

important generalizations were drawn regarding these examples. As was seen in chapter 

2 (§2.1.2), it appears that the C/V distinction or the presence of any phonological feature 

of a segment can condition PCSA. The conditioning segment or feature is always at the 

same edge of the stem where PCSA occurs. Thus, PCSA in a prefix is triggered by stem- 

initial segments or features, while PCSA in a suffix is triggered by stem-final segments or 

features. It was also found that conditioning always comes from an element closer to the 

root; for instance, if a suffix undergoes allomorphy, then this can be triggered by a suffix 

to its left or by the root itself, but it apparently cannot be triggered by a suffix to its right. 

In this chapter, all of the examples involve allomorphy in an affix or clitic conditioned by 

a stem. Another important finding is that there are several cases in which PCSA does not 

258

appear to be optimizing (§2.1.2.5), and where the relationship between the allomorphs 

and their distribution instead seems arbitrary. Finally, examples were found where the 

condition on PCSA was made opaque by a phonological process (§2.1.2.6), such that the 

original condition on allomorphy was lost or changed in the surface form. As was 

discussed, this shows that PCSA is sensitive to input phonological elements rather than 

output elements. 

Chapter 3 presented cases of PCSA conditioned by tone or stress. A general 

observation about these cases is that very few examples were uncovered by the survey. 

Nonetheless, the existence of some examples does demonstrate that tone and stress can 

condition PCSA. As with the examples in chapter 2, conditioning by stress or tone is 

always at the edge of the stem at which the allomorphy occurs. Another interesting 

finding in chapter 3 is that there seem to exist some instances of stem PCSA (§3.1.1.2), 

though they are few. These examples were reconciled with the generalization that 

conditioning comes from the ‘inside’, so the generalization can still be maintained. An 

example from Spanish was discussed (§3.1.1.1) in which the conditions on allomorph 

distribution are opaque on the surface, just as in the examples from chapter 2 mentioned 

above. Finally, some examples of tone or stress-conditioned allomorphy were argued to 

be non-optimizing (§3.1.1.4), since no well-motivated phonological constraint could be 

identified as possibly being responsible for the allomorphy. 

In chapter 4, I discussed examples of PCSA conditioned by prosodic elements, 

namely, moras, syllables, and feet. Many examples were found, and the conditions in 

these examples tended to be overall properties of words rather than individual elements at 

the edge of the stem. Still, the locality generalization discussed above is still maintained 

259

in these examples. Overall properties of the stem count as properties at the edge, since no 

material intervenes between the conditioning factor (e.g., all of the syllables of the stem) 

and the affix undergoing allomorphy. As in chapters 2 and 3, the examples in chapter 4 

involve conditioning by elements closer to the root; no instances of stem allomorphy 

were found here. Some examples were discussed in which, as in chapters 2 and 3, 

conditions on PCSA are made opaque due to regular phonological processes (see, for 

instance, the Ngiyambaa example in §4.2.1). There are also several examples of PCSA in 

chapter 4 in which the allomorphy appears to be non-optimizing (§4.1.1.3). It is 

acknowledged that, relative to chapters 2 and 3, a higher percentage of the examples in 

chapter 4 appear to have some optimizing character. For this reason, in chapter 4 I 

investigated in more detail a large set of related examples from the Pama-Nyungan 

languages of Australia involving allomorphy in the ergative suffix (§4.2). Several of 

these languages exhibit allomorphy that appears to optimize words, but many do not; in 

this discussion I showed how a historical understanding of such examples can explain the 

apparent optimization effects while allowing for the non-optimizing examples. 

I have summarized the cross-linguistic findings for PCSA that were presented in 

chapters 2-4. In the following section, I compare these findings with the predictions 

discussed in §6.1 and discuss the theoretical implications of this comparison. 

6.3 Summary of theoretical implications 

In §6.1, I summarized the predictions for PCSA that are made by P >> M and 

subcategorization. Then, in §6.2, I summarized the results of the cross-linguistic survey 

of PCSA. In this section, I compare how well the predictions in §6.1 correspond to the 

260

cross-linguistic generalizations for PCSA in §6.2, demonstrating that the 

subcategorization approach fares better than the P >> M approach in this comparison. I 

conclude this section by summarizing the theoretical implications of the comparison, and 

also the implications of the findings discussed in chapter 5 for other types of 

phonological conditions on affixation. 

First, regarding the predictions for PCSA, it appears that the subcategorization 

model is a better predictor of the attested examples, based on the present survey. This can 

be assessed in terms of the four major predictions made by each model. The first of these 

deals with phonological optimization. The P >> M approach, since it uses regular 

phonological constraints, predicts that every case of PCSA should be phonologically 

optimizing with respect to an established phonological constraint (prediction (1)a in 

§6.1); the subcategorization approach predicts the existence of non-optimizing examples 

(prediction (2)a). As discussed above, many examples found in this survey appear to be 

non-optimizing. While it is difficult to prove that a particular example is non-optimizing, 

the large number of cases of PCSA for which no phonological constraint is readily 

available (see examples in §2.1.2.5, §3.1.1.4, and §4.1.1.3) to handle the allomorphy 

suggests that the prediction of the subcategorization model is correct. 

The second prediction of each model refers to input/underlying forms vs. surface 

forms. P >> M predicts that PCSA should refer to surface phonological properties of 

words (prediction (1)b), while subcategorization predicts that there should be instances 

where PCSA is clearly sensitive to input or underlying forms, rather than surface forms 

(prediction (2)b). As discussed above, several examples are found in which the conditions 

on PCSA are rendered opaque by the operation of regular phonological processes (see 

261

especially §2.1.2.6). This upholds the prediction of the subcategorization model and is 

problematic for P >> M, since in these examples, PCSA is crucially sensitive to input 

phonological elements, and not to surface elements. 

A third prediction of each model concerns conditioning from the ‘inside’ vs. the 

‘outside’. In a standard P >> M analysis, morphological constituency is not reflected in 

the input form, and therefore each morpheme in the word is equally likely to trigger 

allomorphy in other morphemes. This means that in P >> M, an affix can condition 

PCSA in another affix closer to the root, or even in the root itself (prediction (1)c). The P 

>> M model thus predicts the existence of conditioning from both the ‘inside’ and the 

‘outside’. The subcategorization approach, on the other hand, is more restrictive in this 

regard. As discussed above, subcategorization as implemented within a model of 

morphology such as Lexical Morphology (which is standard for subcategorization 

analyses) predicts the existence only of conditioning from the ‘inside’, never from the 

‘outside’ (prediction (2)c). Based on the survey results throughout chapters 2-4, it appears 

that this more restrictive prediction of subcategorization is upheld cross-linguistically. 

There are no clear-cut cases of conditioning from the ‘outside’. 

Finally, both models make predictions for the location of the condition on 

affixation vis à vis the location of the affix undergoing allomorphy. The P >> M approach 

predicts that the condition may be located anywhere in the word, regardless of the 

location of the affix (prediction (1)d). The subcategorization approach, on the other hand, 

predicts that conditions on PCSA are limited to elements immediately adjacent to the 

262

affix in question (prediction (2)d). 1 The survey data appear to uphold the predictions of 

the subcategorization model, not the P >> M model. Based on this survey, conditions on 

PCSA are always at the same edge of the stem where the affix allomorphs occur; this is 

most apparent in chapter 2 (§2.1.2). There are no examples of the type predicted by P >> 

M in which, for example, prefix allomorphy is conditioned by the stem-final segment. 

Therefore, once again, the survey results point to subcategorization as the superior model. 

