Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Harmony in gestural phonology
(USC Thesis Other)
Harmony in gestural phonology
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
HARMONY IN GESTURAL PHONOLOGY
Caitlin Michele Smith
A Dissertation Presented to the
Faculty of the Graduate School of the University of Southern California
In Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy in Linguistics
August 2018
i
Acknowledgements
At long last, I have written a dissertation. Everyone told me it would be tough going, and
I was still surprised by how difficult it was at times. Luckily, I had many great people (and several
animals) in my life to help me through it.
First, of course, I must acknowledge my family. Thanks to my parents, Greg and Michele,
for saying things like ‘when you go to college’ over and over while I was growing up, and then
sending me to college when the time came. Thanks to Auntie Marcia Hausken for providing me a
home away from home during the many years I’ve been in school. Thanks to Connor, Cristian and
Danielle, Ian, and Elena, for being there and keeping me sane while I did this. Thanks also to
‘Baby’ Madison for making my visits home nice and fun by demanding that I stop working to play
hide and seek or go for a bike ride. Thanks to the pups, Kirby and Peach, for always keeping me
company and getting me out of the house.
I also owe a lot to the linguistics department of UCLA for getting me started in the field
and continuing to contribute to my studies when I moved across town to go to USC. Bruce Hayes
taught the very first linguistics course I ever took and managed to get me hooked. Soon after, Pat
Keating offered me my first job in linguistics and provided me with my first experience with
research. Thanks so much to both of you for getting me started, and for inviting the LA phonology
scene into your home every few months and giving us wine and cheese.
While I was at UCLA, I also made several lifelong (and non-linguist!) friends. I’d like to
especially show my appreciation for the friendship of Emma Sanford Fowler and Nick Koffroth.
Emma, thanks for being my worldwide traveling buddy. I’m so glad you’re still traveling the world
and sending me pictures of every gelato you eat. Nick, thanks for many a diner brunch and
study/work session, for sharing your family with me (and everyone), and for always taking the
ii
time to remind me that my hard work was worth it because, in the words of Joe Biden, ‘this is a
big fucking deal.’
Southern California has been a great place to be a linguist. I’ve so enjoyed the collaborative
spirit that runs between the SoCal linguistics departments, especially those at USC, UCLA, UCSD,
and CSU Long Beach. Thanks to Eric Baković and Jesse Zymet for organizing the first couple of
iterations of SCAMP (Southern California April/Annual Meeting on Phonology) that brought
these departments together. Thanks also to everyone at these schools who has listened to my ideas
(the good ones and the wacky ones) over the last few years. Thanks especially to Adam Chong
and Jesse Zymet at UCLA, and Adam McCollum and Kati Hout at UCSD, for your friendship and
stimulating discussion of our ideas.
A million thanks also go to my friends, classmates, and professors at USC. Thanks to Dani
Byrd, Shri Narayanan, Asterios Toutios, and Mike Proctor for all the hard work you’ve done for
the SPAN (Speech Production and Articulation kNowledge) group at USC. All of the members of
this group have produced so much great MRI work over the years, and it has come from many a
long and grueling Sunday evening (and sometimes ultra-early morning) at LA County Hospital.
This sacrifice of your time is greatly appreciated. I would also like to acknowledge Roumyana
Pancheva and Elsi Kaiser, who took the time to help me with my work, with practice talks, and
with making sure I was on track in the program.
Among my classmates, I first thank my cohort-mates Ulrike Steindl, Peter Guekguezian,
Huilin Fang, Alfredo Garcia Pardo, and Dasha Henderer, for being the first friends I made in the
department. Thanks also to Xiao He, Canan Ipek, Sarah Ouwayda, Ben Parrell, Fangying Hsieh,
Mary Byram Washburn, Iris Chuoying Ouyang, Christina Hagedorn, Priyanka Biswas, and
Mythili Menon for doing this whole linguistics thing before me and showing me how it’s done; to
iii
Michal Temkin Martínez for being an inspiring academic big sis and all-around superstar; and to
Aaron Walker and Yi Hsien Liu Walker for being pals and inviting me into your home on
numerous occasions. Thanks to Mairym Llorens Monteserin for being such an enthusiastic fan of
everyone else’s research, including mine; to Monica Do for the times we spent hanging out in
DTLA; to Maury Lander-Portnoy Courtland for being a lovely, warm, joyful person; to Andres
Benítez for the countless hours spent playing Cards Against Humanity and Grand Theft Auto; to
Reed Blaylock for being a great collaborator and dancer; to Hayeun Jang for all the time we’ve
spent bouncing ideas off of each other; and to all of the members of USC’s Phonlunch group.
Thanks to my dear, dear friends: Brian Hsu, Charlie O’Hara, Syed Saurov, Ksenia Bogomolets
(honorary Trojan!), Ana Besserman, and Cynthia Yoonjeong Lee. You’ve spent entirely too much
time working at the coffee shop with me (brew crew!), taken my late-night panicked calls,
responded to my all-the-time panicked messages, and provided me with more good times than I
thought I could expect as a grad student.
Finally, I absolutely must thank the members of my committee: Rachel Walker, Karen
Jesney, Louis Goldstein, Khalil Iskarous, and Krishna Nayak. Krishna, thank you for serving as
the external member on my committee, and for providing a valuable perspective on my work from
outside of the field of linguistics. Khalil, thank you for guiding me through my first couple of years
in graduate school. You taught me so much about phonetics, physics, dynamics, programming,
and math. Your passion for the field of linguistics (and pretty much every field of science) is
inspiring, and your ability to see how linguistics interfaces within other fields is eye-opening.
Louis, thank you for creating the theory in which I now conduct my research, and thank you for
teaching me about it throughout my time at USC. You are a brilliant, adventurous, and generous
spirit. I have always found inspiration in the way you encourage your students to go out and make
iv
things happen for themselves. Karen, thank you for being astonishingly and unwaveringly good at
your job. I always walk away from our meetings feeling better about my work, and like I’ve learned
so much in a short amount of time. I hope one day I know half as much about phonology as you
do.
Finally, Rachel, I absolutely could not have done this without you. Thank you for showing
me what true hard work and dedication look like. I admire so much how you conduct yourself as
a scholar and how you live your life. Thanks for keeping me on track, for telling me that my ideas
were any good, for reading so many previous drafts of this dissertation, and for being more
encouraging and patient than I had any right to expect. You make me immeasurably better at
everything I do, and I will never stop thanking my lucky stars that I’ve had the privilege of knowing
you.
v
Harmony in Gestural Phonology
Abstract
In this dissertation, I develop the Gestural Harmony Model, a model of harmony situated
within a phonological framework that assumes gestures as the units of subsegmental
representation. Originally developed within Articulatory Phonology (Browman & Goldstein 1986,
1989, et seq.), gestures are dynamically defined units of phonological representation that are
specified for a target articulatory state of the vocal tract. In the Gestural Harmony Model, harmony
is triggered when a gesture extends its period of activation and overlaps other segments in a word.
To model this ability of a gesture to extend its activation, I propose the addition of two new
parameters to the representation of gestures: persistence and anticipation. With the addition of
these parameters, gestures can be specified as either self-deactivating or persistent (non-self-
deactivating), and as either self-activating or anticipatory (early-activating). A persistent gesture
is one that does not self-deactivate when its goal articulatory state is achieved, thus overlapping
following segments and triggering progressive (rightward) harmony. An anticipatory gesture is
one that is activated early, thus overlapping preceding segments and triggering regressive
(leftward) harmony.
In addition to these representational innovations, I develop a phonological grammar,
situated within the framework of Optimality Theory (Prince & Smolensky 1993/2004), that
operates over gestural representations. The presence of harmony in a language is attributed to
whether the segments in a language’s surface phonological inventory contain persistent and/or
anticipatory gestures. As a result, in the Gestural Harmony Model patterns of harmony triggering
result from the interaction of markedness and faithfulness constraints that shape the surface
inventory and determine the distributions of the segments in that inventory. One of the major
vi
advantages of the approach to harmony triggering in the Gestural Harmony Model is that harmony
systems in which bearers of a harmonizing property idiosyncratically trigger or fail to trigger
harmony can be attributed to preservation of a contrast between persistent and self-deactivating
gestures in the case of progressive harmony, and anticipatory and self-activating gestures in the
case of regressive harmony. This approach to harmony triggering avoids the pathological
predictions made by some other analyses of phonological idiosyncrasy and exceptionality.
The Gestural Harmony Model’s representation of harmony also proves advantageous in
the analysis of transparency and blocking. In this model, transparency and blocking are the results
of two distinct theoretical mechanisms, successfully accounting for the distinct crosslinguistic
patterns in the attestation of transparent and blocking segments in some types of harmony. I
analyze transparent segments as undergoers of harmony that include in their representations a
gesture that is antagonistic to a harmonizing gesture. Antagonistic gestures are those that are
specified for directly conflicting target articulatory states of the vocal tract, and as a result enter
into competition with one another. Transparency arises when intergestural competition is resolved
in favor of the gesture of the transparent segment due to its greater specified gestural strength.
Blocking of harmony, on the other hand, results from a different theoretical mechanism:
intergestural inhibition, by which one gesture deactivates another. The Gestural Harmony Model’s
splitting of transparency and blocking among two distinct theoretical mechanisms makes several
advantageous typological predictions. Chief among these is that in some types of harmony, the set
of attested transparent segments is a subset of the set of attested blocking segments. This is
attributed to the idea that only certain types of segments possess the gestural makeup necessary to
surface as transparent to harmony when overlapped by a harmonizing gesture.
vii
Table of Contents
Acknowledgments .......................................................................................................................... i
Abstract ...........................................................................................................................................v
Chapter 1 Introduction ............................................................................................................1
1.1 Overview: A Gestural Model of Harmony ..............................................................1
1.2 Theoretical Background .........................................................................................11
1.2.1 Gestures in Articulatory Phonology ...........................................................11
1.2.2 Optimality Theory and Gestural Representations ......................................19
1.3 Organization of the Dissertation ............................................................................32
Chapter 2 Representing Harmony with Gestures ...............................................................36
2.1 Introduction ............................................................................................................36
2.2 Harmony as Gestural Extension .............................................................................38
2.2.1 Rounding Harmony in Kyrgyz ...................................................................38
2.2.2 Nasal Harmony in Capanahua ...................................................................49
2.2.3 Tongue Root Harmony in Nandi ................................................................55
2.3 Properties of Harmony Triggers ............................................................................60
2.4 Summary ................................................................................................................68
Chapter 3 Patterns of Harmony Triggering ........................................................................70
3.1 Introduction ............................................................................................................70
3.2 Harmony Triggers: Inventory Shaping & Distributional Restrictions ...................73
3.2.1 Revisiting Kyrgyz: Progressive Rounding Harmony ................................74
3.2.2 Revisiting Nandi: Dominant-Recessive Tongue Root Harmony ...............86
3.2.3 Summary ....................................................................................................93
3.3 Contrastive Triggering of Nasal Harmony in Acehnese and Rejang .....................94
3.4 Conditional Triggering of Harmony ....................................................................107
3.4.1 Conditional Triggering via Co-occurrence Constraints ...........................107
3.4.2 Rounding Harmony in Baiyina Oroqen ...................................................110
3.4.3 Revisiting Capanahua: Regressive & Bidirectional Nasal Harmony ......124
3.5 Conditional and Contrastive Triggering: Tongue Root Harmony in Classical
Manchu ................................................................................................................134
3.6 Alternative Accounts of Harmony Triggering .....................................................153
3.6.1 Gestural Spreading via Direct Activation Manipulation .........................154
3.6.2 Constraint Indexation ...............................................................................157
3.6.3 Pre- and Underspecification .....................................................................165
3.7 Summary ..............................................................................................................176
Appendix A: Constraint Definitions ................................................................................177
viii
Chapter 4 Patterns of Transparency & Blocking .............................................................180
4.1 Introduction ..........................................................................................................180
4.2 Typological Patterns of Transparency and Blocking ...........................................183
4.2.1 Nasal Harmony ........................................................................................184
4.2.2 Rounding Harmony ..................................................................................186
4.2.3 Tongue Root and Postvelar Harmonies ...................................................189
4.2.4 Vowel Place Harmonies ...........................................................................191
4.2.5 Summary ..................................................................................................194
4.3 Gestural Antagonism and Gestural Inhibition .....................................................195
4.4 Transparency and Blocking in Nasal Harmony ...................................................208
4.4.1 Sources of Antagonism and Incompatibility ............................................208
4.4.2 Tuyuca: Voiceless Obstruent Transparency ............................................217
4.4.3 Orejón: Obstruent Voicing and Blocking ................................................227
4.4.4 Revisiting Capanahua: Blocking by Obstruents and Liquids ..................233
4.4.5 Summary ..................................................................................................238
4.5 Transparency and Blocking in Rounding Harmony ............................................239
4.5.1 Sources of Antagonism and Incompatibility ............................................239
4.5.2 Halh Mongolian: Transparency and Blocking by High Vowels ..............243
4.5.3 Revisiting Baiyina Oroqen: Non-Triggers as Blockers and Undergoers .253
4.5.4 Tuvan: Blocking by Nonhigh Vowels .....................................................264
4.5.5 Yakut: Blocking of Cross-Height Harmony ............................................270
4.5.6 Summary ..................................................................................................283
4.6 Transparency and Blocking in Tongue Root Harmony .......................................284
4.6.1 Sources of Antagonism and Incompatibility ............................................284
4.6.2 Yoruba: Transparency and Blocking by High Vowels ............................290
4.6.3 Summary ..................................................................................................297
4.7 Comparing Analyses of Transparency and Blocking ..........................................298
4.7.1 Locality of Spreading ...............................................................................299
4.7.2 Asymmetries in Attested Transparent and Blocking Segments ...............301
4.7.3 Harmony Systems with Transparency and Blocking ...............................316
4.7.4 Partial Harmony and Sour Grapes ...........................................................318
4.7.5 Non-Undergoers of Harmony ..................................................................323
4.8 Summary ..............................................................................................................325
Appendix B: Constraint Definitions ................................................................................327
Chapter 5 A Closer Look at Gestural Strength .................................................................329
5.1 Introduction ..........................................................................................................329
5.2 Gestural Strength in the Task Dynamic Model of Speech Production ................331
5.2.1 Formal Definition of Gestural Strength ...................................................331
5.2.2 Computational Modeling of Transparency to Nasal Harmony ................334
5.3 Partial Transparency via Gradient Gestural Strength ..........................................338
5.3.1 Partial Transparency in Coeur d’Alene Salish Faucal Harmony .............338
5.3.2 Computational Modeling of Coeur d’Alene Salish Faucal Harmony ......344
ix
5.4 Contrastive Gestural Strength in Barrow Inupiaq ................................................347
5.5 Summary ..............................................................................................................359
Chapter 6 Conclusion and Further Issues .........................................................................362
6.1 Summary of the Dissertation ...............................................................................362
6.2 Further Issues .......................................................................................................365
6.2.1 Gestural Representation of Vowels .........................................................365
6.2.2 Intergestural Inhibition .............................................................................370
6.2.3 Directionality of Harmony .......................................................................377
References ...................................................................................................................................381
1
Chapter 1
Introduction
1.1 Overview: A Gestural Model of Harmony
Broadly defined, harmony is a process by which a phonological property of one segment
is taken on by one or more additional segments in some domain. The segment that serves as the
source of this phonological property is the trigger, while any segment that takes on that property
is referred to as an undergoer. In some cases, segments in the domain of harmony do not
participate in a harmony process; these are often referred to as neutral segments. These neutral
segments may either be blockers, which stop a phonological property from spreading further
throughout a word, or transparent segments, which allow a phonological property to spread
further while not taking on that property themselves. The figure in (1) illustrates this with a
schematic example of nasal harmony. In this figure, a nasal stop triggers harmony by spreading
its nasality onto a sequence of following sonorants; the obstruent [t] serves as a blocker in (1a)
and as transparent in (1b).
(1) Segmental roles in harmony
a.
b.
2
While assigning segments these roles within a harmony system is often straightforward
from a purely descriptive standpoint, the task of determining what kind of phonological
representations and grammar give rise to attested harmony systems is considerably more
challenging. A successful model of harmony must be equipped to account for typological
asymmetries in the patterning of triggers, undergoers, blockers, and transparent segments. In this
dissertation, I claim that the adoption of a phonological framework in which the units of
representation are gestures (Browman & Goldstein 1986, 1989, et seq.), rather than features
(Jakobson, Fant, & Halle 1952/1963; Chomsky & Halle 1968; Goldsmith 1976; Clements &
Hume 1995), addresses many of the issues that arise in the analysis of harmony systems.
Often, a distinction is made between consonant harmony, in which consonants assimilate
at a distance, and vowel or vowel-consonant harmony, in which segments undergo local
assimilation (see Hansson (2001/2010) and Rose & Walker (2004, 2011) for discussion of this
distinction). In this dissertation, I focus on vowel and vowel-consonant harmonies, which can be
represented as the temporal extension of some phonological property from a triggering segment
to one or more undergoer segments. I introduce the Gestural Harmony Model, and advance two
core components of this model: gestural representational units and a phonological grammar that
operates over these units. This grammar is situated within Optimality Theory (Prince &
Smolensky 1993/2004; henceforth OT) and operates over gestural representations.
Gestures, originally developed within the framework of Articulatory Phonology
(Browman & Goldstein 1986, 1989, et seq.), are goal-based units of phonological representation.
Each gesture is specified for a target articulatory state of the vocal tract. Often, this target state is
described in terms of (1) the degree of a constriction between an active and a passive articulator,
and in some cases (2) the location of that constriction. For instance, the representation of an
3
alveolar stop includes a gesture with a target articulatory state involving full closure between the
tongue tip and the alveolar ridge. Other target articulatory states are described in terms of the
degree of displacement of a primary articulator from a neutral position. For instance, the
representation of a nasal segment includes a gesture with a target articulatory state involving the
lowering and consequent opening of the velum. The achievement of a gesture’s target
articulatory state unfolds throughout a gesture’s period of activation. When sufficient time has
passed for this target state to be achieved, the gesture self-deactivates and its control over the
vocal tract comes to an end. Gestures are also specified for the articulators they may recruit to
achieve their articulatory goals, as well as the strengths with which they can command vocal
tract articulators. This is discussed in greater detail in section 1.2.1.
In the Gestural Harmony Model, harmony is triggered when a gesture extends the period
during which it is active and overlaps the gestures of other segments in a word. In order to model
this potential for a gesture to extend its activation, I propose the addition of two new parameters
to the representation of gestures: persistence and anticipation. A persistent gesture is one that
does not self-deactivate when its goal articulatory state is reached, thus triggering progressive
(rightward) harmony. An anticipatory gesture is one that activates earlier than it should based on
its position in a word, thus triggering regressive (leftward) harmony. These gestural types are
illustrated in the figure in (2), which includes three gestures specified for velum opening. In this
figure, time is represented along the horizontal dimension, and the horizontal length of each
gesture, represented by a box, represents the span of time during which the gesture is active. The
gradually climbing and falling line represents the attainment of each gesture’s target articulatory
state and subsequent return to a neutral, default articulatory state of the vocal tract.
4
(2) Typical, persistent, and anticipatory gestures
The adoption of gestural representations and the development of the Gestural Harmony
Model’s persistent and anticipatory gestures provide a unique perspective through which to
answer many of the questions that arise in the analysis of harmony systems. These questions
include those in (3):
(3) a. What kinds of rules or constraints drive harmony, and how do they relate to attested
patterns of harmony triggering?
b. How can the analysis of harmony account for crosslinguistic asymmetries in the
attested patterns of transparency and blocking?
c. How can transparency to harmony be represented while maintaining the local nature
of a spreading process?
Regarding the question in (3a), a successful model of harmony must account for systems
in which triggers of harmony are restricted to certain positions within a word, and in which only
certain types of segments may trigger harmony (see Archangeli & Pulleyblank (2007) for an
overview). For example, it is common within the Tungusic languages for rounding harmony to
be triggered only by nonhigh round vowels in the initial syllable of a word (Kaun 1995, 2004; Li
5
1996; Zhang 1996; Walker 2001). In the Gestural Harmony Model, harmony is triggered by a
persistent or an anticipatory gesture that surfaces in an output phonological form. Rather than
explicitly driving harmony, the grammar serves only to shape a language’s surface inventory
such that segments that include harmony-triggering gestures may surface in a phonological form,
and to set any distributional restrictions on these segments. Because of this, the Gestural
Harmony Model is able to straightforwardly capture attested, often seemingly complex, patterns
of harmony triggering via the interaction of a small set of markedness and faithfulness
constraints.
It is also necessary to account for cases of apparent exceptionality in the triggering of
harmony. There are numerous examples of harmony systems in which some words
idiosyncratically exhibit harmony while others do not. A well-known case of such a system is
Hungarian backness harmony (Vago 1976; Ringen & Vago 1998; Siptár & Törkenczy 2000;
Gafos & Beňuš 2006; Hayes & Londe 2006), though it is also attested in the nasal harmony
systems of several Malayo-Polynesian languages, as well as various tongue root harmony
systems. As discussed by C. Smith (2017a, 2017b), in the Gestural Harmony Model the ability of
a gesture to trigger harmony is an encoded parameter of that gesture (persistence in the case of
progressive harmony, and anticipation in the case of regressive harmony). In some languages,
this parameter serves a contrastive function, and the result of such contrast is the idiosyncratic
triggering of harmony.
The Gestural Harmony Model is unique in relying on parameters (persistence and/or
anticipation) of a phonological unit to drive harmony, rather than relying on a rule or constraint
in the phonological grammar that explicitly requires the spreading of a harmonizing element.
Feature-based analyses of harmony (see Clements 1976a; Clements & Sezer 1982; Cole & Trigo
6
1988; Piggott 1988, 1992; Archangeli & Pulleyblank 1989, 1994, 2007; Cole & Kisseberth 1994,
1995; Kaun 1995, 2004; van der Hulst & van de Weijer 1995; Walker 1998/2000, 2011; Baković
2000; and Kimper 2011, among many others), on the other hand, do not have these parameters at
their disposal, and instead rely on explicitly driving harmony by rule or constraint. While
feature-based analyses of harmony systems in which conditions are placed on the quality of a
trigger of harmony (see, for example, work by Kaun (1995), Kimper (2011), and Walker (2011,
2014)) are largely successful, such conditions do not address cases of idiosyncratic triggering.
Rather than rely on a contrastive property that distinguishes triggers from non-triggers, these
featural analyses typically rely on mechanisms designed to account for cases of phonological
exceptionality, such as constraint indexation (Pater 2000, 2009a; Flack 2008; Becker 2009). As
discussed by Finley (2010), such analyses often result in the generation of pathological harmony
patterns. The Gestural Harmony Model, on the other hand, does not rely on mechanisms for
phonological exceptionality, or on harmony-driving constraints, in the analysis of idiosyncratic
harmony triggering patterns. I will therefore argue that this model is able to avoid such
pathological predictions.
Turning to the question in (3b), another important contribution of this work concerns the
ability of the Gestural Harmony Model to capture the typological patterns and asymmetries
among attested neutral segments. C. Smith (2016a) states that among rounding harmony and
nasal harmony systems, the set of attested transparent segments is a proper subset of attested
blocking segments. In other words, the ability to surface as transparent to a harmony process is
limited to significantly smaller classes of segments, while the ability of certain classes of
segments to block harmony is less constrained. The Gestural Harmony Model successfully
accounts for this typological asymmetry by regarding transparency and blocking as the results of
7
distinct mechanisms at work within the model. Building upon the insights of Clements (1976b),
Piggott (1988), Cole & Kisseberth (1994, 1995), Walker (1998/2000, 2003), Jurgec (2011), and
others, the Gestural Harmony Model recasts transparent segments as undergoers of harmony,
rather than as truly neutral segments. Because of their unique gestural makeup, transparent
segments may be produced without a harmonizing property despite being overlapped by a
harmonizing gesture. This unique gestural makeup of a transparent segment is based on its
including a gesture whose target articulatory state is in direct conflict with the target state of the
harmonizing gesture. Crucially, only certain segments possess the antagonistic gestural makeup
that allows them to surface as transparent to harmony, successfully limiting the classes of
segments that are predicted to be transparent to harmony to precisely those that are attested
within different harmony systems.
In recasting transparent segments as undergoers of harmony, the Gestural Harmony
Model addresses another of the central debates in the study of vowel and vowel-consonant
harmonies: how to represent transparency to harmony (see question (3c) above). There is a lack
of consensus regarding how phonological forms should be represented when the spreading of a
feature seemingly skips a segment. Configurations in which multiple associations of a single
feature skip over a potential target segment are widely considered to be universally banned (see
Sagey (1988), Archangeli & Pulleyblank (1994), Ní Chiosáin & Padgett (1997, 2001), and
Walker (1998/2000) for discussion), and yet it appears that this is just what has happened in the
case of transparency in harmony. The figure in (4) illustrates.
(4) Gapped representation of transparency to the spread of feature [+nasal]
8
By appealing to the idea that gestures may overlap one another in time, and may resist
one another’s effects when their target articulatory states are in conflict, it is possible to maintain
locality of spreading in the form of uninterrupted activation of a harmonizing gesture while still
permitting a segment to surface as transparent. Such a configuration is schematized in (5), in
which a harmonizing gesture extends to overlap a gesture with a conflicting target articulatory
state.
(5) Schematic representation of transparency in the Gestural Harmony Model
In (5), the gradually climbing and falling line represents the state of the vocal tract along
some relevant articulatory dimension (e.g., velum aperture). The dashed line represents the
neutral, default articulatory state. We see that while the harmonizing gesture is active throughout
the period of time displayed here, it is temporarily pulled away from its target articulatory state
by the conflicting gesture of the transparent segment during the period in which they are
concurrently active.
While transparency arises directly from the gestural representation of transparent
segments, the Gestural Harmony Model relies on phonetically grounded gestural co-occurrence
constraints to motivate the blocking of harmony. Blocking is implemented in the model via a
newly proposed intergestural relation, inhibition, by which one gesture may deactivate another in
order to avoid their concurrent activation. Many feature-based accounts of segments’ failure to
9
undergo harmony also make use of feature co-occurrence constraints (e.g., Smolensky 1993;
Kirchner 1993; Cole & Kisseberth 1994, 1995; Kaun 1995; Walker 1998/2000; Bakovic &
Wilson 2000; Wilson 2003). Often, these constraints are utilized in the analysis of both
transparency and blocking. However, I will show that such an approach is unable to account for
the typological asymmetries in attested blocking and transparent segments identified by C. Smith
(2016a). The Gestural Harmony Model, on the other hand, predicts this asymmetry by providing
distinct analyses of transparency and blocking.
In addition, by dividing the generation of transparency and blocking among two distinct
theoretical mechanisms, the Gestural Harmony Model allows for these mechanisms to operate
independently, and in some cases concurrently. As a result, the model successfully accounts for
harmony systems that exhibit both transparency and blocking of harmony, such as Halh
Mongolian rounding harmony (Svantesson 1985; Svantesson, Tsendina, Karlsson, & Franzén
2005), discussed in section 4.5.2; Coatzospan Mixtec nasal harmony (Gerfen 1999, 2001); and
Menominee tongue root harmony (Cole & Trigo 1988; Archangeli & Pulleyblank 1994;
Archangeli & Suzuki 1995; Walker 2009, 2018).