We have seen that the results for PCSA overwhelmingly support the 

subcategorization approach. In addition, results for other types of phonology-morphology 

interactions appear to favor this approach as well, converging with the results reported 

here. In chapter 5, I covered some other types of phonological conditions on affixation, 

laying out the predictions made by P >> M for each area and discussing, where available, 

the cross-linguistic results for each. The following five logically possible types of 

phonological conditions on affixation (aside from PCSA) were discussed: empty morphs 

(§5.1), phonologically induced morphological gaps (§5.2), reduplication (§5.3), 

phonologically conditioned affix order (§5.4), and infixation (§5.5). For infixation, there 

does exist a large cross-linguistic survey of the phenomenon along with a critique of the 

P >> M approach (Yu 2003). The results in that area converge with the results presented 

here, since P >> M predicts the existence of some unattested effects in affixation, and it 

also fails to predict some effects that are attested. Thus, Yu (2003) argues against the use 

1 

As with the third prediction (regarding the direction of conditioning), this prediction is 

made not by the subcategorization mechanism itself, but by a standard implementation of 

the model. In this case, the implementation would need to incorporate the Generalized 

Determinant Focus Adjacency Condition (GDFAC; Inkelas 1990), discussed in chapters 

1 and 2. As pointed out in chapter 1, the GDFAC is uniquely compatible with the 

subcategorization approach, not the P >> M approach, so from this perspective the 

adjacency prediction is, at least indirectly, a prediction of the subcategorization model 

itself. 

263

of P >> M for modeling infixation, as I have done here for PCSA. Large surveys for the 

purpose of assessing the claims of P >> M have not been conducted in the other areas 

discussed in chapter 5, but I have pointed out claims made by P >> M in each area, and 

have pointed out some unusual and likely false predictions that are made in some cases. 

Therefore, the available data on types of phonological conditions on affixation other than 

PCSA and infixation lend tentative support to subcategorization, though more work is 

needed. 

In this section I have pointed out how subcategorization makes more accurate 

predictions than P >> M for PCSA, and how this is true for other types of phonological 

conditions on affixation for which data are available. The overall result is that the P >> M 

model is not upheld and should be abandoned in favor of the subcategorization approach. 

Thus, cross-linguistic research has given us insight into the best way to model 

phonological effects in morphology. 

6.4 Two types of PCSA? 

Although the cross-linguistic results strongly favor the subcategorization 

approach, one may still find problematic the model’s inability to capture apparent 

optimization effects of PCSA. Some of the apparently optimizing examples discussed in 

the literature are not clear-cut cases of such (e.g. the Dyirbal example claimed by 

McCarthy and Prince (1990) to involve complementarity when in fact no tenable P >> M 

analysis has been able to make use of this insight), and in chapter 5 I discussed at length 

how an understanding of the historical origins of PCSA patterns minimizes the need for 

synchronic explanations of apparent optimization. However, one may wonder whether a 

264

hybrid model could be applied, in which the cases that seem to be optimizing are 

modeled via P >> M, while subcategorization frames are used for the other, non- 

optimizing cases. As discussed for syllable-counting allomorphy in §4.3.8, a proposal 

similar to this was made by Booij (1998), and this also corresponds with Mascaró’s 

(1996) distinction between ‘internal’ and ‘external’ allomorphy. 

There are a number of serious problems with this strategy. First, even the 

‘optimizing’ examples still appear to fulfill predictions (b-d) of the subcategorization 

approach more closely than those of the P >> M approach: they are never demonstrably 

output-based, they are conditioned from the ‘inside’ and not the ‘outside’, and they never 

violate the adjacency requirements of the GDFAC. Thus, even those examples for which 

we would give a P >> M analysis under the dual approach still fail to fulfill three other 

predictions of the P >> M approach that distinguish it from the subcategorization model. 

A second problem has to do with how we might establish the split between 

optimizing and non-optimizing cases to determine which model to use for a given 

example. Optimization vs. non-optimization is not a black-and-white distinction, but 

rather a continuum, as was discussed for syllable-counting allomorphy in §4.3.8. 

The continuum is problematic because it is not clear where optimization ends and non- 

optimization begins. In order to analyze the optimizing examples in one way and the non- 

optimizing examples in a different way, we need a clear dividing line between the two 

types. 

A related problem is that there is no independent basis for two different types of 

PCSA. The split would therefore have to be made based on which examples lend 

themselves to analysis in P >> M and which do not. This is circular because the claim 

265

that there are two types of allomorphy that should be analyzed differently would rest 

upon the fact that we analyzed the two types differently. A useful comparison can be 

made between this situation and the situation faced by Alderete et al (1999) in their study 

of fixed segment reduplication (FSR). As was discussed in §4.3.8, Alderete et al 

proposed that there are two types of FSR, a phonological type and a morphological type; 

the former follows phonological principles and is analyzed phonologically, while the 

latter does not follow phonological principles (e.g., the identity of the fixed segment in 

morphological FSR is an arbitrary segment, not a phonologically unmarked ‘default’ 

segment) and is analyzed morphologically. Unlike in our present study of PCSA, 

Alderete et al (1999: 355-356) identified several independent differences between their 

proposed phonological and morphological types of FSR. We have no such independent 

criteria to distinguish two types of PCSA, and the burden of proof is on those who would 

argue that PCSA can be split into multiple separate phenomena in a principled way. The 

survey presented in this dissertation represents an attempt to find independent criteria on 

which such a distinction might be made, but none were found. 

(footnote 2): 

Bonet et al (in press) advocate a dual approach based on the following argument 

‘The idea that all allomorphy should be explained by the same mechanism 

seems to be assumed by some authors, either implicitly or explicitly (for 

instance, Paster, 2005[a]: section 5, states that subcategorization “avoids 

the problem of having multiple theoretical mechanisms to model a single 

phenomenon”). But ‘allomorphy’ is a (vague) descriptive concept that has 

no privileged theoretical status.’ 

Two points must be made regarding this line of reasoning. First, it must be clarified that 

while some authors may indeed claim that ‘all allomorphy should be explained by the 

same mechanism,’ Paster 2005a argues only that syllable-counting allomorphy, a specific 

266

subtype of PCSA (itself a subtype of ‘allomorphy’) should have a unified analysis; the 

argument is extended in this dissertation to all of PCSA but not to ‘allomorphy’ in 

general. In fact, several specific independent criteria for distinguishing PCSA from non- 

suppletive phonologically conditioned allomorphy are established in §2.1.1. Thus, the 

claim that ‘allomorphy’ has no theoretical status is immaterial to the arguments made in 

Paster 2005a and in this dissertation. Second, the assertion that allomorphy (or more 

specifically, PCSA, if this is what was actually intended in the quote above) is merely a 

‘(vague) descriptive concept’ is easily turned back on Bonet et al, who rely crucially on 

Mascaró’s (1996) distinction between internal and external allomorphy (renamed by 

Bonet et al as ‘arbitrary phonologically conditioned allomorphy’ and ‘regular (natural) 

phonologically conditioned allomorphy’, respectively); the arbitrary ‘type’ of allomorphy 

apparently consists of cases in which ‘[i]t is difficult to see any natural phonological 

connection between the shape of the allomorphs’ (Bonet et al in press: §1). Although 

naturalness is one factor that I have used to determine what constitutes PCSA vs. non- 

suppletive allomorphy (see §2.1.1), this is not the only criterion used, whereas 

naturalness (defined in terms of difficulty for the analyst) does seem to be the sole 

difference between Bonet et al’s two types of PCSA. Classifying a particular case as 

optimizing or non-optimizing allomorphy based solely on the ease of analysis in P >> M 

vs. subcategorization is circular and not insightful. 

A final problem with the dual approach has to do with how to formalize it. Even if 

we decide that there are two types of PCSA that should be analyzed differently, how can 

we ensure that the grammar will ‘do’ the optimizing type using P >> M and the non- 

optimizing type using subcategorization? A possible way to encode the distinction in the 

267

grammar would be to put limits on subcategorization frames so that affixes cannot 

subcategorize for the elements that commonly condition the optimizing type of PCSA. I 

observed earlier that PCSA conditioned by prosodic units such as feet, syllables, and 

moras, as well as the C/V distinction, account for most of the optimizing examples, while 

PCSA conditioned by specific segments and/or features are more commonly non- 

optimizing. Suppose, then, that subcategorization frames were limited so that an affix 

could only subcategorize for small units like segments and features, and not for larger 

prosodic units or C/V. Then suppose that the grammar were set up so that it would ‘try’ a 

subcategorization-based generalization first, and failing this, would then default to the P 

>> M generalization. This would achieve the effect of having the grammar (rather than 

the analyst) distinguish the two types of PCSA, but there is a major problem with this 

move. The problem is that it directly contradicts the set of things that have previously 

been established as elements that affixes can subcategorize for. In his study of infixation, 

Yu (2003, in press) found that the set of phonological elements that can be 

morphologically subcategorized for is exactly: {foot, syllable, mora, C, V}. Thus, 

preventing affixes from subcategorizing for these elements would have unintended 

negative consequences for other types of phonologically conditioned morphology. 

The bottom line is that while the subcategorization model can handle both the 

examples that seem to optimize and those that do not, the P >> M model can handle only 

the optimizing examples well. And with no independent evidence to distinguish two 

separate types of PCSA, the subcategorization model is the one that should be used. 

Assuming that one accepts these arguments for the subcategorization model, one 

may reasonably ask whether this means that the apparent optimization found in so many 

268

examples in the survey is merely a coincidence. The answer is no, not necessarily. As I 

mentioned in §2.2, one way in which PCSA can arise is from a phonological process that 

is lost and becomes morphologized. If the original phonological process was optimizing 

(and it is widely held that phonological processes often do have this characteristic), then 

the resulting PCSA can retain the appearance of optimization without this having to be 

encoded in the synchronic grammar. Therefore, although the subcategorization approach 

does not directly capture the optimizing effects that are apparent in some examples in the 

survey, this does not mean that we cannot explain them. 

6.5 M >> P vs. separate components 

If P >> M is rejected as a way of modeling phonological conditions on affixation, 

how might we rule out this ranking schema? There are at least two possibilities. One, 

proposed by Yu (2003: 108) is to propose a universal ranking M >> P. 2 Alternatively, P 

and M may be distinguished as separate components of grammar that are not evaluated in 

parallel and therefore have no ranking relationship. Both of these options are sufficient to 

rule out P >> M, but they make different predictions, which I will discuss here. 

The M >> P proposal makes at least two important claims. The first is that PCSA 

is surface-based, since if M and P have a ranking relationship, they will be evaluated in 

parallel just as in P >> M. The second is that P constraints will drive morphology 

whenever M constraints underdetermine outputs. 

The ‘separate components’ proposal, on the other hand, makes opposing claims. 

In this approach, assuming that the output of morphology is the input to phonology, 

2 

Note, however, that Yu in press, which is based on Yu 2003, does not pursue this 

option. 

269

PCSA should be an input-based phenomenon. A second claim is that P constraints cannot 

drive morphological processes; instead their effects should be seen on the outputs of 

morphology only. 

There is some evidence suggesting that the ‘separate components’ approach is the 

correct one. First, as we have seen above, PCSA is crucially an input-based phenomenon, 

not output-based. This is demonstrated in the Turkish example discussed in §2.1.2.6. 

Thus, a prediction of the M >> P proposal is contradicted by PCSA. The second type of 

evidence against M >> P comes from affix ordering. Assuming that affix order is handled 

by M constraints (an assumption that is explicitly made in some treatments of putative 

phonologically conditioned affix order, e.g. Hargus and Tuttle 1997), then under M >> P, 

we expect that anytime the M constraints do not fully determine affix order, P constraints 

should be able to step in and select a winner. However, as I discuss in Paster 2006, no 

good cases of phonologically driven affix order are attested. Furthermore, there are at 

least two attested examples of free affix order (in Chintang (Bickel et al in press) and in 

Filomeno Mata Totonac (McFarland 2007)). Under M >> P, we would expect that if the 

morphology left the affix order free, then the lower-ranked P constraints should come 

into play, as in TETU, and select a winner. Free affix order should not exist because the P 

constraints should always pick a consistent winner. Thus, the existence of free affix 

order, in combination with the other observations discussed here, favors the ‘separate 

components’ approach. 

I have elaborated on some aspects of the argument in favor of subcategorization 

and against P >> M, arguing in §6.4 that a hybrid model in terms of ‘two types of PCSA’ 

is untenable and in this section that P and M should be understood as separate 

270

components of grammar rather than having any ranking relation in OT, even one that 

inherently rules out P >> M. In the following section, I suggest a few ways in which the 

present line of research may be advanced. 

6.6 Future directions 

This dissertation contributes to our understanding of the phonology-morphology 

interface by providing a broad empirical basis for the study of PCSA, an important 

instance of a phonological condition on affixation. The generalizations about PCSA that 

have been presented here are based on a large database of examples; previous studies 

have not had a very large set of examples on which to base generalizations. Because of 

this, the generalizations presented here and the theoretical conclusions that follow from 

them are perhaps more significant than those that have been made before. Thus, the 

dissertation makes a theoretical contribution as well as an empirical one. 

The work presented here gives insight into just one part of an overall picture of 

phonological effects in morphology, and the phonology-morphology interface more 

generally. As discussed in chapter 1, the study of phonologically conditioned morphology 

began in earnest relatively recently, and much work in this area remains to be done. 

Regarding the study of PCSA itself, some facts have been left unexplained here. 

One of these, pointed out in chapter 3, is that the survey revealed only a very small 

number of examples of PCSA conditioned by tone or stress. Perhaps this is a gap in the 

survey, or perhaps it reflects some deeper, yet unknown, property of language. This issue 

might fruitfully be probed through careful research on a particular family of languages 

with well documented stress or tone properties (e.g., the Bantu languages of Africa) in 

271

order to find out whether regular phonological processes involving stress and tone can 

evolve into patterns of stress- or tone-conditioned PCSA, and if not, why not. A second 

issue is the appearance of optimization in many examples discussed in the presentation of 

survey results, perhaps more than expected by chance under the subcategorization 

approach. It can still be maintained that the existence of apparent non-optimizing 

examples is sufficient to argue against P >> M, since that model predicts that such 

examples should not exist. However, in the subcategorization model, we do not have a 

ready explanation for the apparent fact that many examples of PCSA (perhaps the 

majority) can be plausibly argued to be optimizing in some way. The detailed look at 

examples of ergative allomorphy in the Pama-Nyungan languages in chapter 4 was 

intended to suggest a general historical approach to optimizing PCSA that complements 

the subcategorization model. This approach could be tested on other examples to see 

whether other apparently optimizing examples can be accounted for successfully in the 

historical realm. Perhaps the apparent optimizing nature of phonological processes is 

sufficient to explain optimization in PCSA, since it seems likely that many cases of 

PCSA evolved from regular phonological processes that were lost. However, analogy 

seems to play a major role as well, and it remains an open question whether analogy can 

involve some type of optimization, even if PCSA itself is not optimizing synchronically. 

In addition, instances of phonological conditions on affixation other than PCSA 

were discussed in chapter 5, and some have not yet been well studied from a cross- 

linguistic perspective. In particular, empty morphs and phonologically induced 

morphological gaps are not well understood. Future research will show whether the 

conclusions regarding the best way to model PCSA hold up for those phenomena as well. 

272

Also, I have focused here only on phonological conditions on affixation, but other 

morphological processes, such as compounding, should be considered as well. Further 

testing based on larger and better cross-linguistic studies will reveal the extent to which 

the claims made in this dissertation can be accepted as general principles of morphology. 

273

References 

Abondolo, Daniel M. 1988. Hungarian Inflectional Morphology. Budapest: Akadémiai 

Kiadó. 

Alderete, John. 2001. Dominance effects as transderivational anti-faithfulness. Phonology 

18.2: 201-253. 

Alderete, John, Jill Beckman, Laura Benua, Amalia Gnanadesikan, John McCarthy, and 

Suzanne Urbanczyk. 1999. Reduplication with fixed segmentism. Linguistic Inquiry 

30.3: 327-364. 

Anderson, Stephen R. 1981. Why phonology isn’t ‘natural’. Linguistic Inquiry 12: 493- 

539. 

Andonian, Hagop. 1999. Beginner’s Armenian. New York: Hippocrene Books. 

Aranovich, Raúl, Sharon Inkelas, Orhan Orgun, and Ronald Sprouse. 2005. Opacity in 

phonologically conditioned suppletion. Paper presented at the Thirteenth Manchester 

Phonology Meeting. 

Arnott, D.W. 1970. The Nominal and Verbal Systems of Fula. Oxford: Clarendon Press. 

Austin, Peter. 1981. A Grammar of Diyari, South Australia. Cambridge: Cambridge 

University Press. 

Bach, Emmon and Robert T. Harms. 1972. How do languages get crazy rules?. Pp. 1-21 

in Robert P. Stockwell and Ronald K.S. Macaulay, eds. Linguistic Change in 

Generative Theory. Bloomington: Indiana University Press. 