1
Feature-based analyses of harmony have met
with mixed success in accounting for such patterns. Some analyses can account for harmony
systems that exhibit both transparency and blocking. These include analyses utilizing Agreement
by Correspondence (Hansson 2001/2010; Rose & Walker 2004), such as those proposed by
Walker (2009, 2018) and Rhodes (2012), as well as those analyses that rely on a continuum or
scale of segments’ propensity to be transparent or to block harmony, as proposed by Kaun (1995)
and Kimper (2011). However, I will show that other analyses, such as those in which the relative
ranking of two constraints determines whether a harmony system exhibits transparency or
1
Cole & Trigo analyze this as a case of height harmony. However, later analyses recast this as a harmony system
based on tongue root position.
10
blocking (e.g., Optimal Domains Theory (Cole & Kisseberth 1994, 1995)), encounter difficulty
in generating such patterns.
In introducing the Gestural Harmony Model, this dissertation has two primary goals. The
first is to provide a novel account of vowel and vowel-consonant harmonies, the full
understanding of which has proven elusive despite decades of study. The Gestural Harmony
Model is designed to solve many of the persistent puzzles that arise in the study of harmony by
drastically reconsidering how to represent harmony, and how to represent phonological forms in
general. The second goal of this dissertation is to use the study of harmony to guide and motivate
advancements within the framework of gestural phonology. The Gestural Harmony Model
redefines several important aspects of gestural representations by proposing new gestural
parameters and expanding the phonological role that individual gestural parameters can play.
Chief among these is a proposed expansion of the role that gestural strength can play in various
phonological processes, including harmony. In addition, in this dissertation I develop a
phonological grammar, based in Optimality Theory, that is designed to operate over gestural
representations. The development of the Gestural Harmony Model, then, represents an important
step forward in the study of both harmony and of gestural phonology.
Before introducing the representation of harmony or the grammatical mechanisms at
work in the Gestural Harmony Model, it is necessary to provide theoretical background on
gestures and gestural phonology. In the following section, I provide an in-depth look at gestures
as the units of representation, as well as the phonological grammar that is assumed to operate
over gestural representations.
11
1.2 Theoretical Background
1.2.1 Gestures in Articulatory Phonology
The analysis of harmony outlined in this dissertation is formulated using gestures as the
units of phonological representational; these units were originally conceived within the
framework of Articulatory Phonology (Browman & Goldstein 1986, 1989, et seq.). In
Articulatory Phonology, gestures are units of representation that are specified for the
achievement of some target articulatory state, usually involving the formation of a constriction of
the vocal tract. For instance, the consonants /t/, /d/, and /n/ all include a gesture whose target
state is closure between the tongue tip and the alveolar ridge. A gesture is specified for certain
target tract variables, or attributes of a vocal tract constriction. These tract variables usually
specify constriction location (labial, alveolar, velar, pharyngeal, etc.) and constriction degree
(wide/narrow constriction, full closure, etc.). Some gestures, such as those governing the velum
or glottis, may be specified for a single tract variable value, such as open or closed.
In addition to being specified for the tract variables that characterize a target articulatory
state, a gesture is specified for a number of additional parameters that determine how and when
that target state is achieved, and in some cases whether it is achieved at all. First, a gesture is
specified for the vocal tract articulators it may recruit to achieve its articulatory target during the
period in which that gesture is active. For instance, the gesture that produces an alveolar
consonant specifies that the tongue tip is involved as the primary articulator, the articulator that
makes contact with the alveolar ridge. In addition, the tongue body and jaw are also recruited for
the production of this tongue tip constriction.
The attainment of a gesture’s target articulatory state is modeled by a dynamically
defined second-order equation of motion that simulates the gradual attainment of a target state
from some initial condition of the vocal tract. The rate at which a dynamical system approaches
12
its goal is determined by its stiffness parameter, often denoted by k. The higher a gesture’s
stiffness, the more quickly it will reach its target articulatory state. When sufficient time has
passed for a gesture’s target state to be achieved, the gesture self-deactivates. The period from a
gesture’s activation to deactivation is referred to as its period of activation. In general,
consonantal gestures tend to have high stiffness, and vocalic gestures tend to have low stiffness.
As a result, consonantal gestures also have a shorter period of activation than vocalic gestures.
Finally, each gesture has a specified blending strength, denoted by α. When the target
articulatory states of two gestures conflict with one another, the gestures’ relative blending
strengths determine the degree to which each gesture must compromise on the achievement of its
target. In general, it is commonly assumed that consonantal gestures have a high specified
strength, while vocalic gestures have a relatively lower strength. Gestural strength and gestural
blending play a large role in the Gestural Harmony Model, particularly in the representation of
transparency; this is discussed throughout chapters 4 and 5.
All of these gestural parameters are illustrated in (6), which depicts the gestural makeup
of the English word ‘comb’ [kom].
13
(6) Gestural makeup of the segments in ‘comb’
/k/ /o/ /m/
In (6), the initial consonant /k/ is made up of three gestures. A tongue body gesture
specifying closure in the velar region of the vocal tract is responsible for the consonant’s primary
place, while a glottal opening gesture is responsible for its voicelessness. Finally, as an obstruent
this /k/ is assumed to include a velum closure gesture; the reasons for this are discussed in
section 4.4.1. The /m/, on the other hand, includes a labial closure gesture and a velum opening
gesture, which is responsible for the segment’s nasality. As a voiced consonant, it does not
include a glottal opening gesture. The stiffness (k) and strength (α) parameters for these
consonantal gestures are both assigned the abstract value ‘high’ here, denoting the relatively high
values for these parameters that are typical of consonantal gestures.
14
As is the case for consonants, vowels in gestural phonology are also specified for
constriction location and constriction degree (though see section 6.2.1 for a possible reanalysis of
vowel place in gestural phonology). In (6), the /o/ of ‘comb’ includes a gesture for a tongue body
constriction in the uvular-pharyngeal region, as well as a lip protrusion gesture that is responsible
for rounding. The stiffness k and strength α parameters of these gestures are assigned the abstract
value ‘low’ here, denoting the relatively low values for these parameters that are typical of
vocalic gestures.
Throughout this dissertation I make the simplifying assumption that all front vowels are
specified for palatal constriction, with constriction degree determining the height of these front
vowels. High back vowels are specified for constriction in the uvular region, while nonhigh back
vowels are specified for constriction in and around the pharyngeal region. The constriction
location of the vowel /o/ falls somewhere within the uvular-pharyngeal or pharyngeal regions. I
assume here that in languages with a single nonhigh back vowel, that vowel has a pharyngeal
constriction location, while in languages that distinguish between mid /o/ and low /a~ɑ/ in terms
of height, /o/ is uvular-pharyngeal while /a~ɑ/ is pharyngeal.
Articulatory Phonology assumes several computational steps in the process of speech
production, from the abstract cognitive representation of a phonological form to a set of
articulatory instructions. These steps are illustrated in (7), adapted from Browman & Goldstein
(1992, p. 160).
(7) Multi-step speech production model assumed by Articulatory Phonology
15
According to the figure in (7), a lexical item first takes the form of a coupling graph,
which specifies the coordination or coupling relations between gestures. Two gestures may be
coupled so that they are produced either at the same time as or in sequence with one another.
Synchronously coordinated gestures are said to be coupled in-phase, while sequentially
coordinated gestures are said to be coupled anti-phase to one another. A coupling graph for
‘comb’ [kom] is provided in the figure in (8). The segmental transcription is included at the top
of the figure, with the subscripts for each segment matching the subscripts of the gestures they
comprise. In the coupling graph, two gestures that are connected by a solid line are coupled in-
phase (synchronously), and a dashed arrow indicates anti-phase (sequential) coupling of two
gestures. Following Browman & Goldstein (2000), a syllable onset is coupled in-phase to a
following vowel, and a syllable nucleus is coupled anti-phase to a following coda consonant.
(8) Coupling graph for ‘comb’ [kom]
2
At this level of representation, gestures are temporally abstract; the time points of their
individual activations and deactivations are not yet set. Once the coupling relations between
gestures have been established, a coupling graph is input to the Coupled Oscillator Model
(Saltzman & Byrd 2000; Nam & Saltzman 2003). At this point, the relative timing of gestures is
stabilized based on the coupling relations that exist between them, and the activation and
2
Often, coupling graphs omit information about gestures’ stiffnesses, strengths, and non-primary articulators for
reasons of space and readability. This practice is adopted here.
16
deactivation time points of each gesture are set. The relative timing of gestures and the resulting
time points of their individual activations and deactivations are calculated by the Coupled
Oscillator Model according to gestures’ relative phases. Each gesture is associated with an
abstract planning oscillator. The coupling relation between two gestures is defined with respect
to the phases of their individual planning oscillators. When two gestures are coupled in-phase,
the 0º phases of their individual oscillators are temporally aligned. This is illustrated in (9) for
the velar closure gesture of [k
1
] and the vocalic constriction gesture of [o
2
] in ‘comb.’
(9) Relative phasing of oscillators of two in-phase coupled gestures
When two gestures are coupled anti-phase, the 0º phase of one gesture and the 180º phase
of another gesture are temporally aligned. The result is a pair of gestures that are produced in
sequence. This is illustrated in (10) for the vocalic constriction gesture of [o
2
] and the lip closure
gesture of [m
3
] in ‘comb.’
17
(10) Relative phasing of oscillators of two anti-phase coupled gestures
Once the Coupled Oscillator Model has temporally aligned all of the gestures in a
coupling graph according to the coupling relations that exist between them, the time points of the
activation and deactivation of each gesture are set. A commonly held assumption within
Articulatory Phonology is that every gesture automatically activates at the 0º phase of its
planning oscillator, and automatically deactivates at some specified phase that corresponds to the
point in time at which the gesture’s target articulatory state is achieved. However, this
assumption is crucially not adopted in the Gestural Harmony Model proposed in this dissertation.
Instead, the workings of the model are centered around the idea that the extension of gestural
activation responsible for harmony is based on the ability of some gestures to activate before
their respective 0º phases, and to remain active after they have achieved their respective target
articulatory states. This proposal and the augmentation of the Coupled Oscillator Model
necessary to implement it are discussed further in chapter 2.
The output of the Coupled Oscillator Model is a gestural score. In a gestural score, each
gesture is represented by a box whose horizontal length represents the period of time over which
a gesture is active. The horizontal positions of these boxes are the result of the coupling relations
18
that exist between these gestures in the coupling graph. The figure in (11) shows the gestural
score for ‘comb’ [kom] corresponding to the coupling graph in (8).
(11) Gestural score for ‘comb’ [kom]
Due to the in-phase (synchronous) coupling between the gestures of [k] and the gestures
of [o], all of these gestures activate at the same point in time. However, the higher stiffness of the
gestures of [k] results in their attaining their respective target articulatory states earlier, and thus
deactivating earlier. The gestures of [o], which have lower stiffness values, remain active longer.
Due to the anti-phase (sequential) coupling between the tongue body gesture of [o] and the labial
gesture of [m], the gestures of [m] activate near the end of the activation period of the vocalic
gestures. While they overlap slightly, the segments [o] and [m] are produced in sequence.
Based on the specified parameters and activation periods of the gestures represented in a
gestural score, the Task Dynamic Model of speech production (Saltzman & Munhall 1989)
determines the trajectories of specific articulators necessary to achieve the gestures’ target
articulatory states. Of particular interest to the Gestural Harmony Model is the resolution of
competition between concurrently active gestures via gestural blending. There are different
modes of blending depending upon the specifications of two competing gestures. When two
concurrently active gestures call for different values of the same tract variable, as is the case for
19
the tongue body gestures of [k] and [o] in (8) and (11), for instance, blending of that tract
variable is determined by each gesture’s specified blending strength (α). The resulting tract
variable value for the two competing gestures is a weighted average of the tract variable values
for each individual gesture, with the gesture’s α values providing the weights. This mechanism
for gestural blending is discussed in greater detail in chapters 4 and 5.
1.2.2 Optimality Theory and Gestural Representations
Articulatory Phonology’s model of speech production assumes several distinct levels of
representation, represented in (7), each directly calculated from the output of a previous level.
The Coupled Oscillator Model builds a gestural score based on the gestural parameters and
coupling relations in a coupling graph, and the Task Dynamic Model in turn determines a set of
articulatory trajectories from the gestural parameters and activation periods in a gestural score.
What this model lacks, however, is a grammar that is responsible for building the original
coupling graph. In this section, I outline such a grammar: Gestural OT (C. Smith 2016b, 2017a).
While Articulatory Phonology provides a rich set of representational primitives, in many
of its implementations it is without a grammatical mechanism to determine which phonological
structures are licit and illicit within a language. Ladefoged (1990) notes that Articulatory
Phonology is not a theory of the relation between underlying and surface phonological forms.
Note that in the speech production model in (7) articulatory trajectories are calculated from a
gestural score, and that gestural score is calculated from a coupling graph; however, there is no
mention of the source of that coupling graph. The coupling graph is usually considered to be
lexically specified in Articulatory Phonology, and therefore the underlying form of a lexical
item; there is no early level of representation at which coupling relations are not yet set. This is a
necessary assumption, as in Articulatory Phonology coupling relations between gestures
20
represent both syllable structure and basic linear ordering. However, by doing away with this
conflation of gestural ordering and prosodic structure, it is possible to construct a coupling graph
through the interaction of constraints in an Optimality Theoretic grammar. On the other hand, the
linear ordering of segments, conceptualized as sets of gestures, is specified in the input for a
lexical item.
The proposals made in this dissertation and by C. Smith (2016b, 2017a) do not represent
the first applications of the framework of OT to gestural representations. Several researchers,
notably Gafos (2002), Davidson (2003), N. Hall (2003), and Bradley (2005), have proposed
analyses in which gestures are coordinated with one another based on constraints on the
alignment of gestural landmarks (onset, achievement of target articulatory state, etc.). In a sense,
this gestural alignment replaces the workings of the Coupled Oscillator Model to coordinate
gestures with one another, and a level of representation analogous to the coupling graph is not
assumed. In contrast to that approach, I assume here a model of Gestural OT that leaves the
Coupled Oscillator Model in place and allows it to determine the relative timing of gestures
automatically from the specifications of a coupling graph. The Gestural OT grammar I assume
here is responsible only for generating the coupling graph.
Another notable application of an OT grammar to gestural representations is developed
by Tejada (2012), who examines tone spreading as the temporal extension of tone gestures.
Tejada’s framework employs constraints that operate over coupling relations between gestures,
an approach that is adopted here and by C. Smith (2016b, 2017a). However, Tejada assumes that
coupling relations are present in the input and that constraints exist to either manipulate or
preserve these relations. In contrast, I do not assume input coupling relations here. Instead, I
assume that OT constraints build and manipulate coupling graphs, but have access to information
21
about gestures’ periods of activation, including any overlap present in the gestural score, as well
as the resulting articulatory trajectories and their acoustic outcomes. The gestural score is
calculated by the Coupled Oscillator Model based on the specified gestural parameters and phase
relations present in the coupling graph. Similarly, the articulatory trajectories produced by the
Task Dynamic Model are calculated based on the gestural parameters and activation periods
present in the gestural score. The outputs of the Coupled Oscillator Model and the Task Dynamic
Model, then, are merely implementations of the information that is present in the coupling graph.
In addition, while it is not part of the formal model of speech production assumed by
Articulatory Phonology (see the figure in (7)), the acoustic output of speech can also be
calculated from the articulatory trajectories generated by the Task Dynamic Model. The content
of the coupling graph, on the other hand, is assumed to be determined by phonological
principles. Therefore, I claim that the phonological grammar should operate to produce these
coupling graphs.
I assume that the input to the OT grammar that produces output coupling graphs is a
string of linearly ordered segments, which are conceptualized here as sets of gestures. There is
little consensus across work conducted within gestural phonology as to whether the segment
should have any theoretical status. A commonly adopted assumption is Byrd’s (1996) claim that
the notion of the segment is epiphenomenal, stemming from the fact that certain sets of gestures
are underlyingly specified for extremely stable coordination relations. Gafos (2002) and N. Hall
(2003, 2006), on the other hand, develop theories that assume that phonological representations
include both segmental and gestural units. Following a similar proposal made by Walker
(2017a), I will assume here that a segment is defined as a set of gestures.
22
In the Gestural OT framework I adopt here, an input is represented by a string of
segments, each of which comprises one or more gestures. Linear ordering is indicated by a
numeric index on each segment. The numeric index on each gesture indicates its affiliation with
a segment bearing the same index. This is illustrated in (12) for the word ‘comb’ [kom], in which
all of the segments are composed of multiple gestures.
(12) Input form of ‘comb’ in Gestural OT, with linear ordering indices indicating segmental
affiliation
/ k
1
o
2
m
3
/
From the underlying linearly ordered segments (sets of gestures), the phonological
grammar will generate a set of output candidates with coupling relations between the gestures.
The computation of stabilized phase relations and periods of gestural activation by the Coupled
Oscillator Model will follow, and application of the Task Dynamic Model will yield articulator
trajectories. These stages of speech production follow directly from the content of a coupling
graph, and therefore their calculations appear to be non-phonological in nature. The phonological
grammar developed here is unable to manipulate gestural representations at any level beyond the
coupling graph, the assumed output of EVAL.
The coupling relations between gestures in a coupling graph are determined by several
constraints in the phonological grammar. Before defining these constraints, it is necessary to
define precisely how the gestures that make up a segment relate to one another. In particular, I
Tongue Body
velar closure
1
Tongue Body
uvular-pharyngeal mid
2
Lip
closure
3
Glottis
open
1
Lip
protrusion
2
Velum
open
3
Velum
closure
1
23
draw a distinction between a segment’s primary gesture and its secondary gesture(s). Gafos
(2002) recognizes this distinction between oral ‘head gestures’ and velic or laryngeal ‘secondary
gestures’ and claims that constraints on particular coordination schemes between gestures should
operate over head (i.e., primary) gestures only. Though intuitively it is not difficult to pick out
the primary gesture of a consonantal or a vocalic segment (for instance, the tongue tip gesture of
[t] should likely be considered its primary gesture rather than its glottal opening gesture), the
terms ‘primary’ and ‘secondary’ remain ill-defined.
It is not within the scope of this dissertation to develop a full definition of what
contributes to the primary versus secondary status of a gesture, but some guidelines can be laid
out. I adopt Gafos’ insight that an oral gesture should gain primary status when it is one of a set
of gestures in which one is oral and the others are not. However, this alone is insufficient to
account for consonantal and vocalic segments with secondary oral gestures, such as those
responsible for rounding or palatalization. In the case of multigestural consonants, we can appeal
to the fact that when a consonant or vowel is composed of two oral gestures, one is a consonantal
gesture and one is a vocalic gesture.
3
Definitions of the terms ‘consonantal’ and ‘vocalic’
deserve much further study, but Sproat & Fujimura (1993) offer these preliminary definitions:
‘Consonantal gestures are those that produce an extreme obstruction in the mid-sagittal plane.
Vocalic gestures are those gestures that do not produce an extreme obstruction; furthermore,
vocalic gestures may actually involve opening of a channel as in the case with velum lowering’
(p. 304). This definition of a vocalic gesture seems able to capture not only vocalic oral gestures
but velic and glottal opening gestures as well. In general, then, it can be assumed that when a
segment comprises two or more gestures, the consonantal gesture as defined by Sproat &
3
Notable exceptions to this generalization are labiovelar consonants such as /k
͡ p/, which involve the concurrent
production of two consonantal constrictions. I set aside the issue of how to classify the consonantal gestures of such
segments, though see Danis (2017) for a recent discussion of the representation of doubly articulated consonants.
24
Fujimura is primary and any vocalic gestures are secondary. In the case of vowels, however, a
vocalic gesture is considered primary; therefore, the tongue body gesture of a multigestural
vocalic segment is considered primary.
Having defined the relation between the gestures that make up a segment as one between
primary and secondary gestures, constraints on the coupling relations between gestures in a
coupling graph can be defined. Following C. Smith (2016b), the coupling relations between
heterosegmental gestures are determined by the three COUPLE constraints in (13), based on
Davidson’s (2003) ASSOCIATE constraints.
(13) COUPLE constraints used to determine intergestural coordination relations
a. COUPLE(C,V): Assign a violation mark for any primary consonantal gesture that is
not coupled in-phase to the primary gesture of the following vocalic segment.
b. COUPLE(C,C): Assign a violation mark for any primary consonantal gesture that is not
coupled anti-phase to the primary gesture of the following adjacent consonantal
segment.
c. COUPLE(V,V): Assign a violation mark for any primary vocalic gesture that is not
coupled anti-phase to the primary gesture of the following vocalic segment.
Davidson’s ASSOCIATE constraints are used to identify pairs of gestures that come under
the influence of gestural landmark-aligning COORDINATE constraints (Gafos 2002). However, the
version of Gestural OT implemented here eliminates this second step of coordination between
the landmarks of associated pairs of gestures, instead leaving the precise coordination of coupled
gestures to the calculations of the Coupled Oscillator Model.
The coupling relations that are present in the coupling graph for ‘comb’ in (8) follow
straightforwardly from the COUPLE constraints in (13). Satisfying COUPLE(C,V), the primary
consonantal tongue body gesture for [k] is in-phase (synchronously) coupled to the primary
vocalic tongue body gesture of [o]. In addition, the tongue body gesture of [o] is coupled anti-
25
phase (sequentially) to the primary consonantal labial closure gesture of the following [m]. This
anti-phase coupling between a primary vocalic gesture and a following primary consonantal
gesture is assumed to be driven not by a COUPLE constraint, but by the universal prohibition
against any gestures remaining uncoupled in an output form. The anti-phase nature of this
coupling preserves the linear ordering of these segments.
The in-phase coupling relations between onset consonants and vowels deserve further
mention. The constraint COUPLE(C,V) is satisfied by a primary consonantal gesture that is
coupled in-phase to a following primary vocalic gesture; however, this in-phase coupling means
that the two gestures will activate synchronously, not sequentially. Therefore, it is not clear that
the vocalic gesture still follows the onset consonantal gesture, as it does in the input. I will
assume that whether a gesture precedes or follows another gesture is defined with respect to the
timepoints of the achievement of each gesture’s target articulatory state, and not its activation.
Therefore, a primary consonantal gesture can be said to precede a primary vocalic gesture to
which it is coupled in-phase, owing to the vocalic gesture’s lower stiffness and later attainment
of its target articulatory state.
The COUPLE constraints defined in (13) will receive little attention here. While they play
a significant role in the analyses outlined by C. Smith (2016b), their impact upon the workings of
the Gestural Harmony Model is minimal. Therefore, they will be considered high-ranked and
inviolable in the analyses that follow.
An additional constraint is necessary to compel the composite gestures of a segment to be
coupled to one another in an output form. Note, for instance, that the secondary gestures of the
initial [k] in ‘comb’ are coupled to the primary consonantal gesture of that segment in the
coupling graph in (8), despite the lack of any COUPLE constraint requiring the presence of these
26
coupling relations. The motivation for the coupling between the composite gestures of a segment
appears to be some form of ensuring that the gestures that are segmentally affiliated in the input
are related in some way in the output. I propose that the lack of a coupling relation between a
segment’s composite gestures should be penalized by a form of INTEGRITY-IO. This constraint
was originally conceived by McCarthy & Prince (1995) as a way to prevent the splitting or
copying of segments between the input and the output. Within Gestural OT, INTEGRITY-IO is
defined as in (14).
(14) INTEGRITY-IO: Assign a violation mark to a primary gesture and a secondary gesture that
are part of the same segment (set of gestures) in the input and are not coupled to one
another in the output.
4
In addition to INTEGRITY-IO, I assume that Gestural OT includes the typical faithfulness
constraints penalizing the epenthesis and deletion of material between input and output, as well
as any changes in the specifications of representational units (McCarthy & Prince 1995). Of
particular importance to the Gestural Harmony Model is the constraint IDENT(parameter
X
)-IO,
which penalizes any changes to some gestural parameter (e.g. constriction location, stiffness,
(de-)activation phase) between the input and the output. It is defined as in (15).
(15) IDENT(parameter
X
)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical specifications for parameter X.
Regarding the epenthesis and deletion of material, Gestural OT admits MAX-IO and DEP-
IO constraints (McCarthy & Prince 1995) that refer both to whole segments (sets of gestures)
and to individual gestures within those segments. The segmental versions of these constraints,
MAX(segment)-IO and DEP(segment)-IO, are defined in (16).
4
It is also conceivable that INTEGRITY-IO could be defined such that all of the gestures that make up a single
segment must be coupled to one another, rather than merely requiring a segment’s primary gesture to be coupled to
all secondary gestures. The workings of the Gestural Harmony Model do not crucially distinguish between these two
possible definitions of INTEGRITY-IO; therefore, the matter is set aside.
27
(16) Constraints on segmental epenthesis and deletion
a. MAX(segment)-IO: Assign a violation mark to an input segment (set of gestures) that
has no output correspondent.
b. DEP(segment)-IO: Assign a violation mark to an output segment (set of gestures) that
has no input correspondent.
Constraints may also penalize the deletion and epenthesis of individual gestures within a
segment. I define these constraints following Pater’s (1999) definitions of IDENT-IO and IDENT-
OI, which penalize changes in segmental quality due to the deletion and epenthesis of privative
features from segments. To avoid confusion with IDENT(parameter
X
)-IO constraints, I will refer
to constraints on gestural deletion and epenthesis as MAX(gesture
X
)-IO and DEP(gesture
X
)-IO,
respectively. These definitions are provided in (17).
(17) Constraints on gestural epenthesis and deletion
a. MAX(gesture
X
)-IO: Assign a violation mark to a segment (set of gestures) that
includes gesture X in the input if its output correspondent does not include gesture X.
b. DEP(gesture
X
)-IO: Assign a violation mark to a segment (set of gestures) that
includes gesture X in the output if its input correspondent does not include gesture X.
In addition to the faithfulness constraints above, gestural representations are subject to
various markedness constraints. As is the case for features, certain incompatible gestures may be
prohibited from occurring with one another, either concurrently or adjacently. These co-
occurrence restrictions are enforced within Gestural OT by one of two basic constraint types:
*COUPLE and *OVERLAP.
*COUPLE constraints, as their name suggests, penalize a coupling relation between two or
more gestures whose co-occurrence is marked in some way. There are two basic forms a
28
*COUPLE constraint can take, schematized in (18). The first schema outlined below refers to a
pair of incompatible gestures, while the second refers to a set of three incompatible gestures.
5
(18) Schemas for *COUPLE constraints
a. *COUPLE(Gest
X
, Gest
Y
): Assign a violation mark for a pair of gestures of type X and
type Y that are in a coupling relation with one another.
b. *COUPLE(Gest
X
, Gest
Y
, Gest
Z
): Assign a violation mark for a gesture of type X that is
in a coupling relation with a gesture of type Y and in a coupling relation with a
gesture of type Z.
*OVERLAP constraints penalize the concurrent activation of two or more incompatible
gestures, with no reference to whether or not those gestures are coupled to one another. In this
sense, they are stricter than *COUPLE constraints. In some cases, including harmony,
incompatible gestures might not be near each other in a coupling graph, let alone coupled to one
another, and yet end up concurrently active in a gestural score. Therefore, *COUPLE constraints
are of no use in capturing their incompatibility. *OVERLAP constraints follow the schema
provided in (19). As with the *COUPLE constraints in (18), there are schemas both for pairs of
incompatible gestures and for sets of three incompatible gestures.
(19) Schemas for *OVERLAP constraints
a. *OVERLAP(Gest
X
, Gest
Y
): Assign a violation mark for a pair of gestures of type X and
type Y that are concurrently active.
b. *OVERLAP(Gest
X
, Gest
Y
, Gest
Z
): Assign a violation mark for a gesture of type X that
is concurrently active with a gesture of type Y and with a gesture of type Z.