Baker, Mark. 1985. The mirror principle and morphosyntactic explanation. Linguistic 

Inquiry 16: 373-415. 

Bastin, Yvonne. 1983. La finale -IDE et l’imbrication en bantou. Tervuren: Annales du 

Musée Royale de l’Afrique Centrale. Série IN-8, Sciences Humaines. N. 114. 

Bauer, Laurie. 1988. Introducing Linguistic Morphology. Edinburgh: Edinburgh 


Beckman, Jill. 1998. Positional Faithfulness. PhD. dissertation, University of 

Massachusetts, Amherst. 

274

Bickel, Balthasar, Goma Banjade, Martin Gaenszle, Elena Lieven, Netra Paudyal, 

Ichchha Purna Rai, Manoj Rai, Novel Kishor Rai, and Sabine Stoll. In press. Free 

prefix ordering in Chintang. Language 83. 

Biggs, Bruce. 1961. The structure of New Zealand Maaori. Anthropological Linguistics 

3: 1-54. 

Blevins, Juliette. 2001. Nhanda: An Aboriginal Language of Western Australia. 

Honolulu: University of Hawai’i Press. 

Blevins, Juliette. 2004. Evolutionary Phonology. Cambridge: Cambridge University 

Press. 

Bonet, Eulàlia. 2004. Morph insertion and allomorphy in Optimality Theory. 

International Journal of English Studies 4.2: 73-104. 

Bonet, Eulàlia, Maria-Rosa Lloret, and Joan Mascaró. In press. Allomorph selection and 

lexical preferences: Two case studies. Lingua. 

Booij, Geert. 1997. Non-derivational phonology meets lexical phonology. Pp. 261-288 in 

Iggy Roca, ed. Derivations and Constraints in Phonology. Oxford: Clarendon Press. 

Booij, Geert. 1998. <strong>Phonological</strong> output constraints in morphology. Pp. 143-163 in 

Wolfgang Kehrein and Richard Wiese, eds. Phonology and Morphology of the 

Germanic Languages. Tübingen: Niemeyer. 

Booij, Geert. 2005. The Grammar of Words. Oxford: Oxford University Press. 

Booij, Geert and Rochelle Lieber. 1993. On the simultaneity of morphological and 

prosodic structure. Pp. 23-44 in Sharon Hargus and Ellen M. Kaisse, eds. 

Phonetics and Phonology 4: Studies in Lexical Phonology. San Diego: Academic 

Press, Inc. 

Breen, J.G. 1976a. Bidjara. P. 339 in R.M.W. Dixon, ed. Grammatical Categories in 

Australian Languages. Canberra: Australian Institute of Aboriginal Studies. 

Breen, J.G. 1976b. Wangkumara. Pp. 336-339 in R.M.W. Dixon, ed. Grammatical 

Categories in Australian Languages. Canberra: Australian Institute of Aboriginal 

Studies. 

Breen, J.G. 1976c. Warluwara and Bularnu. Pp. 331-335 in R.M.W. Dixon, ed. 

Grammatical Categories in Australian Languages. Canberra: Australian Institute of 

Aboriginal Studies. 

275

Broadwell, George Aaron. 2005. The morphology of Zapotec pronominal clitics. Pp. 15- 

35 in Rosemary Beam de Azcona and Mary Paster, eds. Proceedings of the 

Conference on Otomanguean and other Oaxacan Languages. Berkeley: Survey of 

California and Other Indian Languages. 

Buckley, Eugene. 1994. Theoretical Aspects of Kashaya Phonology and Morphology. 

Stanford: CSLI. 

Buechel, Eugene. 1939. A Grammar of Lakota. St. Louis, Missouri: John S. Swift 

Company, Inc. 

Bybee, Joan. 1985. Morphology: A Study of the Relation Between Meaning and Form. 

Amsterdam: Benjamins. 

Bye, Patrik. 2005. Allomorphy – Selection, not optimization. Paper presented at the 

University of Chicago linguistics department colloquium. 

Bye, Patrik. To appear. Allomorphy – Selection, not optimization. Sylvia Blaho, Patrik 

Bye, and Martin Krämer, eds. Freedom of Analysis? Berlin: Mouton de Gruyter. 

Cadely, John-Robert. 2002. Le statut des voyelles nasals en Créole haïtien. Lingua 112: 

435-464. 

Carstairs, Andrew. 1988. Some implications of phonologically conditioned suppletion. 

Yearbook of Morphology 1988: 67-94. 

Carstairs, Andrew. 1990. <strong>Phonological</strong>ly conditioned suppletion. Pp. 17-23 in Wolfgang 

U. Dressler, Hans C. Luschützky, Oskar E. Pfeiffer, and John R. Rennison, eds. 

Contemporary Morphology. New York: Mouton de Gruyter. 

Carstairs-McCarthy, Andrew. 1998. <strong>Phonological</strong> constraints on morphological rules. Pp. 

144-148 in Andrew Spencer and Arnold Zwicky, eds. The Handbook of Morphology. 

Oxford: Blackwell. 

Cohen, Kevin Bretonnel. 2000. Aspects of the Grammar of Kukú. München: Lincom 

Europa. 

Conathan, Lisa J. 2002. Split intransitivity and possession in Chimariko. Pp. 18-31 in 

Lisa Conathan and Teresa McFarland, eds. Survey report #12: Proceedings of the 

50th Anniversary Conference of the Survey of California and Other Indian 

Languages. Berkeley: Survey of California and Other Indian Languages. 

Condoravdi, Cleo and Paul Kiparsky. 1998. Optimal order and scope. Paper presented at 

Lexicon in Focus, Heinrich-Heine Universität Düsseldorf. 

276

Crosswhite, Katherine. 1999. Intra-paradigmatic homophony avoidance in two dialects of 

Slavic. Pp. 48-67 in Matthew K. Gordon, ed. UCLA Working Papers in Linguistics 1: 

Papers in Phonology 2. Los Angeles: UCLA Department of Linguistics. 

Davis, Stuart. 1998. Syllable contact in Optimality Theory. Korean Journal of Linguistics 

23: 181-211. 

de Maria, Jose. 1918. Gramatica y vocabulario jibaros. Quito: Impr. de la Universidad 

Central. 

Dench, Alan. 1995. Martuthunira: A Language of the Pilbara Region of Western 

Australia. Canberra: Pacific Linguistics, Series C-125. 

Dench, Alan. 1998. Yingkarta. München: LINCOM Europa. 

DiFabio, Elvira G. 1990. The Morphology of the Verbal Infix isc in Italian and Romance. 

PhD dissertation, Harvard University. 

Dimmendaal, Gerrit J. 1983. The Turkana Language. Dordrecht: Foris. 

Dimmendaal, Gerrit J. 2000. Morphology. Pp. 161-193 in Bernd Heine and Derek Nurse, 

eds. African Languages: An Introduction. 

Dixon, Roland B. 1910. The Chimariko Indians and language. University of California 

Publications in American Archaeology and Ethnology 5.5. Berkeley: University of 

California Press. 

Dixon, R.M.W. 1972. The Dyirbal Language of North Queensland. Cambridge: 

Cambridge University Press. 

Dixon, R.M.W. 1977. A Grammar of Yidiɲ. Cambridge: Cambridge University Press. 

Dixon, R.M.W. 1980. The Languages of Australia. Cambridge: Cambridge University 

Press. 

Dixon, R.M.W. 2002. Australian Languages: Their Nature and Development. 

Cambridge: Cambridge University Press. 

Dolbey, Andrew. 1997. Output optimization and cyclic allomorph selection. Proceedings 

of the Fifteenth West Coast Conference on Formal Linguistics: 97-112. Stanford: 

CSLI. 

Donaldson, Tamsin. 1980. Ngiyambaa: The Language of the Wangaaybuwan. 


277

Dresher, Elan B. and Xi Zhang. In press. Contrast in Manchu vowel systems. In Carsten 

Naher, ed. Proceedings of the First International Conference on Manchu-Tungus 

Studies. Volume 2: Trends in Tungusic and Siberian Linguistics. Wiesbaden, 

Germany: Harrassowitz. 

Elías-Ulloa, José. 2004. Variable syllable weight and quantity-insensitive allomorphy in 

Shipibo. Paper presented at the 35th meeting of the North Eastern Linguistic Society, 

University of Connecticut. 

Fast Mowitz, Gerhard, Ruby Warkentin de Fast, and Daniel Fast Warkentin. 1996. 

Diccionario Achuar-Shiwiar – Castellano. Serie lingüistica Peruana 36. 

Yarinacocha, Pucallpa, Peru: Instituto Lingüístico de Verano. 

Feldman, Harry. 1986. A Grammar of Awtuw. Pacific Linguistics Series B, number 94. 

Canberra: Australian National University. 

Fulmer, Sandra. 1991. Dual-position affixes in Afar: An argument for phonologicallydriven 

morphology. Pp. 189-203 in Aaron Halpern, ed. Proceedings of the Ninth West 

Coast Conference on Formal Linguistics. Stanford: CSLI. 

Good, Jeff. 2003. Strong Linearity: Three Case Studies Towards a Theory of 

Morphosyntactic Templatic Constructions. PhD. dissertation, UC Berkeley. 

Gordon, Raymond G., Jr., ed. 2005. Ethnologue: Languages of the World. Fifteenth 

edition. Dallas, Tex.: SIL International. Online version: http://www.ethnologue.com. 

Gouskova, Maria. 2004. Relational hierarchies in Optimality Theory: The case of syllable 

contact. Phonology 21.2: 201-250. 

Haig, Geoffrey. To appear. Constraints on morpheme repetition in Turkish?. In 

Proceedings of the Eleventh International Conference on Turkish Linguistics. 

Hale, Kenneth. 1976a. Dja:bugay. Pp. 321-326 in R.M.W. Dixon, ed. Grammatical 


Studies. 

Hale, Kenneth. 1976b. On ergative and locative suffixal alternations in Australian 

languages. Pp. 414-417 in R.M.W. Dixon, ed. Grammatical Categories in Australian 

Languages. Canberra: Australian Institute of Aboriginal Studies. 

Hale, K. and A. Lacayo Blanco. 1989. Diccionario elemental del Ulwa (Sumu 

Meridional). Cambridge: Center for Cognitive Science, MIT. 

278

Hall, Beatrice L. and Eluzai Moga Yokwe. 1981. Bari vowel harmony: The evolution of a 

cross-height vowel harmony system. Pp. 55-63 in Eluzai Moga Yokwe and Wanda 

Pace, eds. Occasional Papers in the Study of Sudanese Languages 1. Juba, Sudan: 

University of Juba Printing Unit. 

Hall, Robert A. 1948. Descriptive Italian Grammar. Ithaca, New York: Cornell 

University Press and Linguistic Society of America. 

Hall, Robert A. 1953. Haitian Creole: Grammar, Texts, Vocabulary. Memoirs of the 

American Anthropological Association 74. Menasha, Wisconsin: American 

Anthropological Association. 

Hansson, Gunnar Ólafur. 2001. Theoretical and Typological Issues in Consonant 

Harmony. PhD. dissertation, UC Berkeley. 

Hargus, Sharon and Siri G. Tuttle. 1997. Augmentation as affixation in Athabaskan 

languages. Phonology 14: 177-220. 

Harlow, Ray. 1996. Māori. München: LINCOM Europa. 

Harrell, Richard S. 1962. A Short Reference Grammar of Moroccan Arabic. Washington, 

DC: Georgetown University Press. 

Harris, James W. 1979. Some observations on ‘Substantive principles in natural 

generative phonology’. Pp. 281-293 in Daniel A. Dinnsen, ed. Current Approaches to 

<strong>Phonological</strong> Theory. Bloomington, Indiana: Indiana University Press. 

Harrison, S.P. 1973. Reduplication in Micronesian languages. Oceanic Linguistics 7: 

407-454. 

Hartell, Rhonda L., ed. 1993. Alphabets of Africa. Dakar: UNESCO/SIL. 

Hayward, Richard J. 1998. Qafar (East Cushitic). Pp. 624-647 in Andrew Spencer and 

Arnold M. Zwicky, eds. The Handbook of Morphology. Oxford: Blackwell Publishers 

Ltd. 

Hendricks, Sean Q. 1999. Reduplication Without Template Constraints: A Study in Bare- 

Consonant Reduplication. PhD. dissertation, University of Arizona. 

Hyman, Larry M. 1995. Minimality and the prosodic morphology of Cibemba 

imbrication. Journal of African Languages and Linguistics 16: 3-39. 

Hyman, Larry M. 2003. Suffix ordering in Bantu: A morphocentric approach. Yearbook 

of Morphology 2002: 245-281. 

279

Hyman, Larry M. and Sharon Inkelas. 1997. Emergent templates: The unusual case of 

Tiene. Pp. 92-116 in Viola Miglio and Bruce Morén, eds. Selected Phonology 

Papers from H-OT-97. University of Maryland Working Papers in Linguistics. 

College Park: University of Maryland Department of Linguistics. 

Inkelas, Sharon. 1990. Prosodic Constituency in the Lexicon. New York: Garland. 

Inkelas, Sharon. 1993. Nimboran position class morphology. Natural Language and 

Linguistic Theory 11: 559-624. 

Inkelas, Sharon. 1998. The theoretical status of morphologically conditioned phonology: 

A case study of dominance effects. Yearbook of Morphology 1997: 121-155. 

Inkelas, Sharon and Cheryl Zoll. 2005. Reduplication: Doubling in Morphology. 


Inkelas, Sharon, Orhan Orgun, and Cheryl Zoll. 1997. The implications of lexical 

exceptions for the nature of grammar. Pp. 393-418 in Iggy Roca, ed. Derivations 

and Constraints in Phonology. Oxford: Clarendon Press. 

Jacobsen, William H. 1973. A rhythmic principle in Washo morphotactics. Paper 

presented at the Symposium on California Indian Linguistics, Southwestern 

Anthropological Association, San Francisco, California. 

Jespersen, Otto. 1904. Lehrbuch der Phonetik. Leipzig and Berlin: Teubner. 

Jintia, Yánkuam, P. Luis Bolla Sartori, and Péas Kantásh Ishtiku. 2000. Achuar 

ujákratmau: Diccionario Achuar. Achuar matsátmau 3. Lima: Centro Amazónico de 

Antropología y Aplicación Práctica. 

Johnston, Raymond Leslie. 1980. Nakanai of New Britain: The Grammar of an Oceanic 

Language. Canberra: Australian National University. 

Kager, Rene. 1996. On affix allomorphy and syllable counting. Pp. 155-171 in Ursula 

Kleinhenz, ed. Interfaces in Phonology. Berlin: Akademie Verlag. 

Karlsson, Fred. 1999. Finnish: An Essential Grammar. London and New York: 

Routledge. 

Kaufman, Terrence. 1971. Tzeltal Phonology and Morphology. University of California 

Publications in Linguistics 61. Berkeley and Los Angeles: University of California 

Press. 

Kenesei, István, Robert Vago, and Anna Fenyvesi. 1997. Hungarian. New York: 

Routledge. 

280

Kennedy, Robert. 2002. Stress and allomorphy in Woleaian reduplication. In Proceedings 

of the Texas Linguistics Society Conference. Austin, Texas: University of Texas. 

Kikuchi, Seiichiro. 2000. Spanish definite article allomorphy: A correspondence 

approach. Paper presented at the Sixth Meeting of the <strong>Phonological</strong> Society of 

Japan, Chiba University. 

Kikuchi, Seiichiro. 2001. Spanish definite article allomorphy: A correspondence 

approach. <strong>Phonological</strong> Studies 4: 49-56. Tokyo: Kaitakusha. 

Kiparsky, Paul. 1982a. Lexical Morphology and Phonology. Pp. 3-91 in I.-S. Yange, ed. 

Linguistics in the Morning Calm. Seoul: Hanshin. 

Kiparsky, Paul. 1982b. Word-formation and the lexicon. Pp. 1-29 in Frances Ingemann, 

ed. 1982 Mid-America Linguistics Conference Papers. Lawrence, Kansas: 

Department of Linguistics, University of Kansas. 