In the version of Gestural OT that I employ here, constraints in the grammar operate to
manipulate output candidate coupling graphs. However, I claim that while constraints may only
5
It may also be desirable to include *COUPLE constraint schemas that refer to sets of four or more incompatible
gestures. However, such configurations are not relevant to the Gestural Harmony Model; therefore, the matter is set
aside.
29
directly manipulate the coupling graph, they may also evaluate information about the gestural
score, articulatory trajectories, and acoustic output that are calculated from this coupling graph.
This crucially includes information about the outcomes of gestural overlap.
The evaluation of a constraint from the *OVERLAP family is demonstrated by the tableau
in (20). Both candidates are displayed both as coupling graphs and as the gestural scores that are
calculated from those coupling graphs.
(20) Tableau illustrating violation profile for *OVERLAP
*OVERLAP
(Gest
X
, Gest
Y
)
a.
b.
*
In candidate (a), Gest
X
and Gest
Y
are antiphase coupled to one another, and Gest
Y
activates just before Gest
X
deactivates. I assume that the slight gestural overlap that arises
between two antiphase coupled gestures does not incur a violation of *OVERLAP. In candidate
(b), Gest
X
and Gest
Y
are also coupled antiphase. However, in this candidate Gest
X
extends in
duration, resulting in its overlap with Gest
Y
. In contrast with candidate (a), the gestural overlap
in candidate (b) is sufficient to violate *OVERLAP. The mechanisms behind extended gestural
30
activation, persistence and anticipation, are introduced and discussed in chapter 2, while the
mechanism for preventing gestural overlap, inhibition, is discussed in chapter 4.
Access to information about gestural activation and overlap is assumed in previous OT-
based analyses in gestural phonology as well. The gestural coordination grammar developed by
Gafos (2002) and adopted by Davidson (2003), N. Hall (2003), and Bradley (2005) relies upon a
series of constraints that coordinate the temporal landmarks of gestures with previously
determined durations. Additionally, total overlap between two gestures has a highly marked
status within these analyses because it impedes perceptual recoverability of the gestural
composition of phonological forms. Gafos (2002) and Bradley (2005) use this principle of
maximization of perceptual recoverability to motivate the inclusion of constraints calling for
little or no gestural overlap. In both of these analyses, constraints in the grammar are able to use
information about the acoustic and perceptual consequences of gestural overlap in order to
evaluate candidate output forms, despite the fact that this acoustic and/or perceptual information
is not explicitly encoded in the candidates’ phonological forms.
Similar abilities of the phonological grammar to access information about the acoustic
and perceptual consequences of phonological forms are also common in most theories of
phonology that assign a large role to speakers’ phonetic knowledge. For instance, Steriade’s
(1995, 1997, 2001, 2009) Licensing by Cue and P-Map theories and Flemming’s (1995/2002,
2004) Dispersion Theory rely heavily on the perceptual outcomes of phonological forms. In
these frameworks, perceptual cues are used to evaluate output candidates, even if those
perceptual cues are not represented in the phonological forms themselves. The phonological
grammar accesses this information by referring to the phonetic knowledge the speaker utilizes to
anticipate the phonetic outcomes of phonological forms. Boersma’s (1998, 2003) Functional
31
Phonology is also notable for the degree to which the phonological grammar may access and
utilize phonetic and perceptual details. This framework recognizes three distinct representations
of a phonological form: the segmental (underlying), the articulatory, and the perceptual.
The ability of the Gestural OT grammar to evaluate the gestural score and articulatory
trajectories that are directly predictable from a candidate coupling graph therefore does not
represent a radical departure from previous theories in terms of what phonetic information a
grammar may access about a phonological form. Within the implementation of Gestural OT I
assume here, the phonological grammar still operates to manipulate output candidate coupling
graphs, but may evaluate candidates based on any information that is available in the gestural
score, articulatory trajectories, and acoustic output that a candidate coupling graph produces.
The final markedness constraint type to be discussed here is gestural licensing. In contrast
to *OVERLAP constraints that penalize the co-occurrence of certain types of gestures, LICENSE
constraints are only satisfied when certain types of gestures do co-occur. They are defined such
that one gesture is specified as licensor and the other as licensee, following the schema in (21).
(21) Constraint schema for gestural LICENSE
LICENSE(Gest
X
, Gest
Y
): Assign a violation mark to an active gesture of type X if a
gesture of type Y is not concurrently active.
According to LICENSE(Gest
X
, Gest
Y
), Gest
X
is unlicensed and therefore incurs a violation
mark unless it is accompanied by another gesture, Gest
Y
. However, the constraint does not
require that Gest
Y
be accompanied by Gest
X
in order to be considered licensed. Unlike
*OVERLAP or *COUPLE constraints, which penalize a combination of gestures without singling
out certain marked gestures within that combination, LICENSE constraints specify a gesture that is
marked in isolation but not in combination within another gesture.
32
In addition to gestural licensing constraints, I also adopt the use of more traditional
positional licensing constraints (Zoll 1996; J. Smith 2002; Walker 2005, 2011; A. Kaplan 2008),
which license certain gestures or segments (i.e., sets of gestures) in prominent positions. These
constraints are defined such that they penalize gestures or segments that occur outside of a
licensed position. The schema for positional licensing is provided in (22).
(22) Constraint schema for positional LICENSE
LICENSE(Gest
X
, Position
Y
): Assign a violation mark to a segment/gesture of type X that is
not in position Y.
The markedness constraints *COUPLE, *OVERLAP, and LICENSE, as well as the
faithfulness constraints IDENT(parameter)-IO, MAX(gesture)-IO, DEP(gesture)-IO, and
INTEGRITY-IO introduced in this section make up the core of EVAL in the version of Gestural
OT I adopt in this dissertation. They share in common the fact that they all have the power to
manipulate the contents of a coupling graph, via either the coupling relations between gestures or
the gestural content of the coupling graph. Additional constraints specific to the analysis of
harmony will be discussed in following chapters as the workings of the Gestural Harmony Model
are introduced.
1.3 Organization of the Dissertation
This chapter has introduced the major motivations for developing the Gestural Harmony
Model, and has also laid out the basic theoretical assumptions necessary to construct an analysis
of harmony within gestural phonology. The remainder of the dissertation is organized as follows.
Chapter 2 introduces the basic representation of harmony within the Gestural Harmony
Model. It also introduces the concepts of gestural persistence (non-self-deactivation) and
anticipation (early activation) by which gestures are able to trigger harmony by extending to
33
overlap other gestures in a word. In addition to introducing the basic gestural representation of
harmony and the gestural parameters of persistence and anticipation, this chapter also discusses
the criteria a gesture must meet in order to surface as persistent or anticipatory and therefore to
serve as a trigger of harmony.
Chapter 3 introduces the constraint set responsible for triggering harmony, and includes
analyses of the different triggering patterns of several harmony systems, both simple and
complex. By treating a gesture’s ability to trigger harmony as the product of a gestural parameter
rather than the satisfaction of a grammatical constraint, the grammar developed within the
Gestural Harmony Model captures patterns of harmony triggering via the shaping of a
language’s surface phonological inventory. Because of this, the model is able to
straightforwardly capture both simple and complex patterns of harmony triggering while
avoiding both over- and under-generation pathologies. Both conditional and contrastive
triggering of harmony are the result of the interaction of markedness and faithfulness constraints,
both general and position-specific. This chapter also discusses how feature-based analyses of
such patterns, which rely on mechanisms such as constraint indexation and contrastive
underspecification in order to account for such patterns, come with undesirable typological
predictions that are avoided by the Gestural Harmony Model. In particular, the Gestural
Harmony Model is able to analyze a gesture’s ability to trigger harmony as a potentially
contrastive property, providing a straightforward account of harmony patterns that are often
treated as cases of phonological exceptionality.
Chapter 4 focuses on transparency and blocking in harmony. The chapter begins with a
description of the typological patterns of transparency and blocking in several different types of
harmony, and goes on to demonstrate how the Gestural Harmony Model successfully accounts
34
for these patterns. Crucial to the success of the Gestural Harmony Model in generating a
constrained typology of transparency and blocking is the division of the analyses of transparency
and blocking among two different theoretical mechanisms. This chapter introduces the idea of
transparency via competition between antagonistic gestures, by which certain segments that are
overlapped by a harmonizing gesture will automatically surface as transparent to harmony due to
their gestural makeup. Blocking, meanwhile, results from high-ranking constraints that penalize
the concurrent activation of incompatible gestures, and is implemented via a newly proposed
relation between gestures: intergestural inhibition. The chapter concludes by comparing the
typological predictions made by various alternative analyses of transparency and blocking.
Chapter 5 presents a closer examination of the concepts of gestural strength and
intergestural competition and blending that are central to the Gestural Harmony Model’s
representation of transparency to harmony. It presents the results of computational modeling of
transparency via intergestural competition. The chapter also investigates the possibility that the
phonological role played by gestural strength is greater than previously assumed by focusing on
cases in which gestural strength appears to serve a contrastive function.
Chapter 6 concludes the dissertation by summarizing the major contributions of the
Gestural Harmony Model to the study of both harmony and the framework of gestural
phonology. It also identifies avenues for future work and ideas for further innovation of the
model. Among these is a possible rethinking of the representation of vowels within gestural
phonology that would make it possible to account for additional types of harmonies based on
vowel place. The chapter also examines further implications of the Gestural Harmony Model’s
development of new theoretical mechanisms, such as intergestural inhibition, as well as whether
35
directional asymmetries can be built into the model via the distinct mechanisms of persistence
and anticipation.
36
Chapter 2
Representing Harmony with Gestures
2.1 Introduction
This chapter introduces the core representational innovations of the Gestural Harmony
Model. In particular, it focuses on developing the concepts of gestural persistence and
anticipation, two parameters that lead to the extended period of activation of a gesture. A trigger
of harmony is a segment that includes a gesture that is either persistent, anticipatory, or both. A
persistent gesture is specified not to self-deactivate once its goal articulatory state is reached.
Instead, a persistent gesture remains active and overlaps the gestures of following segments, the
undergoers of progressive harmony. An anticipatory gesture, on the other hand, is specified to
activate earlier than the specified 0º phase of its planning oscillator. Due to this early activation,
it overlaps the gestures of preceding segments, the undergoers of regressive harmony.
Many types of gestures may surface as persistent (non-self-deactivating) or anticipatory
(early-activating). These include lip protrusion gestures, which trigger rounding harmony; tongue
root advancement gestures, which trigger ATR harmony; and velum opening gestures, which
trigger nasal vowel-consonant harmony. Examples of each of these types of harmony are
illustrated in this chapter. In Kyrgyz (section 2.2.1), rounding harmony is triggered by round
vowels in an initial syllable. In Capanahua (section 2.2.2), nasal stops trigger a process of
regressive (leftward) nasal vowel-consonant harmony. In Nandi (section 2.2.3), ATR vowels
triggers bidirectional ATR harmony.
There are several elements of the Gestural Harmony Model’s representation of harmony
as gestural activation extension that are crucial to the model’s success in accounting for
37
crosslinguistic patterns among harmony systems. First, in the Gestural Harmony Model a
segment’s status as a trigger of harmony is directly encoded in its gestural representation via the
parameters of persistence (non-self-deactivation) and anticipation (early activation). Because of
this, the model is able to accurately and straightforwardly account for complex patterns of
harmony triggering, as detailed in chapter 3. In addition, in the Gestural Harmony Model the
undergoers of harmony include both vowels and consonants, regardless of whether a given type
of harmony is traditionally described as vowel harmony or as vowel-consonant harmony. In the
Gestural Harmony Model, the treatment of these two types of harmony is unified. This is due to
one of the crucial aspects of the Gestural Harmony Model: harmony is always strictly local. In
other words, in this model the period of activation of a harmony-triggering gesture is
uninterrupted throughout a harmony span. Because the Gestural Harmony Model assumes that
harmony is strictly local and affects vowels and consonants alike, there is built into the theory a
restriction on which types of gestures may surface as persistent or anticipatory. In addition, this
local representation of harmony has important consequences for the representation of
transparency within the Gestural Harmony Model and the typological predictions the model
makes about transparent segments. This is discussed in detail in chapter 4.
The chapter is organized as follows. Section 2.2 introduces the proposed gestural
representation of harmony, which is illustrated with examples from rounding harmony, tongue
root harmony, and nasal harmony. It also introduces the proposed gestural parameters of
persistence (non-self-deactivation) and anticipation (early activation) as the drivers of harmony.
Section 2.3 focuses on a discussion of which types of gestures may surface as either persistent or
anticipatory, i.e., which gestures may trigger harmony. Section 2.4 summarizes the proposals
38
made in this chapter and previews the role they play in the development of the Gestural Harmony
Model’s phonological grammar in chapter 3.
2.2 Harmony as Gestural Extension
Harmony can be interpreted as the temporal extension of some phonological property that
is introduced by a triggering segment or morpheme. This section examines how that extension is
represented within the Gestural Harmony Model and which gestural parameters must be
manipulated in order to generate this extended period of gestural activation. These
representational concepts are illustrated by case studies of rounding harmony, tongue root
harmony, and nasal harmony.
2.2.1 Rounding Harmony in Kyrgyz
This section demonstrates the basic representational concepts of the Gestural Harmony
Model with an analysis of rounding harmony in Kyrgyz (Turkic, Kipchak; Kyrgyzstan). Kyrgyz
has a symmetrical surface inventory that distinguishes vowels based on height, backness,
rounding, and length (Comrie 1981), as shown in (23).
(23) Kyrgyz vowel inventory
6
Front Back
Unround Round Unround Round
High i iː y yː ɯ ɯː u uː
Non-High e eː ø øː a aː o oː
As in many Turkic languages, all vowels in Kyrgyz are contrastive for height while
harmonizing for backness and rounding. In this section, I focus on rounding harmony in Kyrgyz;
on the representation of vowel backness in gestural phonology, see section 6.2.1. According to
6
These vowel transcriptions vary somewhat from those used by Comrie (1981). The front rounded vowel
transcribed by Comrie as /ü/ and /ö/ are represented here by /y/ and /ø/, respectively. The back unrounded vowel
transcribed by Comrie as /ɨ/ is represented here by /ɯ/.
39
Comrie (1981), all round vowels in the inventory trigger rounding harmony, and all vowels are
capable of undergoing harmony. Roots surface with vowels that are either all round or all
unround and either all back or all front, and suffixes undergo harmony to match the rounding and
backness specifications of the root. As there are no prefixes in Kyrgyz, these harmonies are
strictly progressive (rightward). In (24a-d), the nonhigh vowel of the ablative suffix surfaces as
round following a root with round vowels, while following roots with unround vowels the same
suffix surfaces with an unround vowel, as in (24e-h). All data are from Comrie (1981).
(24) a. [yj-døn] ‘house (ablative)’ e. [iʃ-ten] ‘work (ablative)’
b. [køl-døn] ‘lake (ablative)’ f. [et-ten] ‘meat (ablative)’
c. [tuz-don] ‘salt (ablative)’ g. [d
͡ ʒɯl-dan] ‘year (ablative)’
d. [tokoj-don] ‘forest (ablative)’ h. [alma-dan] ‘apple (ablative)’
High vowels also undergo rounding harmony, as illustrated by the alternation of the
ordinal suffix in (25a-d).
(25) a. [yc-ynt
͡ ʃy] ‘three (ordinal)’ e. [bir-int
͡ ʃi] ‘one (ordinal)’
b. [tørt-ynt
͡ ʃy] ‘four (ordinal)’ f. [bes-int
͡ ʃi] ‘five (ordinal)’
c. [toguz-unt
͡ ʃu] ‘nine (ordinal)’ g. [altɯ-nt
͡ ʃɯ] ‘six (ordinal)’
d. [on-unt
͡ ʃu] ‘ten (ordinal)’ h. [d
͡ ʒijirma-nt
͡ ʃɯ] ‘twenty (ordinal)’
Comrie also provides several examples that illustrate that harmony also proceeds
throughout multiple suffixes, as in (26).
(26) a. [køz-yn-dø] ‘in his eye’
b. [tuz-un-do] ‘in his salt’
c. [ata-sɯn-da] ‘at his father’
d. [ene-sin-de] ‘at his mother’
As discussed in section 1.2.1, a round vowel is represented by two gestures, one with a
target of tongue body constriction and another with a target of lip protrusion. Based on
experimental findings reported by Boyce (1990), I propose that in a language without rounding
harmony, such as English, the two gestures of a round vowel begin and end at roughly the same
time. The lips return to their neutral position at the end of production of a round vowel.
40
However, in languages that exhibit progressive rounding harmony, such as Kyrgyz, I propose
that the lips do not return to this neutral position upon successful production of a round vowel.
Rather, the lips remain protruded for an extended period of time, resulting in the rounding of
following segments. In the Gestural Harmony Model, rounding harmony is represented as this
extended activation of a lip protrusion gesture that results in its overlap with and rounding of
following segments.
This representation of rounding harmony is demonstrated by the figure in (27), which
shows a gestural score for the Kyrgyz form [tuz-don] ‘salt (ablative).’ The [u] of the initial
syllable, the trigger of rounding harmony, is represented by two gestures: one for tongue body
constriction in the uvular region of the vocal tract, and one for protrusion of the lips.
7
Of note
here is the idea that the lip protrusion gesture (shaded) has an extended period of activation such
that it overlaps all following gestures. The dashed line within the lip protrusion gesture indicates
the time at which a typical lip protrusion gesture would deactivate simultaneously with the
deactivation of the uvular narrowing gesture of [u].
7
For now, it is assumed that the lip protrusion gesture is associated with the first vocalic segment in the word in
Kyrgyz. This is discussed further in section 3.2.1.
41
(27) Gestural score for [tuz-don] ‘salt (ablative)’
In the gestural score in (27), the segments overlapped by the lip protrusion gesture of [u]
are the undergoers of rounding harmony. This includes the suffix vowel, which surfaces as
rounded [o], as well as the consonants [z], [d], and [n]. By adopting the view that rounding
harmony is the result of the extended activation of a harmonizing lip protrusion gesture, the
Gestural Harmony Model contends that harmony is strictly local in the sense that the span of
activation of a harmonizing gesture is uninterrupted. A gesture with an extended period of
activation, such as the lip protrusion gesture in (27), will overlap any segments that come after it.
It makes no distinction between vowel harmony, in which only vowels are potential undergoers
and consonants are assumed to be transparent, and vowel-consonant harmony, in which all
segments are undergoers. In the Gestural Harmony Model, then, consonants are considered
undergoers of what is usually called vowel harmony because they are overlapped by a
harmonizing gesture that originates from a vocalic segment.
This is consistent with claims by Gafos (1996/1999) and Ní Chiosáin & Padgett (1997,
2001) that vowel harmony is local at the segmental level but does not affect consonants in a way
that is perceptible and/or contrastive. Gafos’ model of harmony, like the Gestural Harmony
Model, represents harmony as the extended, uninterrupted activation of a gesture, with
consonantal gestures being overlapped by vocalic gestures in vowel harmony environments.
42
Within feature-based phonology, Ní Chiosáin & Padgett also propose that consonants undergo
harmony, but that a consonant that bears vowel features such as [±round] or [±back] is not
perceptually distinct from a consonant that does not. Likewise, the strict locality inherent in the
Gestural Harmony Model leads to the characterization of consonants as undergoers of so-called
vowel harmony. The appearance of consonants as transparent to such harmonies is merely an
effect of perception; a consonant that has been overlapped by a harmonizing vocalic gesture is
not perceptually distinct from a consonant that has not been overlapped. This is referred to as
‘false transparency’ by Walker (1998/2000) and ‘apparent transparency’ by Archangeli &
Pulleyblank (2007). This can be contrasted with cases of true transparency, in which transparent
segments are produced as if they have not taken on a harmonizing property, discussed in detail in
chapter 4.
The representation of consonants as undergoers of so-called vowel harmony is in contrast
with autosegmental representations of vowel harmony in which only vowels are considered to be
targets of the harmony process. Clements & Sezer (1982) provide such an autosegmental
analysis of rounding harmony and backness harmony in Turkish. In their analysis, [±round] and
[±back] are the relevant harmonizing features, or p-segments, and vowels are the potential
targets, or p-bearing units. Under this analysis, consonants (with the exception of velar and
lateral consonants) are completely inert within the harmony system and are skipped by
autosegmental spreading. Thus, forms in Turkish are represented by Clements & Sezer as in
(28).
43
(28) Autosegmental representation of Turkish rounding harmony
a.
‘end (nom. sg.)’
b.
‘face (nom. sg.)’
However, the results of a number of theoretical and experimental studies support the view
of consonants as undergoers of vowel harmony. In his seminal study of consonant and vowel
coarticulation, Öhman (1966) argues that speech can be characterized as a series of vocalic
articulations over which consonants are superimposed, and that the vocal tract postures necessary
for vowel articulation are adopted by those superimposed consonants. Consonant gestures are
produced concurrently with the vocalic gestures that surround them; therefore, it makes sense
that consonants would also be undergoers of vowel harmony, albeit to a degree that is often not
perceptible. Additional studies have focused specifically on the articulation of consonants within
a harmony span and determined that consonants are affected by harmony processes. For instance,
Zhang (1996) reports that in Classical Manchu, dorsal consonants are produced as velars in ATR
harmonic words and as uvulars in non-ATR words. Similarly, Clements & Sezer (1982) report
that velar and lateral consonants participate in backness harmony in Turkish. Experimental work
by Boyce (1990) has also found that consonants produced within the domain of Turkish rounding
harmony are also produced as round, indicating a continuous ‘plateau’ of lip rounding. English
rounding, on the other hand, exhibits a ‘trough’-like pattern in which the degree of lip protrusion
alternates between consonants and vowels. Another language that has been the subject of this
44
line of study is Kalenjin. The tongue root harmony system of one of the varieties of this
language, Nandi, is the subject of section 2.2.3.
I turn now to the question of what generates the extended activation of the lip protrusion
gesture of round vowels in Kyrgyz and how it can be accounted for within the Gestural Harmony
Model. The literature is replete with analyses that drive harmony via OT constraints, including
featural alignment constraints (e.g., Kirchner 1993; Akinlabi 1994; Cole & Kisseberth 1994,
1995; Pulleyblank 1996), EXTEND(F) (Kaun 1995), SPREAD(F) (Padgett 1995; Walker
1998/2000), feature-driven markedness constraints (Beckman 1997, 1998), and AGREE (Baković
2000), to name only a few. Many of these constraints drive harmony by either directly or
indirectly motivating a harmonizing feature to maximize the number of segments to which it is
associated.
8
However, a parallel approach in which the time span of gestural activation is directly
manipulated by the grammar is not available within the Gestural Harmony Model. As discussed
in section 1.2.2, while temporal information that is present in a candidate form’s gestural score is
available to the phonological grammar, it cannot be manipulated directly. Instead, a candidate’s
gestural score is calculated by the Coupled Oscillator Model based solely on the content of its
coupling graph. Extended gestural activation cannot be achieved by positing a constraint that
requires a gesture to remain active for a certain period of time.
The Gestural Harmony Model instead appeals to a parameter of an extended gesture that
is present within the coupling graph and indicates that the gesture will extend throughout a word
once it has been passed through the Coupled Oscillator Model. Rather than directly requiring the
extended activation of a gesture, constraints in the grammar manipulate the setting of that
8
Baković’s AGREE is a notable exception to this generalization. As their name suggests, AGREE constraints require
segments to agree for some feature value, but are indifferent as to whether or not that agreement stems from the
association of those segments to the same harmonizing feature.
45
parameter. I propose that in the case of progressive harmony the crucial parameter within the
coupling graph is the (self-)deactivation parameter that determines the end phase of a gesture.
Typically, a gesture’s activation begins at its planning oscillator’s 0° phase and extends until it
reaches a specified phase at which it self-deactivates (corresponding to the achievement of that
gesture’s target articulatory state). However, I propose that the gestures that trigger progressive
(rightward) harmony are specified as persistent. A persistent gesture does not perform this self-
deactivation but instead remains active beyond the phase at which it achieves its target
articulatory state. In the case of Kyrgyz rounding harmony, it is the gesture calling for protrusion
of the lips that is persistent. The result is the extended activation of the lip protrusion gesture and
the progressive spread of rounding throughout a word, as in the gestural score in (27).
The figure in (29) contrasts the parameter specifications of a typical (self-deactivating)
gesture with those of a persistent (non-self-deactivating) gesture.
(29) Parameter specifications of a typical self-deactivating gesture and a persistent non-self-
deactivating gesture
I also propose that the workings of the Coupled Oscillator Model must be augmented in
order to accommodate these persistent (non-self-deactivating) gestures. As discussed in section
1.2.1, in its current instantiation the Coupled Oscillator Model comprises a single mechanism
that determines the relative phasing of gestures’ planning oscillators according to the coupling
relations that exist between them. Once the gestures in a phonological form are temporally
46
aligned, determining each gesture’s activation and deactivation time points requires no further
computation on the part of the model. Instead, a gesture automatically activates at its 0º phase
and deactivates at a specified phase corresponding to the achievement of its target articulatory
state. These typical gestures can be called self-activating and self-deactivating; the Coupled
Oscillator Model does not compute the activation and deactivation time points of these gestures,
but rather aligns them and then allows them to activate and deactivate themselves at their own
specified time points.
The proposed augmentation of the Coupled Oscillator Model involves the addition of a
second mechanism that is able to calculate the phases at which a gesture activates and
deactivates, based on its position within a coupling graph and its specified activation and
deactivation parameters. Gestures that are self-activating and self-deactivating do not rely on this
mechanism. However, rather than deactivating itself, a persistent gesture relies on this new
mechanism of the Coupled Oscillator Model to determine when it should deactivate. When the
model encounters a persistent gesture in a coupling graph, it will maintain that gesture’s
activation for as long as possible, i.e., until it reaches a blocking segment or a word boundary, at
which point the persistent gesture deactivates. It is also possible for a smaller domain boundary,
such as a morphemic or prosodic boundary, to deactivate a persistent gesture, though this will
not be explored in great detail here. The mechanism of blocking within the Gestural Harmony
Model is introduced in section 4.3.
The figure in (30) shows the gestures in (29) as they appear in a gestural score,
illustrating the differences in length of activation that result from the differences between the
settings of their self-deactivation/persistence parameters. A typical gesture is specified to
deactivate itself once it has reached its target articulatory state, indicated by the stop sign icon. A
47
persistent gesture, on the other hand, is specified not to self-deactivate when it reaches its target
articulatory state, indicated by the grayed out stop sign icon. This failure to self-deactivate results
in the gesture’s extended activation.
(30) A typical self-deactivating gesture and a persistent (non-self-deactivating) gesture
Most previous work within gestural phonology assumes that all gestures self-deactivate
without fail. A notable exception is found in work by Tejada (2012), who analyzes tone
spreading as the extended activation of tonal gestures. However, Tejada’s proposed grammar
does not evaluate the types of tone gestures (self-deactivating or persistent) but rather whether
those tone gestures are realized with an extended activation. That framework allows for a more
direct manipulation of a gesture’s period of activation within output candidates than what is
assumed by the Gestural Harmony Model. This difference will be discussed further in section
3.6.1.
Appealing to a gestural parameter for self-deactivation versus persistence allows for the
representation of gestures as either having a typical period of activation or the potential for
extended activation while still only manipulating a candidate output coupling graph, in which
information about the temporal extent of gestural activation is not present. When a coupling
graph with a persistent gesture is input to the augmented Coupled Oscillator Model, a gestural
48
score in which that gesture is active throughout an entire word will automatically result
(assuming there is no blocker present). Such a coupling graph, for the Kyrgyz form [tuz-don]
‘salt (ablative),’ is shown in (31). As in the figure in (30), the grayed out stop sign icon indicates
that the lip protrusion gesture is persistent (non-self-deactivating).
(31) Coupling graph for Kyrgyz [tuz-don] ‘salt (ablative)’
9
Note that the period of activation of the lip protrusion gesture for the [u], or any other
gesture for that matter, is not specified in this coupling graph. However, the coupling graph in
(31) will produce the gestural score for [tuz-don] ‘salt (ablative)’ in (27), with the period of
activation of the lip protrusion gesture spanning the entire word, due to that gesture’s not self-
deactivating the way other gestures in the word do.
In the Gestural Harmony Model, a gesture is a trigger of harmony due to its being
specified as persistent (non-self-deactivating). Whenever a round vowel is included in the
coupling graph for a Kyrgyz word, it will be represented by a tongue body gesture accompanied
by a persistent lip protrusion gesture. Therefore, the fact that Kyrgyz exhibits rounding harmony
is due not to the satisfaction of a harmony-driving constraint in its phonological grammar but to a
9
For reasons of space, this coupling graph does not include all gestures that are assumed to be included in the
representations of certain consonants, such as the velum closure gesture that is part of the representation of
obstruents. Such gestures are not crucial to the discussion here, but are the focus of section 4.4.
49
property of the lip protrusion gesture included in the round vowels in its surface phonological
inventory. The inventory of Kyrgyz vowels is represented gesturally as in (32).
(32) Gestural representation of Kyrgyz vowel inventory
/y/ /ø/ /u/ /o/
/i/ /e/ /ɯ/ /a/
Because the presence of harmony in a language is dependent upon whether the segments
in its surface phonological inventory include persistent gestures, in the Gestural Harmony Model,
harmony is driven indirectly via shaping of a language’s surface inventory. The result may be a
uniform inventory, with all gestures of a certain type (e.g., lip protrusion gestures) surfacing as
persistent, as is the case for round vowels in Kyrgyz. However, restrictions on which types of
segments may surface with persistent gestures (and in which contexts), are also common and
result in non-uniform inventories. The interactions of markedness and faithfulness constraints
that result in the shaping of a surface phonological inventory and the placing of distributional
restrictions on the members of that inventory are the subject of chapter 3.
2.2.2 Nasal Harmony in Capanahua
The case of Kyrgyz rounding harmony discussed in section 2.2.1 is a progressive
(rightward) harmony system, and is analyzed as the result of a persistent (non-self-deactivating)
gesture extending to overlap all following gestures in a word. The nasal harmony system of
50
Capanahua (Panoan; Peru; Loos (1967/1969), Safir (1982), Trigo (1988)), on the other hand, is
regressive (leftward), affecting segments that precede the trigger rather than those that follow it.
Loos reports the phonological inventory in (33) for Capanahua.
(33) Capanahua phonological inventory
10
Consonants Vowels
p t k ʔ i ɯ
ts tʃ a o
β s ʃ ʂ
m n
ɾ
j w
h
Nasal harmony in Capanahua is triggered by a nasal consonant and targets preceding
vowels, glides, and glottals, as in (34). All data is from Loos (1967/1969).
(34) a. [h̃ãmawɯ] ‘step on it’
b. [põȷ
̃ ãn] ‘arm’
c. [bãw
̃ĩn] ‘catfish’
d. [cĩʔ̃ĩn] ‘by fire’
e. [cipõnki] ‘downriver’
f. [wɯɾãnwɯ] ‘push it’
g. [wɯɾãnjasãʔ̃nwɯ] ‘push it sometime’
h. [bãnawɯ] ‘plant it’
When a liquid or obstruent precedes a harmony-triggering nasal consonant, it arrests the
spread of nasality, and any consonants and vowels preceding the blocking liquid or obstruent
surface as oral. This is illustrated by the forms in (34e-h). The analysis of blocking in Capanahua
nasal harmony is discussed in section 4.4.4.
In her analysis of tone spreading as the result of the extended activation of tone gestures,
Tejada (2012) claims that all apparently leftward tone spreading is the result of leftward
10
The vowel transcribed here as /ɯ/ is described by Loos as a high back unrounded vowel and transcribed as /ï/,
while Safir transcribes both high vowels as /i/.
51
movement of a tone gesture to a prominent position, accompanied by the rightward extension of
that tone gesture. Therefore, leftward harmony is predicted to always target a prominent position.
Such an analysis could be proposed for harmony systems in which the spread of a harmonizing
property always reaches the beginning of a word, no matter where in that word that property
originates. However, maintaining Tejada’s strict claim that all regressive harmony is the result of
the movement of a triggering gesture and subsequent progressive harmony does not appear to be
tenable. This is evidenced by Capanahua nasal harmony, in which nasality spreads regressively
but targets no particular position in a word. Many other regressive harmony systems exhibit a
similar lack of targeting of prominent positions. Instead, the Gestural Harmony Model must be
able to account for harmony systems in which a harmony triggering gesture extends to overlap
the gestures that precede it.
As discussed in section 1.2.1, a nasal segment is represented by two gestures, one with a
target oral constriction and another with a target of an open velum that will allow air to flow
through the nasal cavity. According to the Gestural Harmony Model, nasal harmony is
represented as the extended activation of a velum opening gesture, resulting in its overlap with
and nasalization of the segments around it. In the case of regressive harmony, it is the segments
preceding the trigger that are overlapped by the harmonizing gesture. This representation of nasal
harmony is demonstrated by the figure in (35), which shows a gestural score for the Capanahua
form [h̃ãmawɯ] ‘step on it.’ In this form, the velum opening gesture of the triggering [m] has an
extended activation such that it overlaps all preceding segments, resulting in their nasalization.
The dashed line indicates the time point at which a typical velum opening gesture would activate
simultaneously with the activation of the lip closure gesture of [m].
52
(35) Gestural score for Capanahua [h̃ãmawɯ] ‘step on it’
As discussed in section 2.2.1, the ability of a gesture to extend and overlap the gestures
that precede it must be represented in a phonological form’s coupling graph. In parallel with the
Gestural Harmony Model’s account of progressive (rightward) harmony, I propose that the
gestures that trigger regressive (leftward) harmony are specified as anticipatory. These
anticipatory gestures rely on the same new mechanism of the augmented Coupled Oscillator
Model (section 2.2.1) that is responsible for progressive (rightward) harmony. While a typical
gesture activates itself at the 0º phase of its planning oscillator, an anticipatory gesture can be
considered non-self-activating. If the Coupled Oscillator Model encounters such a gesture in a
coupling graph, it will calculate the gesture’s earliest possible activation point, beginning at its 0º
phase and moving regressively (leftward) throughout the word until it reaches a blocking
segment or a word boundary. The result of this early activation is the extended activation of a
gesture in the regressive (leftward) direction.
The figure in (36) contrasts the parameter specifications of a typical (self-activating)
velum opening gesture with those of an anticipatory, regressive harmony-triggering velum
opening gesture.
53
(36) Parameter specifications of a typical self-activating gesture and an anticipatory gesture
The figure in (37) shows these same gestures as they appear in a gestural score,
illustrating the differences in the time points of gestural activation and resulting periods of
activation that result from the differences between their activation parameter settings. A typical
gesture is specified to activate itself at the 0º phase of its planning oscillator, indicated by the
flag icon. An anticipatory gesture, on the other hand, is non-self-activating and must be activated
by the Coupled Oscillator Model, indicated by the grayed out flag icon. This gesture’s ability to
be activated earlier than its 0º phase results in the gesture’s extended activation and the potential
overlap of other segments in a word.
(37) A typical self-activating gesture and an anticipatory (early-activating) gesture
The alignment of the phases that correspond to a typical gesture’s activation and
deactivation are determined by that gesture’s position within a coupling graph. Like a typical
gesture, an anticipatory gesture is coupled with other gestures in a coupling graph according to
54
the phases of its planning oscillator. An anticipatory gesture that activates early relative to its
position in a word has not been repositioned within the coupling graph; it has dissociated the
time point at which it actually activates from the oscillatory phase at which a typical gesture
would activate. Unless an anticipatory gesture is also persistent (non-self-deactivating), the time
point of its deactivation is not altered by being anticipatory.
This is illustrated by the coupling graph for the Capanahua form [h̃ãmawɯ] ‘step on it’ in
(38). While the anticipatory velum opening gesture is word-medial in this coupling graph, in the
resulting gestural score in (35) this velum opening gesture extends to activate at the beginning of
the word.
(38) Coupling graph for Capanahua [h̃ãmawɯ] ‘step on it’
The presence of regressive nasal harmony in Capanahua can be attributed to the fact that
the velum opening gesture in its surface phonological inventory is one that is early-activating.
The nasal consonant inventory of Capanahua, then, is posited to contain harmony-triggering /m/
and /n/, as in (39).
55
(39) Gestural representation of Capanahua nasal consonant inventory
/m/ /n/
With the above inventory, whenever a nasal consonant occurs in a coupling graph in
Capanahua, it will be accompanied by an anticipatory (early-activating) velum opening gesture.
As a result, these segments will trigger regressive nasal harmony. There are also cases in which
nasal harmony in Capanahua is bidirectional, suggesting that the velum opening gesture in this
language can surface as simultaneously persistent and anticipatory; such cases will be discussed
in section 3.4.3.
2.2.3 Tongue Root Harmony in Nandi
Kyrgyz rounding harmony (section 2.2.1) and Capanahua nasal harmony (section 2.2.2)
illustrate cases of unidirectional harmony. Progressive harmony in Kyrgyz is triggered by a lip
protrusion gesture that surfaces as persistent (non-self-deactivating) but not anticipatory (early-
activating). Conversely, regressive harmony in Capanahua is triggered by a velum opening
gesture that surfaces as anticipatory but not persistent (though see section 3.4.3 for further
discussion). However, it is also possible for a gesture to surface as both persistent and
anticipatory and to therefore trigger bidirectional harmony. Such a case is illustrated in this
section with an examination of tongue root harmony in Nandi.
Vowel harmonies based upon tongue root position are common throughout the Tungusic
languages of Siberia and China as well as the Niger-Congo and Nilo-Saharan languages of
Africa. In these languages, vowels can be divided into two sets depending on whether they are
56
produced with an advanced tongue root (ATR) or not. This is the case in Nandi, a variety of
Kalenjin (Southern Nilotic; Kenya), which according to Creider & Creider (1989) has the vowel
inventory in (40). (Vowels also contrast for length.)
(40) Nandi (Kalenjin) vowel inventory
ATR non-ATR
Front Back Front Back
High i iː u uː ɪ ɪː ʊ ʊː
Mid e eː o oː ɛ ɛː ɔ ɔː
Low a aː ɑ ɑː
In Nandi, all vowels in a word must be either ATR or non-ATR. Nandi tongue root
harmony is an example of dominant-recessive ATR harmony; if an underlyingly ATR vowel
occurs anywhere in a word, either in a root or an affix, all other vowels surface as ATR as well.
This is illustrated by the data in (41)-(43). Nandi noun roots may be followed by one or more
thematic and number suffixes, many of which agree with roots for tongue root position. The
ATR alternant of a suffix appears after an ATR root (41a-e), while a non-ATR suffix appears
after a non-ATR root (41f-j). Data are from Creider & Creider (1989); roots are underlined.
(41) a. [suːme-eːk] ‘hair’ f. [mʊːsɑr-ɛːk] ‘porridge’
b. [oːrkaːj-jaːt] ‘prophet’ g. [lɔkɔ-jɑːt] ‘fruit’
c. [amt-it] ‘food’ h. [kɑːt-ɪt] ‘neck’
d. [ter-eːnik] ‘pots’ i. [cɛːrɛːr-ɛːnɪk] ‘babies, monkeys’
e. [kirk-it] ‘bull’ j. [kɪ-ɪt] ‘thing’
ATR harmony in Nandi may also be bidirectional, targeting both prefixes and suffixes, as
in (42a-c).
(42) a. [ka-keːr-aːt] ‘see (past amb.)’ d. [kɑ-kɑs-ɑːt] ‘hear (past amb.)’
b. [kaː-puːk-et] ‘sweeping’ e. [kɑː-pʊːs-ɛt] ‘breath’
c. [ka-keːr-eː] ‘see (past instr.)’ f. [kɑ-pɑl-ɛː] ‘dig (past instr.)’
Evidence that this harmony system is dominant-recessive rather than root-controlled
comes from forms in which affixes cause roots to alternate in terms of tongue root position. In
57
the data in (43), alternating roots and affixes are ATR when they occur with ATR affixes (43a-f),
and are RTR when they do not occur with such affixes (43g-l). This is true across the verbal,
nominal, and adjectival morphology.
(43) a. [kiː-laːl-ej] ‘cough (imperf.)’ g. [kɪː-lɑːl] ‘cough (perf.)’
b. [keː-pal-ci] ‘to dig (dat.)’ h. [kɛː-pɑl-ɛː] ‘to dig (instr.)’
c. [intar-o:k] ‘snakes’ i. [ɪntɑr-ɑ] ‘snake’
d. [kweːs-wek] ‘billy-goats’ j. [kwɛs-tɑ] ‘billy-goats’
e. [moː-oːk] ‘wounds’ k. [mɔː-ɛːt] ‘wound’
f. [tuː-in] ‘blackness’ l. [tʊːj] ‘black’
Finally, it is also possible for bidirectional harmony to be triggered by a prefix. Compare,
for example, [ka-ki-kas] ‘hear (past 1p),’ in which the fixed-ATR prefix [-ki-] triggers ATR
harmony, and [kɑ-kɑs] ‘hear (past 3p).’
Within gestural phonology, the two sets of ATR and non-ATR vowels in Nandi can be
divided based upon whether or not they are produced concurrently with a tongue root
advancement gesture. ATR vowels are accompanied by a tongue root advancement gesture,
while non-ATR vowels are not. During the production of non-ATR vowels, the tongue root
assumes its neutral position considerably farther back in the pharynx. The fact that Nandi
exhibits bidirectional tongue root harmony can be attributed to a tongue root advancement
gesture that is both persistent (non-self-deactivating) and anticipatory (early-activating) and that
accompanies all ATR vowels in its inventory. The full surface inventory of Nandi vowels is
illustrated in (44).
58
(44) Gestural representation of Nandi vowel inventory
/i/ /e/ /a/ /o/ /u/
/ɪ/ /ɛ/ /ɑ/ /ɔ/ /ʊ/
When an ATR vowel from this vowel inventory appears in an output coupling graph, it
serves as a trigger of bidirectional tongue root harmony. An illustration of this bidirectional
harmony is provided for the word [ka-ki-kas] ‘hear (past 1p).’ The prefix [-ki-] contains an
underlyingly ATR vowel, and thus all vowels in the word surface as ATR.
(45) Coupling graph for [ka-ki-kas] ‘hear (past 1p)’
In the coupling graph in (45), the tongue root advancement gesture is in word-medial
position. However, due to this gesture’s being both persistent (non-self-deactivating) and
59
anticipatory (early-activating), its activation will extend throughout the entire word. This is
illustrated in the gestural score in (46), which is calculated by the Coupled Oscillator Model from
the coupling graph in (45). The dashed lines within the tongue root gesture indicate the gesture’s
0º phase, i.e. the starting phase for a typical self-activating gesture, as well as the point at which
the gesture reaches its target articulatory state, i.e., the point at which a typical gesture would
self-deactivate.
(46) Gestural score for [ka-ki-kas] ‘hear (past 1p)’
In (46), all vocalic and consonantal gestures are overlapped by the harmonizing tongue
root advancement gesture. This results in all vowels in the word surfacing as ATR. In addition,
the strictly local representation of harmony within the Gestural Harmony Model results in the
treatment of consonants as undergoers of ATR harmony. Work by Lodge (1995) and Local &
Lodge (2004) on the status of Kalenjin consonants under the influence of tongue root harmony
supports this. They find a number of differences in the articulations of consonants in ATR versus
non-ATR words, particularly in terms of duration and degree of intervocalic lenition. This
provides support for the status of consonants as undergoers of vowel harmony in Nandi, rather
than as skipped segments.
60
2.3 Properties of Harmony Triggers
Section 2.2 introduced gestural persistence (non-self-deactivation) and anticipation (early
activation) as a means of representing the ability of a gesture to trigger harmony. This was
illustrated with examples of rounding harmony via a persistent lip protrusion gesture, nasal
harmony via an anticipatory velum opeing gesture, and tongue root harmony via a persistent and
anticipatory tongue root advancement gesture. In all of these cases, the harmonizing gesture
overlaps and affects consonantal and vocalic gestures alike.
However, this is not an exhaustive list of the attested types of harmony. Local harmonies
based on vowel height and backness, tongue dorsum and/or root retraction, and tone are also
attested, and thus the gestures associated with each of these properties may presumably be
persistent and/or anticipatory in some languages. Many other gestures, meanwhile, are not
attested triggers of local harmony. This section addresses why some types of gestures surface as
persistent or anticipatory, and which types of gestures can be triggers of harmony.
A great deal of work has focused on the idea that harmony is perceptually driven. Kaun
(1995, 2004), Flemming (1995), Walker (2005, 2011), and Kimper (2011), among others, argue
that harmony is driven by the desire to maximize the temporal extent of a perceptually
vulnerable feature, thus enhancing the contrast between its presence versus its absence. This
temporal maximization increases a listener’s exposure to this feature and therefore increases the
odds that a listener will correctly perceive it. The maximization of a gesture’s period of
activation is achieved in the Gestural Harmony Model via the gestural parameters of persistence
and anticipation. By surfacing as either persistent or anticipatory (or both), it is ensured that a
gesture will remain active for as long as possible within a word. According to this view, then, a
gesture is predicted to be more likely to trigger harmony if it is perceptually disadvantaged in
some way. In at least some languages, this perceptual disadvantage will motivate a gesture’s
61
surfacing as persistent and/or anticipatory. How this is implemented within the phonological
grammar is the subject of chapter 3.
It is important not just to determine which gestures are most strongly motivated to trigger
harmony but also which gestures are able to trigger harmony. If the extension of a harmonizing
property is a means of perceptual enhancement, then any gesture should be predicted to benefit
from being a trigger of harmony. While some gestures are at more of a perceptual disadvantage
than others, a perceptually enhanced gesture of any kind will always be preferred over its
unenhanced counterpart. By this logic, any gesture should be attested as a trigger of harmony in
some language. However, this is not the case; local harmonies triggered by primary consonantal
gestures and glottal gestures, for instance, are apparently unattested.
11
Recall from section 1.2.1
that primary consonantal gestures are those that are specified for either full closure of the vocal
tract or for constrictions that are narrow enough to result in turbulent airflow. It is necessary to
rule out the possibility of such gestures surfacing as persistent and/or anticipatory and therefore
serving as triggers of harmony.
The strictly local nature of harmony in the Gestural Harmony Model has important
consequences for which types of gestures are predicted to be able to surface as triggers. A
consonantal gesture cannot extend to overlap a vocalic gesture without significantly disrupting
the acoustic transmission of that vocalic gesture. If an alveolar tongue tip closure gesture of a
consonant such as /d/ were persistent or anticipatory, its overlap with any other gestures would
result in a closed vocal tract with a completely silent acoustic output. Under this scenario, this
11
While glottal gestures do not trigger local harmony, consonant harmony based on voicing, aspiration, and
glottalization are attested (see Hansson (2001/2010), Rose & Walker (2004, 2011), and Gallagher (2010) for
discussion). However, such cases are usually analyzed as cases of non-local agreement or assimilation rather than
local spreading of a harmonizing property. I adopt the assumption that such apparently non-local harmony is due to
a phonological mechanism that is distinct from the mechanism responsible for local harmony.
62
period of silence could potentially span multiple syllables and even an entire word. Obviously,
such a form is problematic for a number of reasons, chief among them that a word composed
entirely of a span of silence is unable to transmit any information about phonological contrast
and therefore has no communicative function.
This line of reasoning is promoted by Gafos (1996/1999) and Ní Chiosáin & Padgett
(1997) for explaining the asymmetry between consonants and vowels in their ability to trigger
local harmony. This asymmetry can be understood in terms of what Ní Chiosáin & Padgett call
the ‘bottleneck effect,’ the idea that a segment’s stricture is equal to its most constricted
component. According to this idea, features such as those for rounding, nasality, and vocalic
place are predicted to be able to spread onto a consonant without affecting its stricture or its
identity as a consonant. However, if consonantal place or stricture features were to spread onto a
vowel, the resulting segment would no longer be a vowel but rather some kind of segment
specified for full closure of the vocal tract. Gafos (1996/1999) and Ní Chiosáin & Padgett (1997)
maintain that such a scenario must be universally ruled out.
The asymmetrical triggering ability of consonantal and vocalic gestures can also be
understood in terms of ‘tube geometry,’ a more elaborate model of vocal tract stricture within
gestural phonology proposed by Browman & Goldstein (1989). According to this model, the
vocal tract is abstractly represented as a series of tubes and articulators that are either parallel to
or in sequence with one another. Each gesture is specified for a target articulatory state defined
within one of these tubes, and at any given time each tube has its own constriction degree that is
the result of any constrictions caused by active gestures. This abstract representation of the vocal
tract is shown in (47).
63
(47) Tube geometric representation of the vocal tract (adapted from Browman & Goldstein
(1989, p. 236)
In (47), the vocal tract branches into several tubes that run parallel to one another. Each
of the numbered black discs represents an articulator either within or at the terminus of a tube.
These articulators can create constrictions that determine a tube’s constriction degree. There are
three main tubes represented here: the Central, Lateral, and Nasal tubes. In addition, tubes and
articulators may be grouped together to form compound tubes. The Central and Lateral tubes run
parallel to one another, and together with the tongue tip and tongue body articulators make up
the Tongue tube. The Tongue tube runs in sequence with the Lips articulator, and together they
make up the Oral tube. The Nasal tube, which includes the velum articulator, runs in parallel
with the Oral tube; these tubes combine to form the Supralaryngeal tube. The Supralaryngeal
tube is in sequence with the Glottis articulator.
When two elements combine, the constriction degree of the compound tube is determined
by the constriction degree of each of its composite tubes or terminal articulators. When a
compound tube is made up of two elements in parallel, that compound tube takes on the wider of
64
the two constriction degrees. For instance, if the Nasal tube is open while the Oral tube is closed,
the constriction degree of the Supralaryngeal tube will be open. However, if a compound tube is
made up of two elements in a sequence, the compound tube takes on the narrower of the two
constriction degrees, consistent with Ní Chiosáin & Padgett’s bottleneck effect. For instance, if
the Tongue tube is open but the Lips articulator is closed, then the constriction degree of the Oral
tube is closed. By these principles, the constriction degree of the entire vocal tract at any given
time can be determined by the constriction degrees in each of the individual tubes. Browman &
Goldstein (1989) refer to this as percolation. The overall constriction degree of the vocal tract
takes on one of three values at any given time: open, closed, or critical (a constriction narrow
enough to produce turbulent noise). When multiple gestures are active at the same time, they will
each affect the overall constriction degree of the vocal tract according to the principles of tube
combination and percolation.
This principle of percolation within the tube geometry framework can be used to set a
baseline for determining whether or not a gesture may surface as persistent and/or anticipatory
and therefore whether it may trigger harmony. I propose that a gesture in a surface form may
only be persistent or anticipatory if it is able to overlap other gestures without (1) increasing the
constriction degree of the vocal tract to either closed or critical and/or (2) producing turbulent
noise. These conditions are met when a supralaryngeal gesture contributes an open constriction
degree to a tube. Gestures for lip protrusion and tongue root position each contribute an open
constriction degree, and may therefore overlap both consonantal and vocalic gestures without
affecting the constriction degree of the Oral and Supralaryngeal tubes. Therefore, these gestures
may be persistent and/or anticipatory. A velum opening gesture will always result in an open
Nasal tube and Supralaryngeal tube, no matter the constriction degree of the Oral tube. A velum
65
opening gesture is only able to decrease the constriction degree of the vocal tract, no matter what
it overlaps, and therefore it may extend its period of activation via persistence or anticipation.
On the other hand, the closure gesture of an oral consonant results in full closure in the
Oral tube via percolation from a closure in either the Central tube or at the Lips articulator. If this
consonantal closure gesture were to overlap a vocalic gesture, which is specified for an open
constriction in the Central tube, the result would be a closed constriction degree for the Central,
Oral, and Supralaryngeal tubes, and therefore for the entire vocal tract. In the case in which a
consonantal and a vocalic gesture are concurrently active, it is the consonantal gesture that
determines the constriction degree of the vocal tract. This scenario indicates the asymmetry
between the effects that consonants and vowels can have on each other’s constriction degrees. A
vocalic gesture is permitted to surface as persistent or anticipatory because it does not affect the
constriction degree of a concurrently active consonantal gesture. However, a consonantal gesture
that involves full closure or a critical constriction degree that causes turbulent airflow must
always surface as self-deactivating and self-activating because its overlap with other gestures
increases the constriction degree of the vocal tract.
In addition to oral consonantal gestures, glottal gestures are also ruled out as possible
triggers of local harmony. The glottis is in sequence with the Supralaryngeal tube; therefore,
closure at the glottis results in closure of the vocal tract via percolation. Glottal spreading
gestures that are responsible for voicelessness and/or aspiration are also ruled out as possible
triggers of local harmony, as they result in turbulent noise.
Based on the principles of the bottleneck effect and tube geometry, it is possible to rule
out glottal gestures and oral consonantal gestures as possible triggers of harmony. I assume that
persistent and/or anticipatory consonantal gestures and glottal gestures are universally banned
66
from the set of forms that the phonological grammar can produce. In an OT grammar, this is akin
to stating that GEN does not produce candidate forms that include persistent or anticipatory
consonantal and glottal gestures. Other types of gestures may surface with either setting of the
persistence and anticipation parameters based on the ranking of constraints that make reference
to these parameters. This is the subject of chapter 3.
Alternatively, it may be possible to explain the asymmetry between gestures that may or
may not serve as triggers of harmony by appealing to the split gesture hypothesis (Nam 2007).
Nam proposes that each consonantal constriction should be represented by two gestures, one for
the closure and one for the release of that closure. The effect of this release gesture is a rapid and
controlled movement of an articulator away from the closure gesture’s target articulatory state
and toward the articulator’s neutral default state. The representation of a split consonantal
gesture is provided in (48).
(48) Closure and release gestures
Nam provides evidence from patterns of coordination between an onset consonant and a
following vowel as well as relative speech planning times across the production of different
syllable shapes to support the split gesture representation of consonantal constrictions. Based on
his findings, he proposes that release gestures are unique to the representation of consonants, and
that all consonants should be represented by a closure-release sequence of gestures. Vocalic
gestures are assumed not to be accompanied by such release gestures; instead, when a vocalic
gesture deactivates, the articulators involved in the attainment of that gesture’s target articulatory
state return to their default positions gradually. In the Gestural Harmony Model, the inability of a
67
consonantal gesture to extend its period of activation and trigger harmony could be attributed to
its always being accompanied by a following release gesture. Even if a consonantal closure
gesture were to surface as persistent (non-self-deactivating), the following release gesture would
deactivate the closure gesture upon the attainment of the gesture’s target articulatory state.