Kiparsky, Paul. 1996. Allomorphy or morphophonology?. Pp. 13-31 in Rajendra Singh, 

ed. Trubetzkoy’s Orphan: Proceedings of the Montreal Roundtable 

‘Morphophonology: Contemporary Responses’. Amsterdam: Benjamins. 

Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17: 351-366. 

Kite, Suzanne and Stephen Wurm. 2004. The Duuŋidjawu Language of Southeast 

Queensland. Canberra: Pacific Linguistics. 

Klein, Thomas B. 2003. Syllable structure and lexical markedness in creole 

morphophonology: Determiner allomorphy in Haitian and elsewhere. Pp. 209-228 

in Ingo Plag, ed. The Phonology and Morphology of Creole Languages. 

Tübingen: Max Niemeyer. 

Koch, Harold J. 1980. Kaititj nominal inflection: Some comparative notes. Pp. 259-274 

in B. Rigsby and P. Sutton, eds. Papers in Australian Linguistics 13: Contributions to 

Australian Linguistics. Australian National University: Pacific Linguistics Series A, 

Number 59. 

Kornfilt, Jaklin. 1997. Turkish. London and New York: Routledge. 

Kosch, Ingeborg M. 1998. Thoughts on suppletion in Northern Sotho. South African 

Journal of African Languages 18.2: 33-40. 

Kramer, Ruth. 2005. Root and pattern morphology in Coptic. Ms., University of 

California, Santa Cruz. 

281

Kramer, Ruth. To appear. Root and pattern morphology in Coptic: Evidence for the root. 

In Chris Davis, Amy Rose Deal, and Youri Zabbal, eds. Proceedings of the 36th 

Annual Meeting of the North Eastern Linguistic Society. Amherst, MA: 

University of Massachusetts Graduate Linguistics Student Association. 

Krämer, Martin. 1998. A correspondence approach to vowel harmony and disharmony. 

Collaborative Research Centre 282 Working Paper Number 107, Heinrich-Heine- 

Universität Düsseldorf. 

Ladefoged, Peter. 1982. A Course in Phonetics. New York: Harcourt Brace Jovanovich. 

Lapointe, Steven G. 1999. Stem selection and OT. Yearbook of Morphology 1999: 263- 

297. 

Lauriault, James. 1948. Alternating-mora timing in Shipibo. International Journal of 

American Linguistics 14: 22-24. 

Lazard, Gilbert and Louise Peltzer. 2000. Structure de la langue tahitienne. Paris: 

Peeters. 

le Bleis, Yves and Daniel Barreteau. 1987. Les extensions verbales en Mafa. Pp. 99-114 

in Herrmann Jungraithmayr and Henry Tourneux, eds. Etudes tchadiques: Classes et 

extensions verbales. 

Lee, Hansol H.B. 1989. Korean Grammar. New York: Oxford University Press. 

Leben, William. 1973. Suprasegmental Phonology. PhD. dissertation, MIT. 

Lewis. Geoffrey L. 1967. Turkish Grammar. Oxford: Clarendon Press. 

Li, Bing. 1996. Tungusic Vowel Harmony. The Hague: Holland Academic Graphics. 

Lieber, Rochelle. 1980. On the Organization of the Lexicon. PhD. dissertation, MIT. 

Lindstrom, Lamont and John Lynch. 1994. Kwamera. München: Lincom Europa. 

Lipkind, William. 1945. Winnebago Grammar. New York: King’s Crown Press. 

Mareš, František Václav. 1965. The Origin of the Slavic <strong>Phonological</strong> System and Its 

Development up to the End of Slavic Language Unity. Ann Arbor: Michigan Slavic 

Materials. 

Martin, Samuel E. 1954. Korean Morphophonemics. William Dwight Whitney Linguistic 

Series. Baltimore: Linguistic Society of America. 

282

Martin, Samuel E. and Young-Sook C. Lee. 1964. Beginning Korean. New Haven, 

Connecticut: Yale University Press. 

Martinet, A. 1952. Langues à syllabes ouvertes: Le cas du slave commun. Zietschrift für 

Phonetic und allgemeine Sprachwissenschaft 6: 145-163. 

Mascaró, Joan. 1996. External allomorphy and lexical representation. Ms., Universitat 

Autònoma de Barcelona. 

McCarthy, John J. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 

17: 207-263. 

McCarthy, John J. 2003. OT constraints are categorical. Phonology 20: 75-138. 

McCarthy, John J. and Alan S. Prince. 1990. Foot and word in prosodic morphology: The 

Arabic broken plural. Natural Language and Linguistic Theory 8.2: 209-283. 

McCarthy, John J. and Alan Prince. 1993a. Generalized Alignment. Yearbook of 

Morphology 1993: 79-153. 

McCarthy, John J. and Alan Prince. 1993b. Prosodic Morphology I: Constraint 

interaction and satisfaction. Ms., University of Massachusetts, Amherst and Rutgers 

University. 

McCarthy, John J. and Alan Prince. 1994. The Emergence of the Unmarked. Ms., 

University of Massachusetts, Amherst, and Rutgers University. 

McCarthy, John J. and Alan Prince. 1995. Faithfulness and reduplicative identity. Pp. 

249-384 in Jill Beckman, Laura Dickey and Suzanne Urbanczyk, eds. University of 

Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory. 

Amherst, MA: University of Massachusetts Graduate Linguistics Student 

Association. 

McCawley, James D. 1979. On the role of notation in generative phonology. Pp. 204- 

216 in Adverbs, Vowels, and Other Objects of Wonder. Chicago: University of 

Chicago Press. 

McFarland, Teresa. 2007. Free affix order in Filomeno Mata Totonac. Paper presented at 

the 2007 LSA Annual Meeting, Anaheim, California. 

McGregor, William. 1990. A Functional Grammar of Gooniyandi. Amsterdam: John 

Benjamins Publishing Company. 

Melnar, Lynette R. 2004. Caddo Verb Morphology. Lincoln, Nebraska: University of 

Nebraska Press. 

283

Mester, R. Armin. 1994. The quantitative trochee in Latin. Natural Language and 

Linguistic Theory 12: 1-61. 

Mohanan, K.P. 1986. The Theory of Lexical Phonology. Dordrecht: Reidel. 

Mürk, Harri William. 1997. A Handbook of Estonian: Nouns, Adjectives, and Verbs. 

Bloomington, Indiana: Indiana University Research Institute for Inner Asian Studies. 

Nash, David. 1986. Topics in Warlpiri Grammar. New York: Garland Publishing, Inc. 

Németh, Gyula. 1962. Turkish Grammar. English translation of the German original by 

T. Halasi-Kun. The Hague: Mouton. 

Newman, Stanley S. 1965. Zuni Grammar. University of New Mexico Publications in 

Anthropology 14. Albuquerque: University of New Mexico Press. 

Nikiema, Emmanuel. 1999. De la variation du déterminant /la/ dans les créoles haïtien et 

st-lucien. Lingua 107: 69-93. 

Noyer, Rolf. 1994. Mobile affixes in Huave: Optimality and morphological 

wellformedness. Pp. 67-82 in Eric Duncan, Donka Farkas, and Philip Spaelti, eds. 

Proceedings of the Twelfth West Coast Conference on Formal Linguistics. Stanford: 

CSLI. 

Oates, Lynette F. 1988. The Muruwari Language. Pacific Linguistics C-108. Canberra: 

Australian National University. 

Odden, David. 1993. Interaction between modules in Lexical Phonology. Pp. 111-144 in 

Sharon Hargus and Ellen Kaisse, eds. Phonetics and Phonology 4: Studies in Lexical 

Phonology. San Diego: Academic Press, Inc. 

Odden, David. 1996. The Phonology and Morphology of Kimatuumbi. Oxford: Clarendon 

Press. 

Orgun, Cemil Orhan. 1996. Sign-Based Morphology and Phonology with Special 

Attention to Optimality Theory. PhD. dissertation, UC Berkeley. 

Orgun, Cemil Orhan and Ronald Sprouse. 1999. From MParse to Control: Deriving 

ungrammaticality. Phonology 16: 191-224. 

Osburne, A. 1975. A Transformational Analysis of Tone in the Verb System of Zahao 

(Laizo) Chin. PhD. dissertation, Cornell University. 

Oswalt, Robert L. 1960. A Kashaya Grammar (Southern Pomo). PhD. dissertation, UC 

Berkeley. 