While appealing to an independently motivated aspect of gestural representation such as
the split consonantal gesture in order to capture the asymmetry between triggering and non-
triggering gestures is appealing, there are issues with this approach. Chief among these is a lack
of clarity as to how to use split gesture representations to prevent consonantal gestures from
surfacing as anticipatory and triggering regressive (leftward) harmony. At present, there is no
equivalent of a release gesture that immediately precedes a consonantal gesture and prevents it
from activating early. Without such a gesture, there does not appear to be any way to prevent a
consonantal gesture from triggering regressive harmony based solely on the split gesture
hypothesis. On the other hand, appealing to tube geometry and the bottleneck effect comes with
no such directional asymmetry. Gestures are ruled out as potential triggers of harmony based on
their aerodynamic and acoustic effects on the vocal tract, and these effects hold whether a
gesture extends its activation regressively (leftward) or progressively (rightward).
In addition, it is unclear what predictions the split gesture hypothesis makes regarding the
ability of glottal gestures to act as triggers of local harmony. Glottal gestures often surface as
secondary gestures that are responsible for the voicelessness or glottalization of an oral
consonantal segment. The split gesture, on the other hand, appears to be exclusive to the
representation of primary consonantal gestures. Without accompanying release gestures, these
glottal gestures would not be ruled out as possible triggers of local harmony. This result is
undesirable. By appealing to tube geometry, however, glottal gestures are ruled out as possible
68
triggers of local harmony due to their aerodynamic and acoustic effects. An active glottal closure
gesture produces closure of the entire vocal tract, while an active glottal spreading gesture results
in turbulent noise. This renders both of these types of gestures unable to surface as persistent or
as anticipatory according to the conditions on triggers proposed in this section. Thus, appealing
to tube geometry correctly predicts that glottal gestures are not triggers of any local harmony
processes. However, despite these issues, the split gesture hypothesis offers an intriguing
explanation of the asymmetry in harmony triggering ability between consonantal and vocalic
gestures. Examination of the predictions made by appealing to split consonantal gestures is a
subject worthy of further study.
2.4 Summary
This chapter has introduced the basic representation of harmony within the Gestural
Harmony Model as overlap by a gesture with extended activation. The ability of a gesture to
surface with this extended activation is the result of two new gestural parameters that are central
to the operation of the Gestural Harmony Model. One of these parameters, persistence,
determines whether a gesture self-deactivates upon the achievement of its target articulatory
state, or whether it is a persistent gesture that remains active and triggers progressive harmony.
The other gestural parameter, anticipation, determines whether a gesture activates at the 0º phase
of its planning oscillator, or whether it activates before this phase and triggers regressive
harmony. Both of these parameters necessitate a new mechanism that operates within the
Coupled Oscillator Model. This mechanism calculates the activation and de-activation time
points of anticipatory and persistent gestures, respectively.
As a direct consequence of appealing to the gestural parameters of persistence and
anticipation in the Gestural Harmony Model, a segment’s status as a trigger of harmony is
69
directly encoded in its gestural representation. If a gesture is persistent, it will trigger progressive
harmony; if it is anticipatory, it will trigger regressive harmony. Under this approach, whether a
language exhibits harmony is a property of its surface phonological inventory and whether it
contains segments that include either persistent or anticipatory gestures. Accounting for patterns
of harmony triggering in the Gestural Harmony Model, then, involves developing a phonological
grammar that shapes a language’s inventory and places restrictions on the distributions of the
members of that inventory. This is the subject of chapter 3.
70
Chapter 3
Patterns of Harmony Triggering
3.1 Introduction
As discussed in chapter 2, the presence of harmony in Kyrgyz, Nandi, and Capanahua is
attributed in the Gestural Harmony Model to the idea that these languages include either
persistent (non-self-deactivating gestures) or anticipatory (early-activating) gestures in their
respective surface phonological inventories. This approach to harmony as arising from a property
of a trigger provides an interesting contrast with previous analyses of harmony processes within
feature-based phonological frameworks. In featural analyses, harmony is often analyzed as being
driven by a rule or constraint that requires the addition of association lines between features and
segments. Under this approach, the imperative to spread a feature value comes directly from the
grammar rather than resulting automatically from a property of the trigger itself. In contrast, the
Gestural Harmony Model relies on the gestural parameters of persistence and anticipation to
account for a gesture’s ability to extend its period of activation and trigger harmony either
regressively or progressively.
In the harmony systems presented in chapter 2, the patterns of harmony triggering are
uncomplicated; any segment bearing a harmonizing property serves as a trigger of harmony. In
Kyrgyz, lip protrusion gestures always surface as persistent, and as a result round vowels in this
language always trigger progressive (rightward) rounding harmony. In Nandi, any ATR vowel
will trigger bidirectional ATR harmony due to the inclusion in its gestural makeup of a tongue
root advancement gesture that is both persistent and anticipatory. In Capanahua, the nasal stops
in its inventory include anticipatory velum opening gestures, rendering them triggers of
71
regressive (leftward) nasal harmony. In each of these languages, the ability of all segments of a
certain type to trigger harmony is the result of the language’s surface inventory containing only a
persistent and/or anticipatory gesture of a certain type.
However, there are many harmony systems with more complicated patterns of harmony
triggering, and a successful model of harmony must be able to account for several sources of
these complications. For instance, many languages restrict harmony triggers to certain prominent
positions in a word, such as the initial syllable or the stressed syllable. Other languages place
conditions on the identities of triggers; in particular, they tend to favor triggering by segments
that are perceptually disadvantaged in some way. In addition, some harmony systems involve the
idiosyncratic triggering of harmony. While these more complicated patterns of harmony
triggering bring up issues with many feature-based analyses of harmony, the Gestural Harmony
Model is able to successfully account for all of these complications. In this model, patterns of
harmony triggering arise directly from the shape of a language’s surface phonological inventory,
and the distributional restrictions the phonological grammar places on the members of that
inventory.
The Gestural Harmony Model must account for systems in which restrictions are placed
on the triggers of harmony, in terms of both the position of these triggers within a word and the
quality of the triggering segments. In this model, harmony is triggered by the presence of either a
persistent or an anticipatory gesture in an output phonological form. Because of this, it is
straightforward for the Gestural Harmony Model to capture attested, often seemingly complex
patterns of harmony triggering via co-occurrence constraints that require certain segment types to
be accompanied by gestures of a certain type (typical, persistent, or anticipatory), or licensing
constraints that require these gestural types to appear in certain positions in a word.
72
The Gestural Harmony Model must also be able to account for cases of idiosyncratic
triggering of harmony, in which some forms exhibit harmony while others do not. There are
numerous examples of harmony systems in which some words idiosyncratically exhibit harmony
while others do not. Such patterns are attested in the nasal harmony systems of several Malayo-
Polynesian languages, as well as various tongue root harmony systems. As demonstrated in
section 2.2, in the Gestural Harmony Model the ability of a gesture to trigger harmony is the
result of a specified gestural parameter (persistence in the case of progressive harmony, and
anticipation in the case of regressive harmony). These parameter settings serve a contrastive
function in some languages, and the result of such contrast is the idiosyncratic triggering of
harmony.
This approach to harmony triggering, in which the Gestural Harmony Model relies on a
parameter of a phonological unit to drive harmony, can be contrasted with feature-based analyses
of harmony that do not have access to such parameters and instead rely on rules or constraints
that explicitly drive harmony. While such analyses are able to account for harmony systems that
place conditions on harmony triggers, they do not perform as well in accounting for patterns of
idiosyncratic or contrastive triggering, often over- and/or undergenerating attested patterns.
Instead of relying on a contrastive property, such as persistence versus self-deactivation, to
account for such patterns, featural analyses of harmony triggering typically rely on theoretical
mechanisms designed for dealing with phonological exceptionality. Such mechanisms produce
several undesirable typological predictions that are avoided by the Gestural Harmony Model and
its reliance on gestural parameter settings, rather than grammatical rules or constraints, to drive
harmony.
73
The chapter is organized as follows. Section 3.2 discusses the constraint set that is used to
drive harmony via the shaping of a language’s surface inventory to either include or exclude
harmony-triggering persistent or anticipatory gestures. The following sections present
typological patterns of harmony triggering and present analyses of systems in which harmony
triggering is contrastive, as in the nasal harmonies of Acehnese and Rejang (section 3.3) and
systems in which conditions are imposed on the triggers of harmony, as in Baiyina Oroqen
rounding harmony (section 3.4.2). Section 3.5 demonstrates the ability of the Gestural Harmony
Model to produce even quite complex patterns of harmony triggering by examining the case of
Classical Manchu, which exhibits both conditional and contrastive triggering of tongue root
harmony. Section 3.6 examines some of the issues that beset feature-based analyses of complex
harmony-triggering patterns. The strengths of the Gestural Harmony Model in accounting for
patterns are harmony triggering are summarized in section 3.7. An appendix with the definitions
of all of the constraints used throughout is included at the end of the chapter.
3.2 Harmony Triggers: Inventory Shaping & Distributional Restrictions
Within the Gestural Harmony Model, harmony within a language is not driven directly by
the grammar but by a property of a certain gesture in that language. When a persistent (non-self-
deactivating) gesture occurs in a language’s surface phonological inventory, it will serve as a
trigger of progressive harmony. Likewise, when an anticipatory (early-activating) gesture occurs
in a language’s inventory, it will serve as a trigger of regressive harmony. It is necessary, then,
for a phonological grammar to shape the surface inventory of a language with harmony such that
it includes segments with persistent and/or anticipatory gestures. According to the principle of
Richness of the Base (Prince & Smolensky 1993/2004), neither type of gesture may be excluded
from the set of possible inputs in any language. It is up to the interaction of markedness and
74
faithfulness constraints to shape the surface inventories and distributional restrictions of
languages such that their harmony-triggering patterns are accurately generated by the grammar.
This inventory shaping is the subject of this section, with illustrations of the basic markedness
and faithfulness constraints necessary to capture basic patterns of harmony triggering in the
Gestural Harmony Model.
3.2.1 Revisiting Kyrgyz: Progressive Rounding Harmony
Recall that in Kyrgyz, all vowels harmonize for backness and rounding, and all round
vowels are triggers of rounding harmony (Comrie 1981). In the Gestural Harmony Model, this is
accounted for by positing a phonological inventory in which all round vowels are accompanied
by persistent (non-self-deactivating) lip protrusion gestures, as in the inventory in (49), repeated
from (32) in section 2.2.1.
(49) Gestural representation of Kyrgyz vowel inventory
/y/ /ø/ /u/ /o/
/i/ /e/ /ɯ/ /a/
The grammar of a language like Kyrgyz, in which all round vowels include persistent lip
protrusion gestures on the surface, must ensure that even an underlying self-deactivating lip
protrusion gesture will surface as persistent. This can be achieved by ranking a markedness
constraint that penalizes a self-deactivating lip protrusion gesture over a faithfulness constraint
75
that requires a gesture not to alter its specified deactivation parameter between the input and the
output. This markedness constraint, PERSIST(Gest
X
), is defined following the schema in (50).
(50) PERSIST(Gest
X
): Assign a violation mark to a gesture of type X that is self-deactivating.
Note that this constraint is satisfied whenever a gesture of type X surfaces as persistent
(non-self-deactivating), regardless of whether or not it actually overlaps any following segments.
PERSIST(GestX) evaluates only the setting of a gesture’s deactivation parameter, and not whether
that gesture’s activation actually extends beyond the timepoint at which it achieves its target
articulatory state.
In Kyrgyz, the constraint PERSIST(lip protrusion) penalizes a lip protrusion gesture that
self-deactivates as soon as its target articulatory state (fully protruded lips) is achieved. As
discussed in section 2.3, this constraint’s phonetic grounding is rooted in the claims made by
Kaun (1995, 2004), Flemming (1995), Walker (2005, 2011), Kimper (2011), and others that
temporally extending a feature or a gesture is perceptually advantageous. The non-self-
deactivating nature of a persistent gesture, enforced by PERSIST(Gest
X
), ensures that a gesture
will remain active for as long as possible, deactivating only when it reaches the end of a domain
or a blocking segment (discussed in detail in chapter 4). In doing so, that gesture’s chance of
being perceived correctly increases relative to that of its self-deactivating counterpart.
One crucial distinction in the analysis of harmony within the Gestural Harmony Model
versus those within many featural frameworks lies in how the phonological grammar penalizes
harmony. Many featural analyses of harmony rely on the ranking of a harmony driver over a
faithfulness constraint such as IDENT-IO that assigns violations for any changes in undergoers’
specifications for a harmonizing feature. In contrast, in the Gestural Harmony Model the overlap
of an undergoer segment by a harmonizing gesture does not incur any faithfulness violations. As
76
discussed in section 1.2.2, the Gestural Harmony Model assumes the faithfulness constraints
IDENT(parameter
X
)-IO constraints, which penalize changes in a gesture’s setting for parameter X,
and DEP(gesture
X
)-IO, which penalizes the addition of a gesture of type X to a segment. Neither
of these types of constraints is violated by the undergoers of harmony. DEP(gesture
X
)-IO is not
violated when a segment is overlapped by a harmonizing gesture, because being overlapped by a
gesture does not mean that that gesture is considered to have joined that segment.
IDENT(parameter
X
)-IO, meanwhile, is also not violated by gestural overlap; while this overlap
may alter how a segment is produced, it does not alter any parameters of the gestures of a
segment.
I instead attribute the lack of harmony in a language to the constraint SELFDEACTIVATE.
This constraint penalizes persistent gestures, thus preventing harmony from arising in a
language. In contrast with constraints from the PERSIST(Gest
X
) family, each of which refers to a
specific type of gesture, SELFDEACTIVATE penalizes persistent gestures of all types. This
constraint is defined in (51).
(51) SELFDEACTIVATE: Assign a violation mark to a gesture that is not self-deactivating.
Both PERSIST(Gest
X
) and SELFDEACTIVATE, markedness constraints that each penalize a
specific gestural parameter setting in the output form, potentially conflict with the faithfulness
constraint IDENT(deactivation)-IO, which preserves gestures’ underlying deactivation parameter
settings. It is defined in (52); note that it is defined with respect to the potential trigger of
harmony, and not any potential undergoers.
77
(52) IDENT(deactivation)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical deactivation specifications.
The relative ranking of the constraints in (50)-(52) determines whether or not a language
exhibits harmony of a certain type (e.g., rounding harmony). When SELFDEACTIVATE is ranked
above IDENT(deactivation)-IO and PERSIST(lip protrusion), the result is a language with no
rounding harmony, such as English, in which all lip protrusion gestures surface as typical self-
deactivating gestures. When PERSIST(lip protrusion) is ranked above IDENT(deactivation)-IO and
SELFDEACTIVATE, round vowels in that language will always trigger rounding harmony, whether
they are underlyingly specified as persistent or self-deactivating. All possible input gestures will
neutralize to the persistent type. This is the case in Kyrgyz, as the tableaux in (53) and (54)
illustrate for the form [tuz-don] ‘salt (ablative).’
In order to illustrate the effect of a deactivation parameter setting on the ability of a
gesture to extend its period of activation, these tableaux contain both the candidate forms’
coupling graphs and the gestural scores that are calculated by the Coupled Oscillator Model from
those graphs. It is important to remember that while the temporal information in a gestural score
is visible to the OT grammar assumed by the Gestural Harmony Model, this grammar still
operates only on the material that is present in a coupling graph. For reasons of space and clarity,
only the vowel gestures are included in the following tableaux; however, this should not be taken
to mean that consonants are not undergoers of harmony, as discussed in section 2.2. Lip
protrusion gestures in output candidates are shaded for emphasis.
78
(53) Tableau for Kyrgyz [tuz-don] ‘salt (ablative)’ with underlying self-deactivating lip
protrusion gesture
Input: / t u
1
z - d a
2
n /
PERSIST(lip protrusion)
SELFDEACTIVATE
IDENT(deactivation)-IO
a. [tuz-dan]
*!
F b. [tuz-don]
* *
The input in (53) contains a typical self-deactivating lip protrusion gesture (indicated by
the stop sign) accompanying the tongue body gesture of /u/. In output candidate (a) [tuz-dan] the
Tongue Body
uvular narrow
1
Tongue Body
pharyngeal wide
2
Lip
protrusion
1
79
lip protrusion gesture is also self-deactivating; this satisfies IDENT(deactivation)-IO, which
requires the deactivation specifications of a gesture to be identical between the input and the
output. However, this violates higher-ranked PERSIST(lip protrusion), which penalizes the ability
of a gesture to self-deactivate. The winning candidate (b) [tuz-don] contains a persistent lip
protrusion gesture, violating IDENT(deactivation)-IO but satisfying higher-ranked PERSIST(lip
protrusion). Note that in this candidate the representation of the second vowel is not altered,
despite having undergone harmony. The persistent lip protrusion gesture overlaps that vocalic
segment, but is not associated (i.e., coupled) with it. The lip protrusion gesture remains a part of
the first vocalic segment (set of gestures), and has not in any way joined the second vocalic
segment.
80
(54) Tableau for Kyrgyz [tuz-don] ‘salt (ablative)’ with underlying persistent lip protrusion
gesture
Input: / t
1
u
2
z
3
- d
4
a
5
n
6
/
PERSIST(lip protrusion)
SELFDEACTIVATE
IDENT(deactivation)-IO
a. [tuz-dan]
*! *
F b. [tuz-don]
*
The tableau in (54) uses the same candidate set as in (53), but here the input /u/ is
accompanied by a persistent lip protrusion gesture (indicated by the grayed out stop sign).
Tongue Body
uvular narrow
1
Tongue Body
pharyngeal wide
2
Lip
protrusion
1
81
Candidate (a) [tuz-dan] violates both PERSIST(lip protrusion) and IDENT(deactivation) by altering
the deactivation parameter of the velum opening gesture from persistent to self-deactivating.
Again, the winner is candidate (b) [tuz-don], in which the lip protrusion gesture is persistent in
both the input and the output, thereby violating neither constraint.
One final point must be accounted for in this analysis: according to Comrie (1981), in all
Kyrgyz words, the vowel of the first syllable determines whether all vowels in the word are
round or unround.
12
It is impossible for a Kyrgyz word to contain disharmonic vowel sequences
such as [o-a-a] or [a-o-o]. The first of these sequences, [o-a-a], is ruled out in the Gestural
Harmony Model by representing the triggering (initial) [o] as containing a persistent lip
protrusion gesture. However, the second of these sequences, [a-o-o], cannot be ruled out by the
current constraint set alone. In the Gestural Harmony Model, such a form could be represented
by a coupling graph and gestural score in which the vowel of the second syllable of a word
contains a persistent lip protrusion gesture and therefore acts as a trigger of rounding harmony.
Such forms must be ruled out in Kyrgyz and many other Altaic languages. It appears to be
necessary not only to restrict the inventory of Kyrgyz such that all round vowels are harmony
triggers, but to restrict the distributions of those round vowels as well.
In the Gestural Harmony Model, the distribution of round vowels in Kyrgyz can be
accounted for by restricting lip protrusion gestures to the initial syllable of the word via a
positional LICENSE constraint (section 1.2.2). This constraint is defined in (55).
12
This vowel also determines whether all vowels in a word are front or back; this is set aside here, though see
section 6.2.1 for a discussion of how to represent vowel height and backness, as well as harmonies involving these
properties, in gestural phonology.
82
(55) LICENSE(lip protrusion, first σ): Assign a violation mark to a lip protrusion gesture that is
not in an initial syllable.
In Kyrgyz, the licensing constraint in (55) will ensure that round vowels are restricted to
the privileged position of the initial syllable.
13
When a lip protrusion gesture appears in a non-
initial syllable in an input in Kyrgyz, the high ranking of LICENSE(lip protrusion, first σ) will
ensure that this lip protrusion gesture is either reordered such that it is coupled to the tongue
body gesture of the vowel in the initial syllable, or that it is deleted. In this analysis I assume the
latter; any underlying lip protrusion gesture that is introduced by a vowel outside of the initial
syllable will be deleted rather than moving to the initial syllable. This ensures that suffixes in
Kyrgyz will never introduce their own rounding specifications to a word. In order to favor the
deletion of an unlicensed lip protrusion gesture over its movement to the initial syllable, the
constraint INTEGRITY-IO (section 1.2.2), which penalizes the splitting of gestures that make up a
single segment between the input and the output, must be ranked over a constraint from the
MAX(gesture)-IO family, which penalizes the deletion of gestures between the input and the
output. This constraint is defined in (56).
(56) MAX(lip protrusion)-IO: Assign a violation mark to a segment (set of gestures) that
includes a lip protrusion gesture in the input if its output correspondent does not include
that gesture.
The tableau in (57) demonstrates the result of this ranking for the Kyrgyz word
[alma-dan] ‘apple (ablative)’ with a hypothetical input in which the vowel of the second syllable
includes a lip protrusion gesture. All lip protrusion gestures in the output candidates are assumed
to be persistent, satisfying PERSIST(lip protrusion). Again, for reasons of space, only the vowel
13
Walker (2011) notes that restricting round vowels to the initial syllable of a word can operate independent of a
harmony process, citing the case of Ola Lamut, in which there is no active process of rounding harmony but there is
an active restriction of round vowels to an initial syllable.
83
gestures of the word are included in the candidate coupling graphs and the gestural scores that
are calculated from them.
84
(57) Tableau for Kyrgyz [alma-dan] ‘apple (ablative)’ with hypothetical underlying lip
protrusion gesture in non-initial syllable
Input: / a
1
l m o
2
- d a
3
n /
LICENSE(LP, first σ)
INTEGRITY-IO
MAX(LP)-IO
a. [almo-don]
*!
b. [olmo-don]
*!
F c. [alma-dan]
*
Tongue Body
pharyngeal wide
1
Tongue Body
pharyngeal wide
2
Tongue Body
pharyngeal wide
3
Lip
protrusion
2
85
In (57), the hypothetical input /almo-dan/ includes a round vowel in the second syllable.
In output candidate (a) [almo-don] the gestures of underlying /o
2
/ are coupled to one another,
satisfying INTEGRITY-IO in addition to MAX(lip protrusion)-IO. However, this candidate fatally
violates higher-ranked LICENSE(lip protrusion, first σ), as the lip protrusion gesture of [o
2
] is not
in the first syllable of the word. Candidate (b) [olmo-don] eliminates this violation of
LICENSE(lip protrusion, first σ) by coupling the lip protrusion gesture to the tongue body gesture
of the vowel in the initial syllable. This segmental reassociation of the lip protrusion gesture
earns a fatal violation of INTEGRITY-IO. In the winning candidate (c) [alma-dan] the lip
protrusion gesture has been deleted, thereby vacuously satisfying LICENSE(lip protrusion, first σ)
and INTEGRITY-IO but violating low-ranked MAX(lip protrusion)-IO.
This analysis accounts for the presence of harmony in Kyrgyz, as well as the restriction
on the distribution of harmony-triggering round vowels to the initial syllable of a word. That the
position of a gesture is determined by its position in the coupling graph accounts for the fact that
non-initial round vowels may occur in a word provided that they are undergoers of harmony.
These undergoer vowel segments surface as round because they are overlapped by the lip
protrusion gesture of another vowel segment, and not because they include their own lip
protrusion gestures. As the tableau in (57) shows, an underlying round vowel in a non-initial
syllable will surface without its lip protrusion gesture. Such vowels may only surface as round
when overlapped by the lip protrusion gesture of a triggering round vowel in an initial syllable.
The constraint ranking necessary to capture the patterns of harmony triggering in Kyrgyz
is summarized in the Hasse diagrams in (58).
86
(58) Constraint ranking for Kyrgyz rounding harmony
a. Triggering by round vowels b. Contrastively round vowels only in first
syllable
The relative ranking of the constraint types introduced in this section are able to account
for additional patterns of harmony triggering via restrictions on which gestures may appear in a
language’s surface inventory, and on the positional distributions of the members of that
inventory. One such additional pattern is exemplified by the ATR harmony of Nandi introduced
in section 2.2.3. The following subsection revisits this harmony system and provides an analysis
of its pattern of harmony triggering.
3.2.2 Revisiting Nandi: Dominant-Recessive Tongue Root Harmony
As discussed in section 2.2.3, Creider & Creider (1989) report that Nandi has a
bidirectional dominant-recessive tongue root harmony system. If an underlyingly ATR vowel
occurs anywhere in the word, the entire word takes on that ATR quality. Recall that in Nandi
tongue root harmony is bidirectional, and may be triggered by ATR vowels in roots, prefixes,
and suffixes. Compare, for example, [ka-keːr-aːt] ‘see (past amb.),’ in which the ATR root vowel
[eː] triggers harmony that affects the vowels of surrounding affixes, and [kɑ-kɑs-ɑːt] ‘hear (past
amb.).’ As an example of harmony triggered by an affix, compare [ka-ki-kas] ‘hear (past 1p),’ in
which the underlyingly ATR /i/ of prefix [-ki-] triggers harmony, and [kɑ-kɑs] ‘hear (past 3p).’
In the Gestural Harmony Model, the bidirectional ATR harmony seen in Nandi is
attributed to the fact that the ATR vowels in the language’s surface inventory all include tongue
root advancement gestures that are both persistent and anticipatory. I attribute the surfacing of
87
each of these gestural parameter settings to two distinct sets of constraints. The set of constraints
that reference gestural persistence and self-deactivation were already introduced in the analysis
of Kyrgyz rounding harmony in section 3.2.1. The same basic constraint ranking of
PERSIST(Gest
X
), SELFDEACTIVATE, and IDENT(deactivation)-IO used to shape the Kyrgyz vowel
inventory can be used to ensure that tongue root gestures in Nandi trigger progressive harmony.
The relevant constraint for the progressive component of harmony is PERSIST(tongue root
advancement), which penalizes self-deactivating tongue root advancement gestures. The
constraint ranking in (59) ensures that an ATR vowel will always surface as persistent and will
therefore trigger progressive ATR harmony in Nandi.
(59) Constraint ranking for progressive ATR harmony in Nandi
PERSIST(tongue root advancement) >> SELFDEACTIVATE, IDENT(deactivation)-IO
The grammar must also ensure that tongue root advancement gestures in Nandi always
surface as anticipatory in order to account for the regressive (leftward) component of the ATR
harmony system. In the Gestural Harmony Model, this is achieved by a set of constraints that is
distinct from those responsible for progressive harmony. Paralleling the approach to progressive
harmony triggering, regressive harmony can be achieved via the high ranking of a markedness
constraint that penalizes a non-anticipatory tongue root advancement gesture over a faithfulness
constraint that requires a gesture not to alter its activation parameter between the input and the
output. This markedness constraint, ANTICIPATE(Gest
X
), is defined following the schema in (60).
(60) ANTICIPATE(Gest
X
): Assign a violation mark to a gesture of type X that is non-
anticipatory.
Like PERSIST(Gest
X
), the presence of ANTICIPATE(Gest
X
) in the constraint set is rooted in
the claim that temporally extending some properties is perceptually advantageous. Both
88
constraints motivate the extension of gestural activation, albeit in different directions. Also like
PERSIST(GestX), ANTICIPATE(GestX) is satisfied not based on whether a gesture actually extends
its period of activation, but instead on whether that gesture is specified for a certain setting of a
gestural parameter.
As discussed in section 3.2.1, the overlap of an undergoer segment by a harmonizing
gesture does not incur any violations of faithfulness. This is true of both progressive harmony via
overlap by a persistent gesture, and regressive harmony via overlap by an anticipatory gesture.
Therefore, another markedness constraint penalizing anticipatory gestures (and therefore
regressive harmony) is necessary to account for languages without regressive harmony. This
constraint, SELFACTIVATE, is defined in (61).
(61) SELFACTIVATE: Assign a violation mark to a gesture that is anticipatory (a gesture that
activates before its 0º phase).
The constraints ANTICIPATE(Gest
X
) and SELFACTIVATE are potentially in conflict with
the faithfulness constraint IDENT(activation)-IO, which preserves gestures’ underlying activation
parameter settings. It is defined in (62); again, note that this constraint is defined with respect to
the potential trigger of regressive harmony, and not the potential undergoers of harmony.
(62) IDENT(activation)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical activation specifications.
The relative ranking of these constraints determines whether or not a language exhibits
regressive harmony. When SELFACTIVATE is ranked above ANTICIPATE(Gest
X
) and
IDENT(activation)-IO, the result is a language with no harmony based on property X. However,
when ANTICIPATE(Gest
X
) is ranked above SELFACTIVATE and IDENT(anticipation)-IO, a gesture
of type X will always trigger harmony for property X, whether they are underlyingly specified as
anticipatory or self-activating.
89
The relevant constraints for the regressive component of ATR harmony in Nandi is
ANTICIPATE(tongue root advancement), which penalizes self-activating tongue root advancement
gestures. The constraint ranking in (63) ensures that an ATR vowel will always surface as
anticipatory and will therefore trigger regressive ATR harmony in Nandi.
(63) Constraint ranking for regressive ATR harmony in Nandi
ANTICIPATE(tongue root advancement) >> SELFDEACTIVATE, IDENT(activation)-IO
The rankings in (59) and (63) are demonstrated in the tableau in (64) for the Nandi form
[ka-ki-kas] ‘hear (past 1p),’ in which the vowel of the non-initial prefix [-ki-] contains the tongue
root advancement gesture. In order to demonstrate that a tongue root gesture will always surface
as both persistent (non-self-deactivating) and anticipatory (early-activating) in Nandi, a
hypothetical input is included in which the tongue root gesture is self-deactivating and self-
activating. For reasons of space and clarity, depictions of candidate forms in this tableau include
only vocalic gestures, and only the gestural scores that are calculated from candidate forms’
coupling graphs.
90
(64) Tableau for Nandi [ka-ki-kas] ‘hear (past 1p)’
Input: / k ɑ
1
- k i
2
- k ɑ
3
s /
PERSIST(ATR)
SELFDEACTIVATE
IDENT(deactivation)-IO
ANTICIPATE(ATR)
SELFACTIVATE
IDENT(activation)-IO
a. [kɑ-ki-kɑs]
*! *!
b. [kɑ-ki-kas]
* * *!
c. [ka-ki-kɑs]
*! * *
d. [ka-ki-kas]
* * * *
In (64), candidate (a) [kɑ-ki-kɑs] preserves the underlying activation and deactivation
parameter settings of the tongue root advancement gesture of the medial [i]. In this candidate,
neither regressive nor progressive harmony is triggered, fatally violating both PERSIST(ATR) and
Tongue Body
pharyngeal wide
1
Tongue Body
palatal narrow
2
Tongue Body
pharyngeal wide
3
Tongue Root
advanced
2
91
ANTICIPATE(ATR). In candidate (b) [kɑ-ki-kas] the tongue root gesture surfaces as persistent but
not anticipatory, triggering progressive but not regressive ATR harmony. This satisfies
PERSIST(ATR) at the expense of lower-ranked SELFDEACTIVATE and IDENT(deactivation)-IO, but
still incurs a fatal violation of high-ranked ANTICIPATE(ATR). In candidate (c) [ka-ki-kɑs] the
tongue root gesture surfaces as anticipatory but not persistent, triggering regressive but not
progressive ATR harmony. This satisfies ANTICIPATE(ATR) while violating lower-ranked
SELFACTIVATE and IDENT(activation)-IO. However, it still fatally violates high-ranked
PERSIST(ATR). The winning candidate (d) [ka-ki-kas], in which the tongue root gesture surfaces
as both persistent and anticipatory and triggers bidirectional ATR harmony, satisfies both
PERSIST(ATR) and ANTICIPATE(ATR).
While Nandi and many other languages exhibit bidirectional tongue root harmony due to
the presence in their respective surface phonological inventories of tongue root gestures that are
both persistent and anticipatory, it should be noted that the Gestural Harmony Model also
predicts the existence of unidirectional harmony. As demonstrated in this section, bidirectional
harmony in this model is actually the result of concurrently operating progressive and regressive
harmony processes. The model relies on the distinct gestural parameter settings of persistence
and anticipation, which are subject to different sets of constraints, to drive regressive and
progressive harmony. Because the constraint rankings in (59) and (63) above do not interact with
one another, it is possible within the Gestural Harmony Model to generate systems in which
progressive and regressive harmony operate entirely independently from one another, or in
which harmony operates in one direction and not another. For instance, purely regressive ATR
would result if the constraints in (59) were reranked such that SELFDEACTIVATE outranked
PERSIST(tongue root advancement) and IDENT(deactivation)-IO, resulting in a language’s ATR
92
vowels surfacing with tongue root advancement gestures that were anticipatory but not
persistent.
In this way, the Gestural Harmony Model mirrors featural approaches to harmony that
rely on constraints that not only drive harmony but also specify the direction in which a feature
should spread. These include constraints on featural alignment (Kirchner 1993; Akinlabi 1994;
Cole & Kisseberth 1994, 1995; Pulleyblank 1996), which specify both the harmonizing feature
and the edge of a domain to which that feature should be aligned (e.g., ALIGN(+F,R) for
progressive harmony). Walker (1998/2000) also proposes directional versions of the constraint
SPREAD(F), which drive spreading of a feature to segments preceding the trigger (SPREAD(F)-L)
and following the trigger (SPREAD(F)-R). Similarly, the Gestural Harmony Model makes use of
two distinct sets of constraints that determine the triggering of harmony: one for the surfacing of
anticipatory gestures for regressive harmony, and one for the surfacing of persistent gestures for
progressive harmony. By using distinct constraints or sets of constraints to account for
progressive and regressive harmony, these theories are able to account for harmony systems in
which harmony is unidirectional, or in which patterns of triggering of harmony differ between
the two directions of a bidirectional harmony system.
There is some debate as to whether this ability to independently derive progressive and
regressive harmony is a desirable aspect of a theory of harmony. Baković (2000) claims that the
directionality of harmony need not be specified within the phonological grammar, because the
directionality of tongue root harmony in a language is directly predictable from its
morphological structure. According to this idea, harmony by default operates bidirectionally, and
any appearance of unidirectionality is attributed to a lack of possible undergoers to one side of a
trigger. For instance, an ATR harmony process that is triggered by root vowels and appears to be
93
purely regressive (leftward) can be attributed to the language’s only having prefixes and not
suffixes.
However, there are a number of harmony systems that call this proposal into question and
point to the importance of the phonological grammar’s ability to derive progressive and
regressive harmony independently of one another. Tongue root harmony systems with regressive
directionality that is unpredictable from morphological structure have been reported in Pulaar
(Niger-Congo; Paradis (1986)), Assamese (Indo-Aryan; Mahanta (2007)), and Karajá (Macro-Jê;
Brazil; Ribeiro (2002)). Unidirectional harmonies that are not derivable from morphological
structure, both progressive and regressive, are also widely reported among nasal harmony
systems (see Walker (1998/2000) for an overview). Therefore, the ability of the Gestural
Harmony Model to independently derive progressive and regressive harmony from distinct
gestural parameter settings is an asset, making it possible to account for both unidirectional and
bidirectional harmony processes. This also allows the model to account for harmony systems in
which the conditions placed on the triggers of progressive and regressive harmony are distinct.
One example of such a system is Capanahua nasal harmony, discussed in section 3.4.3.
3.2.3 Summary
This section has demonstrated the Gestural Harmony Model’s account of the presence
versus absence of harmony in a language as a consequence of the phonological grammar shaping
the surface inventory of a language such that it contains harmony-triggering gestures. As is
typical in OT grammars, surface inventory shaping and the restriction of the distributions of
harmony-triggering gestures in the Gestural Harmony Model is achieved via the relative ranking
of markedness and faithfulness constraints. The relative ranking of the constraints
PERSIST(Gest
X
), SELFDEACTIVATE, and IDENT(deactivation)-IO determines whether a language’s
94
phonological inventory contains persistent (non-self-deactivating) gestures and therefore whether
or not that language exhibits progressive harmony. Similarly, the relative ranking of
ANTICIPATE(Gest
X
), SELFACTIVATE, and IDENT(activation)-IO determines whether a
phonological inventory contains anticipatory (early-activating) gestures and therefore whether or
not that language exhibits regressive harmony.
In addition, the relative ranking of the constraints LICENSE(Gest
X
, prominent position),
INTEGRITY-IO, and MAX(gesture)-IO determine the positional distributions of the gestures in a
language’s inventory. The rounding harmony system of Kyrgyz restricts a harmony-triggering
gesture to the privileged position of the initial syllable, which is enforced via positional
licensing. Analyzing triggering patterns as the result of both inventory shaping and distributional
restrictions also proves useful when analyzing more complex patterns of harmony triggering.
These include patterns in which the ability to trigger harmony is restricted to a certain class of
segments, as well as patterns in which the ability to trigger harmony appears to be contrastive in
a language. Such cases are the focus of sections 3.3 to 3.5.
3.3 Contrastive Triggering of Nasal Harmony in Acehnese and Rejang
Thus far the discussion of how harmony is driven in the Gestural Harmony Model has
focused on cases in which constraints require a gesture of a certain type to surface as persistent
and/or anticipatory. In the straightforward case of progressive rounding harmony in Kyrgyz, for
instance, this constraint is PERSIST(Gest
X
). When it outranks IDENT(deactivation)-IO and
SELFDEACTIVATE, all bearers of a harmonizing property act as triggers. Not yet discussed is the
result of ranking the faithfulness constraint IDENT(deactivation)-IO over the markedness
constraints PERSIST(Gest
X
) and SELFDEACTIVATE. While the ranking of these markedness
constraints over faithfulness favors neutralization of contrast in favor of some uniform setting of
95
a gesture’s deactivation parameter, the ranking of faithfulness over markedness can be used to
capture instances of contrast preservation. The high ranking of the faithfulness constraint
IDENT(deactivation)-IO then introduces the possibility that a gesture’s deactivation parameter
may be contrastive in some languages. This is a desirable prediction that falls out of the Gestural
Harmony Model’s representation of harmony triggering as the result of a gestural parameter.
Such cases of contrastive triggering in the nasal harmony systems of several Malayo-Polynesian
languages are examined in this section.
Many languages with progressive nasal harmony can be analyzed straightforwardly
within the Gestural Harmony Model as having surface inventories containing only persistent
velum opening gestures. However, the model also predicts that languages with both self-
deactivating and persistent velum opening gestures should be able to occur in the same
phonological inventory and act as contrastive phonological units. Such a system is exemplified
by Acehnese (Malayo-Polynesian, Chamic; Indonesia; Durie (1985), Cowan (1981)), whose
inventory is provided in (65).
(65) Acehnese phonological inventory
14
Consonants Vowels
p t c k ʔ i ɯ u
b d ɟ g e ə o
s ʃ ɛ ɔ
m n ɲ ŋ a
m ̂ n ̂ ɲ ̂ ŋ ̂
ɾ
l
j w
h
14
The vowel transcribed by Cowan as /ə/ is transcribed by Durie as /ʌ/. Both describe it as a central vowel.
96
In Acehnese, some forms show progressive nasal harmony triggered by a nasal consonant
and proceeding among vowels, glides, and glottals (66a-c), while others do not (66d-f). All data
are from Durie (1985) and Cowan (1981).
(66) a. [mãw
̃ə
̃ ] ‘rose’ d. [miəb] ‘suck’
b. [mɯ
̃ h̃ãȷ
̃ ] ‘expensive’ e. [tinaj] ‘to dwell’
c. [nãw
̃ɔ
̃ŋ] ‘soul’ f. [mon] ‘cloud’
Durie (1985) attributes this pattern to a distinction between plain nasals, which trigger the
progressive nasal harmony that is common among Austronesian languages, and so-called ‘funny’
nasals, which do not trigger harmony. Similarly, McGinn (1982) and Coady & McGinn (1982)
propose a contrast in Rejang (Malayo-Polynesian, Land Dayak; Indonesia) between plain and
‘barred’ nasals. McGinn reports the phonological inventory in (67).
(67) Rejang phonological inventory
15
Consonants Vowels
p t c k ʔ i u
b d ɟ g e ə o
s a
m n ɲ ŋ
m ̄ n ̄ ɲ ̄ ŋ ̄
ɾ
l
j w
h
The inventory in (67) contains two series of nasal consonants, which McGinn (1982) and
Coady & McGinn (1982) claim contrast in their ability (68a-d) or inability (68e-h) to trigger
nasal harmony.
15
The palatal stop transcribed here as /ɟ/ is transcribed by McGinn as /j/, while the palatal glide transcribed here as
/j/ is transcribed by McGinn as /y/. Also, the vowel transcribed by McGinn as /e/ and described as a central vowel is
transcribed here as /ə/.
97
(68) a. [ɟamẽw
̃ ] ‘guava’ e. [ɟamew] ‘party’
b. [taŋẽn] ‘hand’ f. [tuŋew] ‘wait’
c. [mĩȷ
̃ õw
̃ ã] ‘coconut’ g. [muʔ] ‘eat’
d. [mĩȷ
̃ ẽ] ‘cooked rice’ h. [net] ‘sew’
The distinction between plain and funny/barred nasals in Acehnese and Rejang and their
ability or inability to trigger nasal harmony is captured straightforwardly in the Gestural
Harmony Model. C. Smith (2017b) proposes that the only difference between these two nasal
stop series lies in which velum opening gesture, self-deactivating or persistent, accompanies a
nasal stop’s oral closure gesture. Plain (harmony-triggering) nasals in both Acehnese and Rejang
can be represented as segments with an oral closure gesture accompanied by a persistent velum
opening gesture. Acehnese and Rejang words that include these gestures will exhibit nasal
harmony. This is illustrated in the figure in (69), which provides a gestural score for the Rejang
form [ɟamẽw
̃ ] ‘guava.’
(69) Gestural score for Rejang [ɟamẽw
̃ ] ‘guava’
In contrast, the funny or barred nasals that do not trigger nasal harmony can be
represented as segments with an oral closure gesture accompanied by a self-deactivating velum
opening gesture. This is shown in the gestural score for the Rejang form [ɟamew] ‘party’ in (70).
98
(70) Gestural score for Rejang [ɟamew] ‘party’
These two Rejang words form a minimal pair that contrasts only in the presence versus
absence of nasal harmony. Consistent with previous analyses of nasal harmony in Rejang and in
Acehnese, the Gestural Harmony Model represents this as a contrast between triggering nasal
stops, which are accompanied by persistent velum opening gestures, and non-triggering nasal
stops, which are accompanied by self-deactivating velum opening gestures. The patterns of
harmony triggering in these languages, then, are the result of a property of the surface
phonological inventories of these languages. This is illustrated in (71), which depicts the Rejang
nasal consonant inventory as containing segments with both persistent and self-deactivating
velum opening gestures.
99
(71) Gestural representation of Rejang nasal stop inventory
Non-triggers
/m/ /n/ /ɲ/ /ŋ/
Triggers
/m/ /n/ /ɲ/ /ŋ/
The preservation of the contrast between the two nasal stop series in (71) is enforced in
the Gestural Harmony Model by the high ranking of IDENT(deactivation)-IO, which will allow
both self-deactivating and persistent velum opening gestures to surface when they appear in
input forms. Before presenting an analysis of contrastive triggering of nasal harmony in
Acehnese and Rejang, however, a final aspect of the distribution of triggering and non-triggering
nasals must be mentioned. Durie (1985) states that in Acehnese ‘funny’ nasals may only occur in
the final (stressed) syllable of a word, while plain nasals are unrestricted in their distribution.
While McGinn (1982) and Coady & McGinn (1982) do not explicitly say that the same
restriction holds for barred nasals in Rejang, the data they provide shows a distributional pattern
identical to that in Acehnese.
It appears, then, that the contrast between self-deactivating and persistent velum opening
gestures in Acehnese and Rejang is limited to the privileged position of the final syllable. This
100
type of pattern in which a contrast is preserved only in privileged positions is common among
phonological systems, and can be captured by relativizing faithfulness constraints to those
privileged positions (Beckman 1997, 1998). In this case, it is a specific version of
IDENT(deactivation)-IO that holds only over final syllables (following work by Hyman (1998),
Krämer (2003), Sasa (2009), Walker (2011) and others) that preserves the deactivation parameter
contrast in that position.
16
This relativized constraint, IDENT(deactivation)-IO
Finals
, must be
ranked above PERSIST(velum opening) and SELFDEACTIVATE.
In non-final syllables, the contrast between triggering and non-triggering nasals is
neutralized; in these positions, only triggering nasals may appear. Therefore, PERSIST(velum
opening), which penalizes nasals that do not trigger harmony, must be ranked above the general
constraint IDENT(deactivation)-IO, as well as SELFDEACTIVATE. The full ranking of constraints
necessary to generate the pattern of context-specific contrastive triggering in Acehnese and
Rejang is shown in (72).
(72) Constraint ranking for nasal harmony triggering in Acehnese and Rejang
IDENT(deactivation)-IO
Finals
>> PERSIST(velum opening) >>
IDENT(deactivation)-IO
SELFDEACTIVATE
The results of this ranking are demonstrated in the following tableaux, beginning with the
final-syllable preservation of contrast of a gesture’s deactivation parameter setting. This is
exemplified by the Rejang minimal pair [ɟamẽw
̃ ] ‘guava’ and [ɟamew] ‘party.’ The tableau in
(73) demonstrates the evaluation for [ɟamẽw
̃ ] ‘guava,’ in which [m] serves as a trigger of nasal
harmony.
16
In Acehnese and Rejang, the final syllable is also privileged by its status as the main stressed syllable. Whether
IDENT(deactivation)-IO is relativized to this syllable based on its stress or its position at the end of the word has no
effect on the outcome of the analysis presented in this section.
101
(73) Tableau for Rejang [ɟamẽw
̃ ] ‘guava’
Input: / ɟ
1
a
2
m
3
e
4
w
5
/
IDENT(deactivation)-IO
Finals
PERSIST(velum opening)
IDENT(deactivation)-IO
SELFDEACTIVATE
a.
*
b.
*! * *
In the tableau in (73), the input contains an /m/ that includes a persistent velum opening
gesture. The winning candidate (a) [ɟamẽw
̃ ] surfaces with that persistent gesture, resulting in
nasal harmony throughout the final syllable of the word. It satisfies high-ranked
IDENT(deactivation)-IO
Finals
while violating low-ranked SELFDEACTIVATE. In candidate (b)
[ɟamew], the velum opening gesture is self-deactivating and does not trigger harmony, resulting
in a fatal violation of IDENT(deactivation)-IO
Finals
. It also satisfies SELFDEACTIVATE while
violating PERSIST(velum opening).
Tongue Tip
palatal clo
1
Tongue Body
pharyng wide
2
Lip
closure
3
Tongue Body
palatal mid
4
Lip
protr
5
Velum
open
3
102
Turning to non-triggering nasals, the tableau in (74) demonstrates the evaluation for
[ɟamew] ‘party,’ in which [m] does not trigger nasal harmony.
(74) Tableau for Rejang [ɟamew] ‘party’
Input: / ɟ
1
a
2
m
3
e
4
w
5
/
IDENT(deactivation)-IO
Finals
PERSIST(velum opening)
IDENT(deactivation)-IO
SELFDEACTIVATE
a.
*! * *
b.
*
In (74), the input contains an /m/ that includes a self-deactivating velum opening gesture.
For this input, candidate (a) [ɟamẽw
̃ ] fatally violates IDENT(deactivation)-IO
Finals
by surfacing
with a persistent velum opening gesture that triggers nasal harmony. Winning candidate (b)
Tongue Tip
palatal closure
1
Tongue Body
pharyngeal wide
2
Lip
closure
3
Tongue Body
palatal mid
4
Lip
protrusion
5
Velum
open
3
103
[ɟamew] satisfies IDENT(deactivation)-IO
Finals
by surfacing with a self-deactivating velum
opening gesture that is faithful to its input deactivation parameter.
As for the distribution of nasals in non-final syllables, the tableau in (75) shows the
evaluation of the Rejang form [mĩȷ
̃ ẽ] ‘cooked rice.’ In this form, [m] in a non-final syllable
triggers harmony. In order to demonstrate that an /m/ in this position will surface as a trigger of
harmony no matter its input specification, the tableau in (75) assumes a hypothetical input in
which /m/ includes a self-deactivating velum opening gesture.
104
(75) Tableau for [mĩȷ
̃ ẽ] ‘cooked rice’
Input: / m
1
i
2
j
3
e
4
/
IDENT(deactivation)-IO
Finals
PERSIST(velum opening)
IDENT(deactivation)-IO
SELFDEACTIVATE
a.
* *
b.
*!
In (75), the high-ranked constraint IDENT(deactivation)-IO
Finals
is not active, as neither
output candidate contains a nasal in the privileged position of the final syllable. Winning
candidate (a) [mĩȷ
̃ ẽ], in which the velum opening gesture of [m] is persistent, violates only the
low-ranked general version of IDENT(deactivation)-IO, as it is not faithful to the self-deactivating
velum opening gesture of the input. The deactivation parameter of the velum opening gesture of
candidate (b) [mije] is faithful to the hypothetical input, and therefore satisfies
IDENT(deactivation)-IO. However, it fatally violates higher-ranked PERSIST(velum opening). This
Lip
closure
1
Tongue Body
palatal narrow
2
Tongue Body
palatal narrow
3
Tongue Body
palatal mid
4
Velum
open
1
105
tableau demonstrates, then, that outside of the final syllable, a nasal must surface as a trigger of
nasal harmony no matter the deactivation parameter specification of its velum opening gesture.
The contrastiveness of this deactivation parameter is only preserved in the privileged position of
the final syllable, and neutralized elsewhere in favor of persistent velum opening gestures.
Patterns similar to those in Acehnese and Rejang, in which nasal harmony is triggered by
one nasal consonant series and not another, have also been noted in Bukar-Sadong (Land Dayak;
Indonesia, Malaysia) and Iban (also known as Sea Dayak; Indonesia, Malaysia) by Scott (1957,
1964) and Court (1970). However, in these languages the non-triggering nasals have usually
been described as either homorganic nasal-stop sequences or as prenasalized stops. If this is the
case, the inability of these nasals to trigger progressive nasal harmony can be attributed to the
blocking of nasal harmony by the following obstruent rather than to a difference in the
representation of a nasal consonant’s velum opening gesture. On the other hand, many sources
that described nasal-stop sequences and prenasalized stops in these languages also note that they
are usually produced as something approximating a simple nasal stop. Scott (1957) describes the
production of voiced obstruents following nasals in Iban as ‘very gentle.’ Taking this a step
further, Court (1970) claims that in many of these Indonesian languages the voiced stop portion
of a homorganic nasal-stop sequence disappears entirely. Similarly, ‘funny’ nasals in Acehnese
are analyzed by Cowan (1981) as prenasalized stops that have fused into a simple nasal
consonant.
While the synchronic status of non-triggering nasal consonants in these languages
remains uncertain, they are likely historically derived from prenasalized stops or nasal-stop
clusters. McGinn (1982) and Coady & McGinn (1982) point to a number of cognate pairs
between Rejang and Indonesian (a variety of Malay) to show that the ‘barred’ (non-triggering)
106
nasal consonants of Rejang correspond to nasal-stop clusters in related languages. For instance,
the word for ‘guava’ is [jamew] in Rejang, and [jambu] in Indonesian. In earlier forms of
languages with contrastively triggered nasal harmony, then, there may have been a single
persistent velum opening gesture in the phonological inventory, resulting in nasal harmony in
that language. However, some nasal consonants that included this persistent velum opening
gesture were immediately followed by obstruents, which blocked the spread of nasality. Because
in these forms the extended activation of the velum opening gesture never surfaced, a learner
would have had no reason to posit that the velum opening gesture in these forms was persistent
rather than self-deactivating. A possible result of such a pattern is the assumption by the learner
of an allophonic distribution of velum opening gestures, with the self-deactivating velum
opening gesture present before obstruents (however weakly they were produced), and the
persistent gesture elsewhere. It is possible that this is the case in some present-day Malayo-
Polynesian languages. In others, it appears that the obstruent portion of these prenasalized stops
has been lost entirely, but the ability of the velum opening gesture to self-deactivate, rather than
spreading nasality throughout a word, has remained. In these languages, there are now two
contrastive types of nasal consonants: one accompanied by a self-deactivating velum opening
gesture, and one accompanied by a persistent velum opening gesture.
The attestation of patterns of contrastive triggering in Acehnese and Rejang lends further
support to the Gestural Harmony Model’s treatment of harmony triggering ability as part of a
gesture’s representation. By admitting a gesture’s ability or inability to trigger harmony into the
set of parameters that make up a gesture’s representation, the model admits the possibility that
such a parameter should be able to serve a contrastive function in some languages. This
107
prediction is borne out, as evidenced by the patterns of contrastive triggering in Acehnese,
Rejang, and several other Malayo-Polynesian languages.
The question remains as to why such systems are not more widespread. The cluster of
Malayo-Polynesian languages discussed in this section is the only set of languages to receive
analyses explicitly based on the contrastive ability to trigger harmony. C. Smith (2017b)
proposes that one possibility for their relative rarity is that a harmony system with contrastive
triggering represents an intermediate stage in the diachronic development or loss of a harmony
system, suggesting that such systems might be unstable. Alternatively, the apparent rarity of
contrastive triggering may be an artifact of how such harmony systems are described. There are
considerably more harmony patterns that are described in terms of exceptionality in harmony
triggering. One harmony system that has previously been promoted as an example of
exceptionality in harmony is Classical Manchu tongue root harmony, whose reanalysis within
the Gestural Harmony Model as a case of contrastive triggering is presented in section 3.5.
3.4 Conditional Triggering of Harmony
3.4.1 Conditional Triggering via Co-occurrence Constraints
Among all types of harmony, the sort of symmetry of triggering exhibited by a language
like Kyrgyz (sections 2.2.1 and 3.2.1), in which all round vowels trigger rounding harmony, is
relatively uncommon. Far more common are systems in which conditions are placed on the types
of segments that may serve as triggers of harmony. The Gestural Harmony Model is also
successful in generating these conditional triggering patterns.
In section 3.2, PERSIST(Gest
X
) is proposed to be a perceptually grounded constraint. By
surfacing as persistent, a gesture of type X has the opportunity to extend its period of activation
and improve its perceptual salience. The same is true of gestures that surface as anticipatory and
108
extend in the regressive (leftward) direction. This is in keeping with assertions by Kaun (1995,
2004), Flemming (1995), Walker (2005, 2011), and Kimper (2011) that harmony is perceptually
motivated as a means to increase a listener’s exposure to a perceptually difficult element. There
is a constraint PERSIST(lip protrusion), for example, because the contrast between a round vowel
(with a lip protrusion gesture) and an unround vowel (without a lip protrusion gesture) is
potentially difficult to perceive unless a lip protrusion gesture is of a sufficient length.
However, it is not the case that all round vowels are equally perceptually disadvantaged
or that they trigger harmony with equal frequency across languages. Kaun (1995) discusses this
at length and claims that vowels that are perceptually weak transmitters of a rounding contrast,
i.e., nonhigh vowels and front vowels, are more likely to act as triggers of harmony, a principle
she summarizes with the statement that ‘Bad Vowels Spread.’ If harmony is a phenomenon that
extends the period of activation of hard-to-perceive gestures, it is predicted that the harder it is to
accurately perceive the presence of a gesture, the more likely it will be to trigger harmony. In the
case of rounding harmony, a lip protrusion gesture is more difficult to perceive when it
accompanies a nonhigh vowel or a front vowel. Therefore, when included in the representation
of such vowels, a lip protrusion gesture will be more likely to surface as persistent and/or
anticipatory.