284

Parker, E.M. and R. J. Hayward. 1985. An Afar-English-French Dictionary (with 

Grammatical Notes in English). London: School of Oriental and African Studies, 

University of London. 

Paster, Mary. 2001. What determines suffix ordering in Fula? Paper presented at the 32nd 

Annual Conference on African Linguistics, UC Berkeley. 

Paster, Mary. 2004. Vowel height harmony and blocking in Buchan Scots. Phonology 

21.3: 359-407. 

Paster, Mary. 2005a. Subcategorization vs. output optimization in syllable-counting 

allomorphy. Pp. 326-222 in John Alderete, Chung-Hye Han, and Alexei Kochetov, 

eds. Proceedings of the Twenty-Fourth West Coast Conference on Formal 

Linguistics. Somerville, MA: Cascadilla Proceedings Project. 

Paster, Mary. 2005b. Pulaar verbal extensions and phonologically driven affix order. 

Yearbook of Morphology 2005: 155-199. 

Paster, Mary. 2006. A survey of phonological affix order with special attention to Pulaar. 

Leah Bateman and Cherlon Ussery, eds. Proceedings of the 35th Annual Meeting of 

the North Eastern Linguistic Society: Volume 2: 491-506. Amherst, MA: University 

of Massachusetts Graduate Linguistics Student Association. 

Paster, Mary and Rosemary Beam de Azcona. 2005. A phonological sketch of the 

Yucunany dialect of Mixtepec Mixtec. Pp. 61-76 in Proceedings of the Seventh 

Annual Workshop on American Indigenous Languages. 

Pater, Joe. 1999. Austronesian nasal substitution and other *NC̥ effects. Pp. 310-343 in 

Rene Kager, Harry van der Hulst, and Wim Zonneveld, eds. The Prosody- 

Morphology Interface. Cambridge: Cambridge University Press. 

Patz, Elizabeth. 1991. Djabugay. Pp. 245-347 in Robert M.W. Dixon and Barry J. Blake, 

eds. The Handbook of Australian Languages 4. Melbourne: Oxford University Press. 

Patz, Elizabeth. 2002. A Grammar of the Kuku Yalanji Language of North Queensland. 

Canberra: Pacific Linguistics. 

Payne, David L. 1981. The Phonology and Morphology of Axininca Campa. Dallas, 

Texas and Arlington, Texas: Summer Institute of Linguistics and University of Texas 

at Arlington. 

Pike, Eunice V. and Thomas Ibach. 1978. The phonology of the Mixtepec dialect of 

Mixtec. Pp. 271-285 in Mohammed Jazayery, Edgar C. Polomé, and Werner Winter, 

eds. Linguistic and Literary Studies in Honor of Archibald A. Hill, Volume 2: 

Descriptive Linguistics. The Hague: Mouton. 

Poser, William. 1985. There is no domain size parameter. GLOW Newsletter 14: 6-67. 

285

Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14: 19-100. 

Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in 

Generative Grammar. Technical Report 2. New Brunswick, New Jersey: Rutgers 

Center for Cognitive Science, Rutgers University. 

Pulleyblank, Douglas. 1996. Neutral vowels in Optimality Theory: A comparison of 

Yoruba and Wolof. Canadian Journal of Linguistics/Revue canadienne de 

linguistique 41.4: 295-347. 

Pulleyblank, Douglas. 2002. Harmony drivers: No disagreement allowed. Pp. 249-267 in 

Julie Larson and Mary Paster, eds. Proceedings of the Twenty-Eighth Annual 

Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society. 

Radhakrishnan, R. 1981. The Nancowry Word: Phonology, Affixal Morphology, and 

Roots of a Nicobarese Language. Edmonton, Alberta: Linguistic Research, Inc. 

Rice, Keren. 2000. Morpheme Order and Semantic Scope: Word Formation in the 

Athapaskan Verb. Cambridge: Cambridge University Press. 

Riggs, S.R., ed. 1852. Grammar and Dictionary of the Dakota Language. New York: 

G.P. Putnam. 

Rose, Sharon. 2000a. Multiple correspondence in reduplication. Pp. 315-326 in M. Juge 

and J. Moxley, eds. Proceedings of the Twenty-Third Annual Meeting of the Berkeley 

Linguistics Society. Berkeley: Berkeley Linguistics Society. 

Rose, Sharon. 2000b. Epenthesis positioning and syllable contact in Chaha. Phonology 

17.3: 397-425. 

Rounds, Carol. 2001. Hungarian: An Essential Grammar. New York: Routledge. 

Sands, Kristina. 1996. The Ergative in Proto-Australian. München: LINCOM Europa. 

Schuh, Russell G. 1998. A Grammar of Miya. University of California Publications in 

Linguistics 130. Berkeley and Los Angeles: University of California Press. 

Schwarze, Christoph. 1999. Inflectional classes in Lexical Functional Morphology: Latin 

-sk- and its evolution. In Miriam Butt and Tracy Holloway King, eds. The 

Proceedings of the LFG ’99 Conference. Stanford: CSLI. 

Selkirk, Elisabeth. 1982. The Syntax of Words. Cambridge, Massachusetts: MIT Press. 

Selkirk, Elisabeth. 1984. Phonology and Syntax. Cambridge, Massachusetts: MIT Press. 

286

Sezer, Engin. 1981. The k/Ø alternation in Turkish. Pp. 354-382 in G.N. Clements, ed. 

Harvard Studies in Phonology. Bloomington: Indiana University Linguistics 

Club. 

Sharp, Janet Catherine. 2004. Nyangumarta: A Language of the Pilbara Region of 

Western Australia. Canberra: Pacific Linguistics. 

Shaw, Patricia A. 1980. Theoretical Issues in Dakota Phonology and Morphology. New 

York: Garland. 

Sibanda, Galen. 2004. Verbal Phonology and Morphology of Ndebele. PhD. dissertation, 

UC Berkeley. 

Simpson, Jane and Margaret Withgott. 1986. Pronominal clitic clusters and templates. 

Syntax and Semantics 19: The Syntax of Pronominal Clitics: 149-174. 

Slocum, Marianna C. 1948. Tzeltal (Mayan) noun and verb morphology. International 

Journal of American Linguistics 14: 77-86. 

Sohn, H.M. 1975. Woleaian Reference Grammar. Honolulu: University Press of Hawaii. 

Sohn, H.M. and A.F. Tawerilmang. 1976. Woleaian-English Dictionary. Honolulu: 

University Press of Hawaii. 

Spagnolo, L. M. 1933. Bari Grammar. Verona, Italy: The Nigrizia Printing Press School. 

Speas, Margaret. 1990. Phrase Structure in Natural Language. Dordrecht: Kluwer. 

Steriade, Donca. 1988. Reduplication and syllable transfer in Sanskrit. Phonology 5: 73- 

155. 

Stump, Gregory. 1992. On the theoretical status of position class restrictions on 

inflectional affixes. Yearbook of Morphology 1991: 211-241. 

Stump, Gregory. 1993. Position classes and morphological theory. Yearbook of 

Morphology 1992: 129-180. 

Terrill, Angela. 1998. Biri. München: LINCOM Europa. 

Thompson, D.A. 1976. Kuuku Yaʔu. Pp. 329-331 in R.M.W. Dixon, ed. Grammatical 


Studies. 

Timberlake, Alan. 2004. A Reference Grammar of Russian. Cambridge: Cambridge 


287

Tranel, Bernard. 1996a. French liaison and elision revisited: A unified account within 

Optimality Theory. Pp. 433-455 in Claudia Parodi, Carlos Quicoli, Mario Saltarelli, 

and María-Luisa Zubizarreta, eds. Aspects of Romance Linguistics. Washington, DC: 

Georgetown University Press. 

Tranel, Bernard. 1996b. Exceptionality in Optimality Theory and final consonants in 

French. Pp. 275-291 in Karen Zagona, ed. Grammatical Theory and Romance 

Languages. Amsterdam and Philadelphia: John Benjamins. 

Tryon, Darrell T. 1970. Conversational Tahitian: An Introduction to the Language of 

French Polynesia. Berkeley and Los Angeles: University of California Press. 

Underhill, Robert. 1976. Turkish Grammar. Cambridge, Massachusetts: MIT Press. 