This sort of conditional triggering is not limited to rounding harmony. In nasal harmony,
triggers are sometimes limited to perceptually weak bearers of nasality. Perhaps the most well-
known example of a triggering condition in nasal harmony comes from Inor (also known as
Ennemor; Semitic, Western Gurage; Ethiopia). In Inor, nasal stops do not trigger nasal harmony,
while nasalized continuants (derived by a process of nasal stop spirantization) act as triggers of
bidirectional nasal harmony (Chamora & Hetzron 2000). There are also cases in which nasal
109
harmony is triggered by nasal vowels and not nasal consonants, as in Moba Yoruba (Ajíbóyè
2001; Ajíbóyè & Pulleyblank 2008; Walker 2014).
Many feature-based analyses of harmony capture these sorts of trigger asymmetries by
including restrictions on triggering segments directly within the definition of a harmony-driving
constraint. Kaun (1995), for instance, assumes a general harmony constraint EXTEND(round) for
driving rounding harmony, as well as the specific constraints EXTEND(round)IF[-high] and
EXTEND(round)IF[-back]. Because the Gestural Harmony Model does not utilize any constraints
that serve directly as harmony drivers, this strategy for capturing triggering asymmetries is not
available.
However, it is still possible to capture patterns of conditional harmony triggering in the
Gestural Harmony Model via the shaping of surface phonological inventories. Simple inventory
shaping was discussed in section 2.3 for Kyrgyz and Nandi, whose inventories are shaped by the
high ranking of the constraints PERSIST(Gest
X
) and (in the case of Nandi) ANTICIPATE(Gest
X
).
Conditions on triggers can be captured by assuming additional markedness constraints that
require persistent or anticipatory gestures to co-occur only with certain types of segments. The
analysis of conditional harmony triggering via co-occurrence constraints is the subject of this
section. A simple case of conditional harmony triggering is exemplified by the conditional
triggering observed in Baiyina Oroqen rounding harmony. In addition, in this section I return to
the case of Capanahua nasal harmony (section 2.2.2), which presents an interesting case of
bidirectional harmony in which a triggering condition is placed on one direction of harmony and
not the other.
110
3.4.2 Rounding Harmony in Baiyina Oroqen
Baiyina Oroqen
17
rounding harmony (Li 1996; Kaun 2004; Walker 2014; Dresher &
Nevins 2017) presents a unique case in which only a subset of round vowels trigger rounding
harmony, and in which non-triggers of rounding harmony may either propagate or block
harmony. By representing harmony as the result of overlap of undergoing segments by a single
persistent (non-self-deactivating) gesture, the Gestural Harmony Model is able to
straightforwardly account for both of these aspects of harmony triggering in the language.
Baiyina Oroqen (Northern Tungusic; China) exhibits harmony for both tongue root
position and rounding. The vowel inventory can be split into symmetric sets of ATR and non-
ATR vowels, while rounding is present only on the back vowels. Length is contrastive on all but
the front nonhigh vowels, which are diphthongized. The vowel inventory in (76) is reported by
Li (1996).
(76) Baiyina Oroqen vowel inventory
ATR Vowels Non-ATR Vowels
Front Back Front Back
High i iː u uː ɪ ɪː ʊ ʊː
Nonhigh ie ə əː o oː ɪɛ a aː ɔ ɔː
As in many Tungusic languages, harmony in Baiyina Oroqen is triggered by a vowel in
the initial syllable of a word. Following vowels may agree with the initial vowel in both tongue
root position and rounding, the latter of which is the focus of this section. Rounding harmony in
Baiyina Oroqen holds among the set of non-high back vowels. The vowels /ə/ and /əː/ alternate
with /o/ and /oː/, respectively, with respect to rounding, while /a/ and /aː/ alternate with /ɔ/ and
/ɔː/. While there are a number of round vowels in the language’s inventory, both high and
17
This variety of Oroqen should not be confused with the standard dialect, for which the patterns of triggering of
rounding harmony are somewhat different.
111
nonhigh, harmony is triggered only by the nonhigh short vowels /o/ and /ɔ/, as in (77). All data
are from Li (1996).
(77) a. [somsok-jo] ‘pasture (indef. acc.)’ d. [ɔlɔ-jɔ] ‘fish (indef. acc.)’
b. [ɲoɲo-xoːn-mo] ‘bear (dim. def. acc.)’ e. [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’
c. [oloː-wkoːn-no] ‘to cook (caus. pres.)’ f. [bɔdɔ-wkɔːn-nɔ] ‘to think (caus. pres.)’
Nonhigh round vowels may surface in non-initial syllables only as the product of
harmony, i.e., after triggers /o/ and /ɔ/. If the initial syllable contains an unround vowel, all
following nonhigh vowels must be unround as well. The long vowels /oː/ and /ɔː/ may surface as
the products of harmony, as in (77b-c) and (77e-f). However, when they occur in an initial
syllable they do not trigger rounding harmony. Any nonhigh vowels that follow /oː/ or /ɔː/ in an
initial syllable must surface as their unround alternants, as in (78).
(78) a. [koːməxə] ‘windpipe’ c. [kɔːŋakta] ‘handbell’
b. [boːl-jə] ‘slave (indef. acc.)’ d. [gɔːl-ja] ‘policy (indef. acc.)’
High round vowels are unrestricted in their distribution and do not participate in rounding
harmony in Baiyina Oroqen. Along with the high unround vowels, they act as blockers of
harmony. The analysis of the status of high vowels as blockers within this harmony system is left
until section 4.5.3. The analysis presented in this section focuses on the distribution of the
nonhigh round vowels and their ability or inability to trigger harmony.
The pattern of rounding harmony triggering in Baiyina Oroqen represents a case of
conditional triggering, in which restrictions are imposed on the types of vowels that may act as
harmony triggers. While the Baiyina Oroqen inventory includes a number of round vowels, only
the short nonhigh /o/ and /ɔ/ trigger round harmony. This is in keeping with the widespread
assertion, discussed in section 2.3, that the intent of harmony is to increase the temporal extent of
a perceptually vulnerable harmonizing property in order to maximize its perceptibility. In the
case of Baiyina Oroqen, however, it is not rounding in general that is deemed perceptually
112
vulnerable and therefore targeted for temporal extension, but rather rounding that accompanies
nonhigh vowels. As discussed at length by Kaun (1995, 2004), nonhigh round vowels are at a
perceptual disadvantage in comparison with high round vowels due to the lowered position of the
jaw and its interference in the production of lip protrusion. According to Kaun, this accounts for
the fact that high round vowels do not trigger rounding harmony in Baiyina Oroqen and many
other languages with rounding harmony.
Furthermore, within the class of nonhigh round vowels, short /o/ and /ɔ/ are at a
perceptual disadvantage relative to long /oː/ and /ɔː/. While short /o/ and /ɔ/ trigger harmony in
order to increase the perceptibility of rounding, long /oː/ and /ɔː/ need not trigger harmony as
they are already long enough to render rounding sufficiently perceptible. The perceptual
weakness of round vowels based on height and length serves as the basis of the analyses of the
pattern of rounding harmony triggering in Baiyina Oroqen proposed by Kaun (2004) and Walker
(2014).
Within the Gestural Harmony Model, the triggering of Baiyina Oroqen rounding
harmony can be accounted for via the interaction of markedness and faithfulness constraints that
shapes its surface inventory of vowels with persistent and self-deactivating lip protrusion
gestures. The inventory of round vowels in Baiyina Oroqen is represented gesturally as in (79).
113
(79) Baiyina Oroqen round vowel inventory represented gesturally
/u/ /ʊ/ /o/ /ɔ/
/uː/ /ʊː/ /oː/ /ɔː/
In the inventory in (79), all high vowels include self-deactivating lip protrusion gestures
that will not trigger rounding harmony. Among the nonhigh vowels, short /o/ and /ɔ/ are triggers
of harmony and thus include persistent lip protrusion gestures, while non-triggers /oː/ and /ɔː/ are
accompanied by self-deactivating lip protrusion gestures.
This inventory can be generated within the Gestural Harmony Model as follows. Because
round vowels in general do not trigger harmony, SELFDEACTIVATE must be ranked above
IDENT(deactivation)-IO and PERSIST(lip protrusion). However, the short round vowels /o/ and /ɔ/
always trigger harmony; therefore, they must be banned from co-occurring with self-deactivating
lip protrusion gestures. This can be achieved via a constraint from the *COUPLE family.
114
Introduced in section 1.2.2, *COUPLE constraints ban gestures of certain types from being
coupled to one another in a coupling graph. The *COUPLE constraint relevant to Baiyina Oroqen
rounding harmony is provided in (80).
(80) *COUPLE(short nonhigh V, self-deactivating lip protrusion): Assign a violation mark to a
short nonhigh vocalic gesture and a self-deactivating lip protrusion gesture that are
coupled to one another in an output.
The constraint in (80) penalizes a short nonhigh vowel gesture that is coupled to a self-
deactivating lip protrusion gesture. When this constraint is ranked above SELFDEACTIVATE, a
short nonhigh round vowel will be compelled to surface with a persistent lip protrusion gesture,
and will therefore trigger harmony.
In addition, tongue root harmony in Baiyina Oroqen is triggered by all ATR vowels in the
inventory, indicating that PERSIST(ATR) is also ranked above SELFDEACTIVATE, though this will
not be a major focus of the analysis presented in this section. In order to capture the fact that
rounding and ATR harmony are triggered by a vowel in an initial syllable and that only roots
may introduce harmonizing gestures, I assume the high ranking of two positional licensing
constraints. The constraint LICENSE(nonhigh round V, first σ) penalizes a nonhigh vowel that
includes a lip protrusion gesture that occurs outside of the first syllable of a word, and
LICENSE(ATR, first σ) penalizes a tongue root gesture that occurs outside of the first syllable of a
word. These must both be ranked above INTEGRITY-IO, which must itself outrank MAX(lip
protrusion)-IO and MAX(ATR)-IO. The full constraint ranking is given in the Hasse diagrams in
(81).
115
(81) Constraint ranking for Baiyina Oroqen rounding and ATR harmony
a. Constraint ranking for triggering of rounding harmony by nonhigh short round vowels
b. Constraint ranking limiting rounding and ATR harmony triggers to initial syllable
The workings of the constraint ranking in (81a) are demonstrated by the following
tableaux. In order to focus solely on the pattern of rounding harmony triggering in Baiyina
Oroqen, I consider only non-ATR forms that do not include a tongue root advancement gesture
going forward. Illustrating the importance of high-ranked *COUPLE(short nonhigh V, self-
deactivating lip protrusion) to the triggering of rounding harmony by a short nonhigh round
vowel, the tableau in (82) for the form [ɔlɔ-jɔ] ‘fish (indef. acc.)’ includes a hypothetical input in
which the lip protrusion gesture that accompanies the initial /ɔ/ is self-deactivating. Again, for
reasons of space and clarity, only the vocalic gestures are included for each candidate, which is
represented by the gestural score that is calculated from an output coupling graph. Recall from
the discussion in chapter 2, however, that consonants are considered to be undergoers of
rounding and tongue root harmony.
116
(82) Tableau for Baiyina Oroqen [ɔlɔ-jɔ] ‘fish (indef. acc.)’
Input: /ɔ
1
l a
2
- j a
3
/
*COUPLE(short nonhigh V)
SELFDEACTIVATE
IDENT(deactivation)
PERSIST(lip protrusion)
a. [ɔlɔ-jɔ]
* *
b. [ɔla-ja]
*! *
In (82), winning candidate (a) [ɔlɔ-jɔ] violates SELFDEACTIVATE by including a persistent
lip protrusion gesture in the representation of the word-initial harmony-triggering /ɔ/. In addition,
this candidate violates low-ranked IDENT(deactivation)-IO due to the change in the lip protrusion
gesture’s deactivation parameter between input and output. However, this candidate crucially
satisfies high-ranked *COUPLE(short nonhigh V, self-deactivating lip protrusion). Candidate (b),
on the other hand, in which the nonhigh vowel is accompanied by a self-deactivating lip
protrusion gesture, fatally violates *COUPLE while satisfying SELFDEACTIVATE.
Tongue Body
pharyngeal wide
1
Tongue Body
pharyngeal wide
2
Tongue Body
pharyngeal wide
3
Lip
protrusion
1
117
When a word contains a long round vowel in the initial syllable, rounding harmony is not
triggered, as in the tableau for [kɔːŋakta] ‘handbell’ in (83), which includes a hypothetical input
with an initial persistent lip protrusion gesture.
(83) Tableau for Baiyina Oroqen [kɔːŋakta] ‘handbell’
Input: /k ɔː
1
ŋ a
2
k t a
3
/
*COUPLE(short nonhigh V)
SELFDEACTIVATE
IDENT(deactivation)-IO
PERSIST(lip protrusion)
a. [kɔːŋɔktɔ]
*!
b. [kɔːŋakta]
* *
Because the round vowel in this form is long, the constraint *COUPLE(short nonhigh V,
self-deactivating lip protrusion) is not relevant and is therefore unable to compel the vowel to
surface with a harmony-triggering persistent lip protrusion gesture. In candidate (a) [kɔːŋɔktɔ]
rounding harmony is triggered by a non-self-deactivating lip protrusion gesture, fatally violating
SELFDEACTIVATE. Candidate (b) [kɔːŋakta], which contains a self-deactivating lip protrusion
Tongue Body
pharyngeal wide
1
Tongue Body
pharyngeal wide
2
Tongue Body
pharyngeal wide
3
Lip
protrusion
1
118
gesture accompanying [ɔː], violates only the low-ranked IDENT(deactivation)-IO and PERSIST(lip
protrusion) and is therefore the winner.
The constraint set used here accurately captures the patterns of harmony triggering and
non-triggering by nonhigh round vowels in the initial syllable in Baiyina Oroqen. The language
imposes a condition on which round vowels may trigger rounding harmony, restricting triggering
to the set of nonhigh short round vowels. This conditional triggering is captured
straightforwardly in the Gestural Harmony Model via the use of a constraint on gestural co-
occurrence, in this case a constraint from the *COUPLE family.
In Baiyina Oroqen, nonhigh round vowels may only occur in non-initial syllables as the
result of rounding harmony, i.e., if harmony-triggering [o] or [ɔ] occur in the initial syllable.
Thus, sequences of vowels such as [a-o-o] or [a-a-o] are disallowed while sequences such as
[o-o-o] are tolerated. In the Gestural Harmony Model, this can be captured straightforwardly by
restricting lip protrusion gestures to the initial syllable of the word via a positional LICENSE
constraint. This effectively restricts round vowels from occurring in non-initial syllables unless
they receive their rounding as a result of overlap from a persistent lip protrusion gesture in an
initial syllable. The vowels that undergo rounding harmony will surface as round without being
coupled to their own lip protrusion gestures or being associated with the triggering vowel’s lip
protrusion gesture in any way. A similar account of the distributional restrictions of round
vowels in Kyrgyz is provided in section 3.2.1.
This is demonstrated by examining in greater detail the gestural representation of
harmony in the word [ɔlɔ-jɔ] ‘fish (indef. acc.).’ Its gestural score is provided in (84).
119
(84) Gestural score for vocalic portion of Baiyina [ɔlɔ-jɔ] ‘fish (indef. acc.)’
In this gestural score, there is a single lip protrusion gesture that overlaps all other
segments in the word. As a result, these segments surface as rounded. However, none of these
segments has taken on this property as a result of any kind of association (i.e., coupling) with this
lip protrusion gesture. This is apparent when examining the coupling graph from which the
Coupled Oscillator Model computes the gestural score in (84). This coupling graph is provided
in (85).
(85) Coupling graph for vocalic portion of [ɔlɔ-jɔ] ‘fish (indef. acc.)’
In the coupling graph for [ɔlɔ-jɔ] ‘fish (indef. acc.)’ in (85), the word’s sole lip protrusion
gesture is coupled only to the first vocalic gesture. The following vowels surface as rounded
because they are overlapped by this lip protrusion gesture and not because they are coupled to it,
nor are they coupled to lip protrusion gestures of their own. These vowels are able to surface as
rounded without violating the ban on lip protrusion gestures being coupled to nonhigh vowels in
non-initial syllables only.
120
The fact that undergoers of harmony are overlapped by a harmonizing element without
becoming associated (i.e., coupled) with it has important consequences for the behavior of long
[oː] and [ɔː] in non-initial syllables in Baiyina Oroqen. The nonhigh long vowels are not
harmony triggers when they occur in initial syllables; however, in non-initial syllables they will
propagate harmony that is triggered by a preceding [o] or [ɔ], as in [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim.
def. acc.)’ (cf. [bɪra-xa:n-ma] ‘river (dim. def. acc.)’). The Gestural Harmony Model predicts this
behavior from the non-triggering long nonhigh vowels. In an initial syllable, a long nonhigh
vowel is coupled to its own lip protrusion gesture and is thus subject to the constraint
SELFDEACTIVATE. This constraint ensures that the long nonhigh vowel will not trigger rounding
harmony. However, in a non-initial syllable [oː] and [ɔː] are round because of overlap by another
vowel’s lip protrusion gesture; they are not coupled to their own lip protrusion gestures in this
position due to the high ranking of LICENSE(nonhigh round V, first σ) in (81b). This is illustrated
by the coupling graph and gestural score for [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’ in (86). For
reasons of space and clarity, only the gestures for the vowels are included.
121
(86) Coupling graph and gestural score for vowels of [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’
In (86), the initial [ɔ]
1
is coupled to a persistent lip protrusion gesture, which overlaps all
other gestures and causes following vowels, including long nonhigh vowels, to surface as
rounded. This satisfies high-ranked *COUPLE(short nonhigh V, self-deactivating lip protrusion).
Crucially, the single lip protrusion gesture of the triggering initial [ɔ]
1
is the source of rounding
for all of the undergoers of harmony; rounding does not spread iteratively from vowel to vowel.
Thus, the vowel [ɔː]
3
is not a source of rounding for the final [ɔ]
4
, but rather a fellow undergoer
of rounding harmony. The fact that [ɔː] does not trigger harmony is irrelevant to its ability to
propagate harmony to following vowels.
As pointed out by Walker (2014), an alternative analysis that views harmony as a process
of local iterative feature spreading faces difficulty in accounting for the patterning of the long
nonhigh vowels in Baiyina Oroqen. This difficulty is due to such models’ representation of
harmony as the step-by-step addition of association lines between segments and a harmonizing
122
feature in order to satisfy a harmony-driving constraint or due to the application of a feature-
spreading rule. In Baiyina Oroqen, such a rule could be defined as in (87).
(87) Iterative [+round] spreading rule for Baiyina Oroqen
Iterative
This rule states that the feature [+round] will spread from one nonhigh vowel (the trigger)
to another (the undergoer), provided that the trigger is short. Because it is specified as an
iterative rule, it may apply to its own output provided the structural description of the rule is met.
The derivation for the Baiyina Oroqen form [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’ proceeds
according to this rule as in (88).
(88) Derivation of [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’ via feature spreading rule
Input
First rule application
Second rule application
Structural description not met;
rule does not apply
*[dʒɔlɔ-xɔːn-ma] Output (ill-formed)
123
At each step of the derivation in (88), the segments that meet the structural description of
the rule in (87) change, and a different segment acts as the trigger of harmony. As a result,
prohibiting a long nonhigh round vowel from triggering harmony will also prohibit it from
propagating harmony that originates from another round vowel, contrary to the observed pattern
in Baiyina Oroqen. After the second application of the rule, its structural description is no longer
met and it will no longer apply. The iterative spreading rule in (87), then, is unable to produce
forms such as [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.),’ in which a non-triggering long nonhigh
vowel such as [ɔː] propagates harmony.
Walker (2014) concludes that harmony should be modeled as the result of relations
between a single trigger and multiple undergoers, rather than a sequence of local triggers and
targets. The Gestural Harmony Model shares this representation of the relation between triggers
and undergoers of harmony, and as a result faces no such trouble in analyzing harmony patterns
in which some segment types are capable of undergoing and propagating harmony but are not
capable of serving as the initial trigger of harmony, as in Baiyina Oroqen. This is a consequence
of the fact that harmony is represented as the overlap of potentially many undergoers by a
persistent or an elastic gesture that serves as the singular source of a harmonizing property.
According to this representation, a so-called propagator of harmony is not actually the source of
a harmonizing property, but rather just another undergoer, and thus its status as a trigger or non-
trigger of harmony is irrelevant to the other undergoers around it.
Dresher & Nevins (2017) propose an alternative account of Baiyina Oroqen forms such
as [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’ that circumvents the difficulty encountered by the rule
in (87) while still maintaining a local iterative model of harmony. They propose that in Baiyina
Oroqen, stem-internal and suffix-targeting rounding harmony are two distinct processes, and that
124
suffix-targeting rounding harmony is triggered only by nonhigh round vowels, either short or
long, in a non-initial syllable. While this account generates the correct surface forms for Baiyina
Oroqen, it relies on a more complicated view of the patterns of vowel alternation observed in the
language. The one-to-many relationship between trigger and undergoers adopted by the Gestural
Harmony Model and by Walker (2014), on the other hand, is able to avoid such complications,
instead maintaining a unified view of Baiyina Oroqen rounding harmony as a single
phonological process.
In addition to illustrating the relation between triggers and undergoers assumed within the
Gestural Harmony Model, Baiyina Oroqen rounding harmony also provides a good test case for
the model’s ability to produce an analysis of a more complex pattern of harmony triggering. By
appealing to the use of a gestural co-occurrence constraint from the *COUPLE family, the surface
inventory of round vowels in Baiyina Oroqen can be shaped such that only the perceptually
disadvantaged short nonhigh round vowels /o/ and /ɔ/ include persistent lip protrusion gestures.
In addition, the representation of harmony as gestural overlap rather than feature association has
proven advantageous in analyzing the distributions of the non-triggering long nonhigh round
vowels. Not yet discussed is the role played by high vowels, which block rounding harmony in
Baiyina Oroqen; this is taken up in section 4.5.3.
3.4.3 Revisiting Capanahua: Regressive & Bidirectional Nasal Harmony
Capanahua nasal harmony (section 2.2.2) presents a particularly interesting case of
harmony triggering, as it exhibits harmony that is in some cases purely regressive and in other
cases bidirectional. As such, an analysis of Capanahua must make use of constraints on both
persistent and anticipatory gestures. In addition, bidirectional nasal harmony in Capanahua
exhibits a pattern of harmony triggering that can be described as unconditional in the regressive
125
direction but conditional in the progressive direction. This pattern is accounted for
straightforwardly in the Gestural Harmony Model, in which regressive and progressive harmony
are the results of two different types of gestures, persistent and anticipatory, being permitted to
surface in a language’s phonological output forms.
Recall that in Capanahua, nasal harmony is triggered by the nasal stops /n/ and /m/ and
affects preceding vowels, glides, and glottals. The data in (89) is reported by Loos (1967/1969)
and is repeated from (34) in section 2.2.2.
(89) a. [h̃ãmawɯ] ‘step on it’
b. [põȷ
̃ ãn] ‘arm’
c. [bãw
̃ĩn] ‘catfish’
d. [cĩʔ̃ĩn] ‘by fire’
e. [cipõnki] ‘downriver’
f. [wɯɾãnwɯ] ‘push it’
g. [wɯɾãnjasãʔ̃nwɯ] ‘push it sometime’
h. [bãnawɯ] ‘plant it’
Bidirectional nasal harmony also arises in Capanahua under certain circumstances. In fast
or casual speech, a coda nasal consonant’s oral closure component is lost, but its nasality remains
and triggers nasal harmony, as in (90).
18
In such cases, nasality spreads both regressively and
progressively from the original position of the deleted nasal consonant; progressive harmony can
be observed in the forms in (90d-e).
(90) a. [põȷ
̃ ã] ‘arm’ (cf. [põȷ
̃ ãn])
b. [bãw
̃ĩ] ‘catfish’ (cf. [bãw
̃ĩn])
c. [cĩʔ̃ĩ] ‘by fire’ (cf. [cĩʔ̃ĩn])
d. [wɯɾãw
̃ɯ
̃ ] ‘push it’ (cf. [wɯɾãnwɯ])
e. [wɯɾãȷ
̃ ãsãʔ̃w
̃ɯ
̃ ] ‘push it sometime’ (cf. [wɯɾãnjasãʔ̃nwɯ])
18
This process, referred to as ‘nasal loss’ by Loos, does not apply to coda consonants that occur before a stop
consonant, either oral or nasal. The details of this process are not a major focus of this section, and are not crucial to
the analysis presented here.
126
Focusing first on the regressive component of this harmony process, in the Gestural
Harmony Model this is accounted for by positing a surface phonological inventory in which
these triggering nasal stops are accompanied by anticipatory (early-activating) velum opening
gestures, as in the inventory in (91), repeated from (39) in section 2.2.2.
(91) Capanahua nasal consonant inventory
/m/ /n/
The grammar of a language like Capanahua, in which all nasal stops include anticipatory
velum opening gestures on the surface, must ensure that even an underlying self-activating
velum opening gesture will surface as anticipatory. As already demonstrated for Nandi tongue
root harmony in section 3.2.2, this can be achieved by the high ranking of a markedness
constraint from the ANTICIPATE(Gest
X
) family. With ANTICIPATE(velum opening) ranked above
SELFACTIVATE and IDENT(activation)-IO in Capanahua, all velum opening gestures will surface
as anticipatory and will trigger regressive (leftward) harmony. The lack of progressive
(rightward) harmony in this language, meanwhile, can be attributed to the high ranking of the
constraint SELFDEACTIVATE. The full ranking used to generate regressive but not progressive
harmony in Capanahua is provided in (92).
127
(92) Constraint ranking for regressive nasal harmony in Capanahua
a. ANTICIPATE(velum opening) >> SELFACTIVATE, IDENT(activation)-IO
b. SELFDEACTIVATE >> PERSIST(velum opening), IDENT(deactivation)-IO
The effect of this ranking is demonstrated by the tableau in (93) for Capanahua
[h̃ãmawɯ] ‘step on it.’ To demonstrate that a velum opening gesture will surface as anticipatory
no matter its input activation parameter setting, the hypothetical input in this tableau includes a
non-anticipatory velum opening gesture. In addition, to demonstrate that a velum opening
gesture will surface as self-deactivating no matter its input deactivation parameter setting, the
velum opening gesture in this hypothetical input is persistent.
128
(93) Tableau for regressive nasal harmony in [h̃ãmawɯ] ‘step on it’
Input: / h
1
a
2
m
3
a
4
w
5
ɯ
6
/
ANTICIPATE(velum opening)
SELFACTIVATE
IDENT(anticipation)-IO
SELFDEACTIVATE
PERSIST(velum opening)
IDENT(deactivation)-IO
a.
*!
* *
F b.
* *
* *
c.
* *
*!