Valentine, J. Randolph. 2001. Nishnaabemwin Reference Grammar. Toronto: University 

of Toronto Press. 

van Marle, J. 1985. On the Paradigmatic Dimension of Morphological Creativity. 

Dordrecht: Foris. 

van Marle, J. 1986. The Domain Hypothesis: The study of rival morphological processes. 

Linguistics 24: 601-627. 

van Wijk, Judith. 2002. The Dutch plural landscape. Linguistics in the Netherlands 2002: 

211-221. 

Vaux, Bert. 1998. The Phonology of Armenian. Oxford: Oxford University Press. 

Vaux, Bert. 2003. Syllabification in Armenian, Universal Grammar, and the Lexicon. 

Linguistic Inquiry 34.1: 91-125. 

Wade, Terence. 2002. Russian Grammar and Verbs. Oxford: Oxford University Press. 

Walsh Dickey, Laura. 1999. Syllable count and Tzeltal segmental allomorphy. Pp. 323- 

334 in John R. Rennison and Klaus Kühnhammer, eds. Phonologica 1996: 

Syllables!?. The Hague: Thesus. 

Watson, Janet C.E. 2002. The Phonology and Morphology of Arabic. Oxford: Oxford 


Werner, Roland. 1993. Tidn-Aal: A Study of Midob. Berlin: Dietrich Reimer Verlag. 

Wiering, Elisabeth, and Marinus Wiering. 1994. The Doyayo Language: Selected Studies. 

Arlington, Texas: SIL and University of Texas at Arlington. 

288

Wolf, Matthew and John J. McCarthy. To appear. Less than zero: Correspondence and 

the null output. In Curt Rice, ed. Modeling Ungrammaticality in Optimality Theory. 

London: Equinox. 

Wordick, F.J.F. 1982. The Yindjibarndi Language. Canberra, Australia: Pacific 

Linguistics. 

Yip, Moira. 2004. <strong>Phonological</strong> markedness and allomorph selection in Zahao. Language 

and Linguistics 5.4: 969-1001. 

Yu, Alan C.L. 2003. The Morphology and Phonology of Infixation. PhD. dissertation, UC 

Berkeley. 

Yu, Alan. In press. A Natural History of Infixation. Oxford University Press. 

Zoll, Cheryl. 1996. Parsing Below the Segment in a Constraint Based Framework. PhD. 

dissertation, UC Berkeley. 

Zoll, Cheryl. 2002. Clash-driven tone mapping. MIT Working Papers in Linguistics 42: 

377-404. Cambridge, Massachusetts: MIT Working Papers in Linguistics. 

Zwicky, Arnold. 1985. How to describe inflection. Pp. 372-386 in Mary Niepokuj, Mary 

VanClay, Vassiliki Nikiforidou, and Deborah Feder, eds. Proceedings of the Eleventh 

Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics 

Society. 

Zwicky, Arnold M. 1986. The general case: Basic form versus default form. Pp. 305- 314 

in Vassiliki Nikifordou, Mary VanClay, Mary Niepokuj, and Deborah Feder, eds. 

Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society. 

Berkeley: Berkeley Linguistics Society. 

289

Appendix: Surveyed languages 

Note: references to languages in the text included what was deemed the most useful or relevant 

level of specificity in classification. Here, I list the phylum to which each language belongs, in 

order to facilitate an assessment of the range of languages surveyed. Linguistic affiliations are 

from Ethnologue (Gordon 2005). 

Abbreviations: AA = Afro-Asiatic; IE = Indo-European; MP = Malayo-Polynesian; NC = Niger- 

Congo; NS = Nilo-Saharan; PN = Pama-Nyungan. 

Language name Phylum Location Reference 

Armenian IE: Armenian Armenia Andonian 1999 

Axininca Campa Arawakan: Maipuran Peru Payne 1981 

Bari NS: Eastern Sudanic Sudan Spagnolo 1933 

Biak Austronesian: MP New Guinea Booij 2005 

Bidjara Australian: PN Australia Breen 1976a 

Biri Australian: PN Australia Terrill 1998 

Caddo Caddoan: Southern Oklahoma Melnar 2004 

Chimariko Hokan: Northern Northwestern California Conathan 2002 

Coptic AA: Egyptian Egypt Kramer 2005 

Dakota Siouan: Siouan Northern United States Shaw 1980 

Proper 

Diyari Australian: PN Australia Austin 1981 

Dja:bugay Australian: PN Australia Patz 1991 

Dutch IE: Germanic Netherlands Booij 1997 

Duuŋidjawu Australian: PN Australia Kite and Wurm 2004 

Dyirbal Australian: PN Australia Dixon 1972 

English IE: Germanic United Kingdom Zwicky 1986 

Estonian Uralic: Finnic Estonia Mürk 1997 

Finnish Uralic: Finnic Finland Karlsson 1999 

Gooniyandi Australian: Bunaban Australia McGregor 1990 

Haitian Creole creole Haiti Hall 1953 

Hungarian Uralic: Finno-Ugric Hungary Rounds 2001 

Italian IE: Italic Italy Hall 1948 

Jivaro Jivaroan Ecuador de Maria 1918 

Kaititj Australian: PN Australia Koch 1980 

Kashaya Hokan: Northern Northern California Oswalt 1960 

Kimatuumbi NC: Atlantic-Congo Tanzania Odden 1996 

Korean isolate Korea Lee 1989 

Kuku Yalanji Australian: PN Australia Patz 2002 

Kuuku Yaʔu Australian: PN Australia Thompson 1976 

Kwamera Austronesian: MP Vanuatu Lindstrom and Lynch 1994 

Latin IE: Italic Vatican State Mester 1994 

Mafa AA: Chadic Cameroon Le Bleis and Barreteau 1987 

Manchu Altaic: Tungus China Li 1996 

Maori Austronesian: MP New Zealand Biggs 1961 

Martuthunira Australian: PN Australia Dench 1995 

290

Midob NS: Eastern Sudanic Sudan Werner 1993 

Mixtepec Mixtec Oto-Manguean: Mexico Paster and Beam de Azcona 2005 

Mixtecan 

Miya AA: Chadic Nigeria Schuh 1998 

Moroccan Arabic AA: Semitic Morocco Harrell 1962 

Muruwari Australian: PN Australia Oates 1988 

Nancowry Austro-Asiatic: Nicobar Islands Radhakrishnan 1981 

Mon-Khmer 

Ngiyambaa Australian: PN Australia Donaldson 1980 

Nhanda Australian: PN Australia Blevins 2001 

Nishnaabemwin Algic: Algonquian Ontario Valentine 2001 

Northern Sotho NC: Atlantic-Congo South Africa Kosch 1998 

Nyangumarta Australian: PN Australia Sharp 2004: 117 

Qafar AA: Cushitic Ethiopia Parker and Hayward 1985 

Russian IE: Slavic Russia Timberlake 2004 

Saami Uralic: Sami Norway Dolbey 1997 

Sa’ani Arabic AA: Semitic Egypt Watson 2002 

Shipibo Panoan: Peru Elías-Ulloa 2004 

North-Central 

Spanish IE: Italic Spain Kikuchi 2001 

Tahitian Austronesian: MP French Polynesia Tryon 1970 

Turkana NS: Eastern Sudanic Kenya Dimmendaal 1983 

Turkish Altaic: Turkic Turkey Lewis 1967 

Tzeltal Mayan: Mexico Slocum 1948 

Cholan-Tzeltalan 

Wangkumara Australian: PN Australia Breen 1976b 

Warlpiri Australian: PN Australia Nash 1986 

Warluwara Australian: PN Australia Breen 1976c 

Warrgamay Australian: PN Australia Dixon 1980 

Winnebago Siouan: Siouan Wisconsin Lipkind 1945 

Proper 

Woleaian Austronesian: MP Micronesia Sohn 1975 

Yidiɲ Australian: PN Australia Dixon 1977 

Yindjibarndi Australian: PN Australia Wordick 1982 

Yingkarta Australian: PN Australia Dench 1998 

Zahao Sino-Tibetan: Burma Osburne 1975 

Tibeto-Burman 

Zuni isolate New Mexico Newman 1965 

291

Phonological Conditions on Affixation

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?