In (93), candidate (a) [hamawɯ] surfaces faithfully with a velum opening gesture that is
neither anticipatory nor persistent, and therefore triggers no nasal harmony. This fatally violates
high-ranked ANTICIPATE(velum opening). Candidate (b) [h̃ãmawɯ] and candidate (c) [hãmãw
̃ɯ
̃ ]
Glottis
open
1
Tongue
Body
pharyng
wide
2
Lip
closure
3
Tongue
Body
pharyng
wide
3
Lip
protr
5
Tongue
Body
uvular
narrow
6
Velum
open
3
129
both satisfy this constraint by surfacing with an anticipatory velum opening gesture that acts as a
trigger of regressive harmony. The choice between these two candidates falls to the constraints
on persistent gestures. Candidate (c) [hãmãw
̃ɯ
̃ ], in which nasal harmony is bidirectional,
violates high-ranked SELFDEACTIVATE, while winning candidate (b) [hãmawɯ], in which
harmony is only regressive, satisfies SELFDEACTIVATE. This tableau demonstrates that even
when an underlying form contains a non-anticipatory, persistent velum opening gesture, the high
ranking of ANTICIPATE(velum opening) and SELFDEACTIVATE in Capanahua ensures that nasal
consonants will surface with an anticipatory, self-deactivating velum opening gesture and will
therefore trigger regressive (leftward), but not progressive (rightward), nasal harmony.
Because of the high ranking of SELFDEACTIVATE, nasal consonants /n/ and /m/ do not
trigger progressive (rightward) nasal harmony. However, as mentioned above, there is one
specific instance in which progressive harmony is triggered: when a nasal consonant’s primary
consonantal gesture is deleted during fast or casual speech, leaving only its velum opening
gesture. This stranded velum opening gesture that results from nasal consonant deletion must be
coupled to some other gesture in the word in order to be pronounced. I assume that as a result,
the velum opening gesture is coupled to the preceding vowel, where it surfaces as persistent,
serving as a trigger of progressive nasal harmony.
It takes the addition of only one constraint to successfully account for this pattern of
conditional bidirectional harmony. The triggering of progressive nasal harmony only when a
velum opening gesture is coupled to a vocalic gesture and not to a consonantal gesture can be
analyzed as satisfying a high-ranked *COUPLE constraint that penalizes the co-occurrence of a
vocalic gesture with a self-deactivating velum opening gesture. This constraint is defined in (94).
130
(94) *COUPLE(vowel, self-deactivating velum opening): Assign a violation mark to a vocalic
gesture that is coupled to a self-deactivating velum opening gesture in the output.
Section 3.4.2 introduces the idea that cases of conditional triggering can be accounted for
within the Gestural Harmony Model via the use of constraints that prevent perceptually weak
bearers of some property from co-occurring with gestures that do not trigger harmony for that
property. In the case of nasal harmony, vowels are claimed to be preferred triggers because they
are perceptually weak bearers of nasality (Cole & Kisseberth 1995; Walker 2014). By extending
its period of activation, the velum opening gesture of a vowel increases its likelihood of being
correctly perceived. As a result of the high ranking of the *COUPLE constraint in (94), in
Capanahua a velum opening gesture will only surface as persistent when it is coupled to a
vocalic gesture. Therefore, nasal consonants will not serve as triggers of progressive harmony,
but vowels that become nasalized due to nasal consonant deletion will trigger progressive
harmony. The triggering of regressive nasal harmony, meanwhile, remains unaltered, as it is the
result of a distinct set of constraints on gestural anticipation.
The full set of constraints necessary for this analysis of bidirectional nasal harmony in
Capanahua is provided in (95).
(95) Constraint ranking for bidirectional nasal harmony in Capanahua
a. ANTICIPATE(velum opening) >> SELFACTIVATE, IDENT(anticipation)-IO
b. *COUPLE(vowel, self-deactivating velum opening) >> SELFDEACTIVATE >>
PERSIST(velum opening), IDENT(deactivation)-IO
The interaction of the constraints ANTICIPATE(velum opening), SELFACTIVATE, and
IDENT(anticipation)-IO has already been demonstrated by the tableau in (93), and is not affected
by any of the constraints in (95b). The workings of the constraints in (95b) are illustrated below
for the Capanahua form [wɯɾãnwɯ]~[wɯɾãw
̃ɯ
̃ ] ‘push it.’ The tableau in (96) includes a form in
131
the typical register, in which nasal coda deletion does not take place, and only regressive
harmony takes place. For reasons of space, the constraints that select an anticipatory velum
opening gesture are not included, and only candidates with anticipatory gestures are considered.
Note also that in this form the tap [ɾ] blocks nasal harmony; this is set aside until section 4.4.4.
(96) Tableau for Capanahua [wɯɾãnwɯ] ‘push it’ (neutral register)
Input: / w
1
ɯ
2
ɾ
3
a
4
n
5
w
6
ɯ
7
/ (neutral register)
*COUPLE(V, self-deact. velum)
SELFDEACTIVATE
PERSIST(velum opening)
IDENT(deactivation)-IO
a.
*
b.
* *
Lip
protr
1
TB
uvu nar
2
TT
alv tap
3
TB
phar
wide
4
TT
alv clo
5
Lip
protr
6
TB
uvu nar
7
Velum
open
5
132
In (96), both candidates contain a velum opening gesture that is coupled to a primary
consonantal gesture; therefore, high-ranked *COUPLE(vowel, self-deactivating velum opening) is
not relevant. In candidate (a) [wɯɾãnw
̃ɯ
̃ ] the velum opening gesture is persistent and triggers
progressive nasal harmony, satisfying PERSIST(velum opening) but fatally violating higher-
ranked SELFDEACTIVATE. In the winning candidate (b) [wɯɾãnwɯ] there is no progressive nasal
harmony, as the velum opening gesture is self-deactivating, satisfying SELFDEACTIVATE. This
ranking successfully captures the ability of a nasal consonant to trigger regressive, but not
progressive, nasal harmony in the neutral register in Capanahua.
When the alveolar closure gesture of coda /n/ is deleted in the fast or casual register, the
picture is different. In this case, harmony is both regressive and progressive, as demonstrated by
the tableau in (97) for [wɯɾãw
̃ɯ
̃ ] ‘push it.’ Here, the alveolar closure gesture of /n/ is deleted in
both of the output candidates. As a result, the velum opening gesture of /n/ couples to the gesture
of the preceding vowel. As it is not the focus of this analysis, I set aside determining what drives
the deletion of nasal consonants in fast and/or casual speech.
133
(97) Tableau for Capanahua [wɯɾãw
̃ɯ
̃ ] ‘push it’ (fast/casual register)
Input: / w
1
ɯ
2
ɾ
3
a
4
n
5
w
6
ɯ
7
/ (fast/casual register)
*COUPLE(V, self-deact. velum)
SELFDEACTIVATE
PERSIST(velum opening)
IDENT(deactivation)-IO
a.
*
b.
*! * *
In (97), high-ranked *COUPLE(vowel, self-deactivating velum opening) is active as the
velum opening gesture is coupled to the preceding vowel gesture in both candidates. Now, it is
candidate (a) [wɯɾãw
̃ɯ
̃ ] that is the winner. While this candidate still violates SELFDEACTIVATE,
it satisfies high-ranked *COUPLE(vowel, self-deactivating velum opening). Candidate (b)
[wɯɾãwɯ], in which the velum closure gesture is self-deactivating and there is no progressive
nasal harmony, fatally violates this high-ranked *COUPLE constraint.
Lip
protr
1
TB
uvu nar
2
TT
alv tap
3
TB
phar
wide
4
TT
alv clo
5
Lip
protr
6
TB
uvu nar
7
Velum
open
5
134
This analysis bears some similarity to Safir’s (1982) analysis of Capanahua nasal
harmony. He claims that bidirectional nasal harmony is triggered by a vocalic segment
immediately preceding a nasal consonant, which optionally deletes only after a process of local
nasalization of the vocalic segment. In the gestural analysis proposed here, bidirectional (as
opposed to strictly regressive) nasal harmony is triggered when a velum opening gesture is
coupled to a vocalic gesture as a result of the deletion of a nasal consonant’s oral constriction. In
a sense, then, it is the nasalized vocalic element that triggers harmony.
19
Nasal harmony in Capanahua is an interesting example of a system of harmony triggering
in which a condition on the identity of a triggering segment is enforced in one direction of
harmony but not the other. This is accounted for straightforwardly in the Gestural Harmony
Model, in which progressive and regressive harmony are the results of two distinct gestural
parameter settings and are therefore subject to different sets of constraints. While these
constraints parallel one another in their definitions and in the ways in which they shape
phonological inventories to include or exclude gestures with certain parameter settings, the two
sets of constraints do not interact with one another. In Capanahua nasal harmony, in which the
triggering patterns of regressive and progressive harmony are distinct from one another, this
independence of the constraint sets is an asset of the model.
3.5 Conditional and Contrastive Triggering: Tongue Root Harmony in Classical Manchu
Throughout this chapter, the Gestural Harmony Model has proven itself capable of
accounting for multiple sources of complexity in patterns of harmony triggering. As discussed in
sections 3.3 and 3.4, such complexity includes cases in which conditions are placed on trigger
19
Trigo (1988) offers a slightly different analysis of the progressive component of Capanhua nasal harmony,
proposing that harmony is triggered by a debuccalized nasal consonant that fully deletes later in the derivation.
135
position and identity, as well as cases of contrastive triggering. Classical Manchu (Tungusic; 17
th
to 19
th
century China) provides a unique case of a tongue root harmony system characterized by
contrastive triggering, conditional triggering, and restrictions on the position of triggers. This
makes it a good case with which to test the claim that within the Gestural Harmony Model even
quite complex harmony triggering patterns can be reduced to relatively simple principles of
inventory shaping and restrictions on the distributions of certain types of segments or gestures.
While earlier research by Vago (1973), Odden (1978), and others treats the harmony
system of Classical Manchu as one based on vowel backness, Zhang (1996) and Li (1996) argue
that the assignment of vowels into one of two harmony classes is best understood based on
tongue root position. The vowel system of Classical Manchu according to Zhang is provided in
(98).
(98) Classical Manchu vowel inventory
ATR Vowels non-ATR Vowels
High i u ʊ
Nonhigh ə a ɔ
Classical Manchu has a tongue root harmony system in which non-ATR /ʊ/ alternates
with ATR /u/, and non-ATR /a/ alternates with ATR /ə/.
20
As in other Tungusic languages, the
tongue root specification of the initial syllable determines whether following vowels in the root
and any suffixes are ATR or non-ATR. In an initial syllable, the vowel [ə] always triggers ATR
harmony (99a-c); non-ATR [a] and [ʊ] may never follow [ə]. [ə] may never occur in a non-initial
syllable except as a product of harmony, i.e., following a harmony-triggering ATR vowel. All
data are from Zhang (1996).
20
Classical Manchu also exhibits rounding harmony among nonhigh vowels, causing /a/ to alternate with /ɔ/.
However, I set aside discussion of rounding harmony in order to focus on the complexities of the language’s patterns
of ATR harmony triggering.
136
(99) a. [səbɟə-ŋgə] ‘joyous’ d. [kimu-ŋgə] ‘harboring hatred’
b. [hərə-ku] ‘ladle’ e. [sisə-ku] ‘sieve’
c. [hətu-kən] ‘somewhat stocky’ f. [uɟə-kən] ‘somewhat heavy’
If the non-ATR vowels [a] or [ʊ] occur in an initial syllable, all following nonhigh
vowels must surface as non-ATR [a], as in (100). High vowels in non-initial syllables are
unrestricted in terms of tongue root specification, with [i], [u], and [ʊ] all occurring in those
positions (100c-g). When [i] and [u] occur non-initially, they never trigger ATR harmony in
following vowels (100d-g).
(100) a. [aga-ŋga] ‘of rain’
b. [hʊla-ŋga] ‘crying’
c. [malhʊ-ŋga] ‘frugal’
d. [kani-ŋga] ‘agreeing’
e. [dacu-kan] ‘somewhat sharp’
f. [gʊni-ŋga] ‘thoughtful’
g. [hʊdu-kan] ‘somewhat fast’
Of particular interest here is the fact that the high ATR vowels [i] and [u] do not
uniformly trigger ATR harmony when they occur in an initial syllable, as is the case for non-high
[ə]. Instead, harmony is triggered by some high vowels and not others. In (101a-e), [i] and [u]
trigger ATR harmony within and across morpheme boundaries, while in (101f-i) ATR harmony
is not triggered.
(101) a. [sitə-ku] ‘bed-wetter’ f. [nimasa-kʊ] ‘two-man boat’
b. [cisu-də] ‘act for interest’ g. [nilhʊ-da] ‘be slick’
c. [hurə-nə] ‘arch’ h. [tuwa-na] ‘go to look’
d. [urgu-ŋgə] ‘joyous’ i. [mukʊ-ha-] ‘held a liquid in the mouth’
e. [uŋgi-hə] ‘sent off’
Multiple high ATR vowels may occur in a root without being the product of harmony;
these roots will take non-ATR suffixes, as in (102). It is possible for high vowels with
independent ATR specifications to surface in a sequence, as in (102a-d), giving the appearance
of a harmonizing root while failing to trigger harmony in suffixes. While such a sequence of
137
ATR vowels may appear to be the result of within-root harmony, there is evidence suggesting
that it is not; each high ATR vowel is independently specified as ATR. This is easier to see in
(102e-f), whose roots contain multiple high ATR vowels separated by a non-ATR vowel. Since
the high ATR vowels in the initial syllables of these forms do not trigger harmony, these roots
take non-ATR suffixes.
(102)
a. [sifi-kʊ] ‘hairpin’
b. [bilu-ŋga] ‘pacified’
c. [fuli-ŋga] ‘lucky’
d. [usu-kan] ‘somewhat fussy’
e. [mijali-kʊ] ‘measurer’
f. [duwali-ŋga] ‘of the category’
The pattern of harmony triggering exhibited by the high vowels in Classical Manchu
represents another case of contrastive triggering similar to those seen in Acehnese and Rejang
nasal harmony. Following C. Smith (2017a), I analyze Classical Manchu as having two series of
high ATR vowels: one that triggers ATR harmony, and one that does not. The Gestural Harmony
Model is able to capture this straightforwardly by allowing these vowels to contrast with respect
to the deactivation parameters of their accompanying tongue root advancement gestures. The
high ATR vowel inventory of Classical Manchu, which contains both persistent (non-self-
deactivating) and self-deactivating tongue root advancement gestures, is provided in (103).
138
(103) Gestural representation of Classical Manchu high ATR vowel inventory
Harmony Triggers (persistent) Harmony Non-Triggers (self-deactivating)
/i/ /u/ /i/ /u/
The nonhigh ATR vowel /ə/, on the other hand, is always a trigger of ATR harmony, and
shows no such deactivation parameter contrast. This vowel always surfaces with a persistent
tongue root advancement gesture, as in (104).
(104) Gestural representation of Classical Manchu /ə/
The non-ATR vowels /ʊ/, /a/, and /ɔ/, meanwhile, are represented by vowel gestures that
are not accompanied by tongue root gestures, as in (105).
(105) Gestural representation of Classical Manchu non-ATR vowel inventory
/ʊ/ /a/ /ɔ/
139
In addition to shaping the surface vowel inventory of Classical Manchu such that it
includes the vowels in (103)-(105), the phonological grammar must account for their different
distributional restrictions. It is not the case that any vowel may appear in any position in a word.
Non-triggering /i/ and /u/ are unrestricted, appearing both in initial and non-initial syllables.
Triggering /i/ and /u/, however, are restricted to the initial syllable of a word. As in Acehnese and
Rejang nasal harmony, the contrast between self-deactivating and persistent gestures is restricted
to a privileged position. In addition, while exhibiting no harmony triggering contrast, the
nonhigh ATR /ə/ is also subject to a positional restriction, occurring only in the initial syllable of
a word or as a product of ATR harmony. Despite their complexity, the inventory shaping and
distributional restrictions exhibited by the Classical Manchu vowel system can be achieved
within the Gestural Harmony Model via the interaction of a small set of markedness and
faithfulness constraints. These constraints and their interaction are described in the remainder of
this section.
Focusing first on non-initial syllables, i.e. non-privileged positions, a high ATR vowel
may never trigger harmony and therefore must always surface with a self-deactivating tongue
root advancement gesture. Therefore, the constraint SELFDEACTIVATE must be ranked above
PERSIST(ATR) and a general version of IDENT(deactivation)-IO. This ranking eliminates the
contrast between the self-deactivating and persistent tongue root advancement gesture in
Classical Manchu in all positions within a word. To preserve the underlying specification of a
gesture’s deactivation parameter only in the privileged position of the initial syllable,
IDENT(deactivation)
Firsts
, a positional faithfulness constraint (Beckman 1997, 1998), must be
ranked above SELFDEACTIVATE. With this ranking, the contrast between self-deactivating and
140
persistent tongue root advancement gestures will be preserved in the first syllable of a word, and
neutralized elsewhere.
Classical Manchu exhibits not only contrastive harmony triggering but a form of
conditional triggering as well. Unlike its high ATR counterparts, in an initial syllable the
nonhigh ATR /ə/ does not contrast for its ability to trigger harmony, but instead triggers harmony
exceptionlessly. It is necessary, then, to restrict /ə/ from being accompanied by a self-
deactivating tongue root advancement gesture. As discussed in section 3.4, such a pattern can be
achieved via a gestural co-occurrence constraint from the *COUPLE family. In the case of
Classical Manchu, this constraint penalizes a nonhigh vowel that is accompanied by a self-
deactivating tongue root advancement gesture, as defined in (106).
(106) *COUPLE(nonhigh vowel, self-deactivating ATR): Assign a violation mark to a nonhigh
vowel gesture and a self-deactivating tongue root advancement gesture that are coupled
to one another.
This constraint must outrank IDENT(deactivation)-IO
Firsts
; even in the privileged position
of the initial syllable, triggering is not contrastive for the vowel /ə/ in Classical Manchu.
Finally, recall that [ə] may not occur in a non-initial syllable except as a product of
harmony, i.e., except when it is overlapped by the persistent tongue root advancement gesture of
a harmony-triggering vowel in an initial syllable. This distributional restriction is achieved via
the inclusion of two MAX(gesture)-IO constraints, one general and one position-specific. In a
non-initial syllable, only self-deactivating tongue root advancement gestures are permitted due to
the ranking of SELFDEACTIVATE over PERSIST(ATR) and the general version of
IDENT(deactivation)-IO. As a result, high ATR vowels are non-triggers in these positions. The
nonhigh ATR vowel [ə], on the other hand, may never be accompanied by a non-triggering, self-
deactivating tongue root gesture due to high-ranked *COUPLE(nonhigh V, self-deact. ATR).
141
Instead, any nonhigh vowel in a non-initial syllable that is not an undergoer of harmony must
surface as non-ATR [a]. The fact that a nonhigh vowel must satisfy both *COUPLE and
SELFDEACTIVATE, even if it requires deletion of an underlying tongue root advancement gesture,
indicates that both *COUPLE and SELFDEACTIVATE must outrank a constraint from the
MAX(gesture)-IO family. This constraint is defined in (107).
(107) MAX(ATR)-IO: Assign a violation mark to a segment (set of gestures) that includes a
tongue root advancement gesture in the input if its output correspondent does not include
that gesture.
This constraint must be ranked above IDENT(deactivation)-IO. This ranking will ensure
that the tongue root gesture of a high vowel is never deleted in order to satisfy SELFDEACTIVATE;
it is only manipulated with respect to its gestural deactivation parameter.
In an initial syllable, however, the tongue root advancement gesture of nonhigh /ə/ is
permitted to surface. Therefore, it is necessary to include a position-specific version of
MAX(ATR)-IO, defined in (108).
(108) MAX(ATR)-IO
Firsts
: Assign a violation mark to a segment (set of gestures) in an initial
syllable that includes a tongue root advancement gesture in the input if its output
correspondent does not include that gesture.
With both MAX(ATR)-IO
Firsts
and *COUPLE dominating SELFDEACTIVATE, the tongue
root advancement gesture of underlying /ə/ will always surface as persistent, and will therefore
always trigger ATR harmony. In addition, by ranking MAX(ATR)-IO
Firsts
over
IDENT(deactivation)-IO
Firsts
, the grammar will compel any underlying /ə/ that is in an initial
syllable and is accompanied by a self-deactivating tongue root advancement gesture to change
the deactivation parameter of that gesture rather than deleting it.
The full ranking of constraints necessary to account for Classical Manchu tongue root
harmony is provided in (109).
142
(109) Constraint ranking for Classical Manchu tongue root harmony
The analysis of the distribution of harmony triggering and non-triggering vowels in
Classical Manchu using the constraint ranking in (109) is demonstrated in the following
tableaux. The first tableau in (110) is for the form [sisə-ku] ‘sieve,’ in which an /i/ in the initial
syllable triggers ATR harmony throughout the rest of the word. For the sake of space and clarity,
only the vowel gestures for the input and output candidates are included.
143
(110) Tableau for Classical Manchu [sisə-ku] ‘sieve’
Input: /s i
1
s a
2
– k ʊ
3
/
MAX(ATR)-IO
Firsts
*COUPLE(nonhigh V, self deact. ATR)
IDENT(deactivation)-IO
Firsts
SELFDEACTIVATE
MAX(ATR)-IO
IDENT(deactivation)-IO
PERSIST(ATR)
a. [sisə-ku]
*
b. [sisa-kʊ]
*! * *
c. [sɪsa-kʊ]
*! *
The input of the tableau in (110) contains the vowel /i/ with a persistent tongue root
advancement gesture. In outputs (a) and (b), this vowel is in the initial syllable. The winning
candidate (a) [sisə-ku] surfaces with a faithfully realized persistent tongue root advancement
gesture, violating SELFDEACTIVATE but satisfying higher-ranked IDENT(deactivation)
Firsts
. In
candidate (b) [sisa-kʊ] the tongue root advancement gesture of [i] is self-deactivating, resulting
Tongue Body
palatal narrow
1
Tongue Body
palatal wide
2
Tongue Body
uvular narrow
3
Tongue Root
advanced
1
144
in the following vowels surfacing as their non-ATR alternants. This satisfies SELFDEACTIVATE
but fatally violates higher-ranked IDENT(deactivation)
Firsts
, as well as the low-ranked general
IDENT(deactivation)-IO and PERSIST(ATR). The candidate that is faithful to the input’s specified
deactivation parameter for tongue root advancement is the winner. Candidate (c) [sɪsa-kʊ]
vacuously satisfies the markedness constraints and both IDENT(deactivation)-IO constraints by
deleting the tongue root advancement gesture altogether. However, in doing so it fatally violates
both high-ranked MAX(gesture)-IO
Firsts
as well as general MAX(gesture)-IO.
The faithful candidate is also selected when the input contains a self-deactivating tongue
root advancement gesture in an initial syllable. This is demonstrated in the tableau in (111) for
[nilhʊ-da] ‘be slick.’
145
(111) Tableau for Classical Manchu [nilhʊ-da] ‘be slick’
Input: /n i
1
l h ʊ
2
– d a
3
/
MAX(ATR)-IO
Firsts
*COUPLE(nonhigh V, self deact. ATR)
IDENT(deactivation)-IO
Firsts
SELFDEACTIVATE
MAX(ATR)-IO
IDENT(deactivation)-IO
PERSIST(ATR)
a. [nilhu-də]
*! * *
b. [nilhʊ-da]
*
c. [nɪlhʊ-da]
*! *
In (111), the input contains an /i/ with a self-deactivating tongue root advancement
gesture. In candidate (a) [nilhu-də] this gesture surfaces as persistent, violating both
IDENT(deactivation)-IO constraints as well as SELFDEACTIVATE. The winning candidate (b)
[nilhʊ-da] faithfully realizes the tongue root advancement gesture as self-deactivating, crucially
satisfying high-ranked IDENT(deactivation)
Firsts
and violating only low-ranked PERSIST(ATR).
Tongue Body
palatal narrow
1
Tongue Body
uvular narrow
2
Tongue Body
palatal wide
3
Tongue Root
advanced
1
146
Again, candidate (c) [nɪlhʊ-da] deletes the tongue root advancement gesture and fatally violates
both versions of MAX(gesture)-IO. The ranking in (109), then, has successfully preserved the
contrast between self-deactivating and persistent tongue root advancement gestures in an initial
syllable.
When a high vowel occurs in a non-initial syllable, it never triggers harmony, regardless
of what its input specification for self-deactivation or persistence may be. This is demonstrated
by the tableau in (112) for [kani-ŋga] ‘agreeing.’ In order to demonstrate that a tongue root
advancement gesture will always surface as self-deactivating in a non-initial syllable, a
hypothetical input with a persistent tongue root advancement gesture is posited.
147
(112) Tableau for Classical Manchu [kani-ŋga] ‘agreeing’
Input: / k a
1
n i
2
– ŋ g a
3
/
MAX(ATR)-IO
Firsts
*COUPLE(nonhigh V, self deact. ATR)
IDENT(deactivation)-IO
Firsts
SELFDEACTIVATE
MAX(ATR)-IO
IDENT(deactivation)-IO
PERSIST(ATR)
a. [kani-ŋgə]
*!
b. [kani-ŋga]
* *
c. [kanɪ-ŋga]
*!
In (112), IDENT(deactivation)-IO
Firsts
is not relevant, as the tongue root advancement
gesture of [i] does not occur in the initial syllable of either candidate output. Therefore, in this
form the choice between candidates falls to the lower-ranked constraints. Candidate (a)
[kani-ŋgə], in which the tongue root advancement gesture of [i] is persistent and therefore a
trigger of harmony, fatally violates SELFDEACTIVATE. Winning candidate (b) [kani-ŋga], on the
Tongue Body
palatal wide
1
Tongue Body
palatal narrow
2
Tongue Body
palatal wide
3
Tongue Root
advanced
2
148
other hand, satisfies this constraint with its self-deactivating tongue root advancement gesture at
the expense of violating the low-ranked general IDENT(deactivation)-IO, as well as
PERSIST(ATR). Candidate (c) [kanɪ-ŋga] avoids violation of both the IDENT(deactivation)-IO
constraints and SELFDEACTIVATE by deleting the tongue root advancement gesture altogether,
but is ruled out by violation of the general version of MAX(gesture)-IO. This tableau
demonstrates that no matter a tongue root advancement gesture’s input deactivation parameter
setting, it will always surface as self-deactivating in a non-initial syllable. This accounts for the
inability of vowels to trigger ATR harmony outside of the initial syllable.
Turning now to the pattern of harmony triggering exhibited by nonhigh /ə/, the tableau in
(113) demonstrates the evaluation of the form [hərə-ku] ‘ladle.’ In order to show that /ə/ in an
initial syllable will always surface with a persistent tongue root advancement gesture and will
therefore always trigger harmony, a hypothetical input containing a self-deactivating tongue root
advancement gesture is posited.
149
(113) Tableau for Classical Manchu [hərə-ku] ‘ladle’
Input: / h ə
1
r a
2
– k ʊ
3
/
MAX(ATR)-IO
Firsts
*COUPLE(nonhigh V, self-deact. ATR)
IDENT(deactivation)-IO
Firsts
SELFDEACTIVATE
MAX(ATR)-IO
IDENT(deactivation)-IO
PERSIST(ATR)
a. [hərə-ku]
* * *
b. [həra-kʊ]
*! *
c. [hara-kʊ]
*! *
In (113), the tongue root advancement gesture in winning candidate (a) [hərə-ku] is
persistent and triggers ATR harmony, violating SELFDEACTIVATE as well as both
IDENT(deactivation)-IO constraints while satisfying highest-ranked MAX(ATR)-IO
Firsts
and
*COUPLE. The faithful candidate (b) [həra-kʊ] surfaces with a self-deactivating tongue root
advancement gesture, satisfying all faithfulness constraints but fatally violating high-ranked
Tongue Body
palatal wide
1
Tongue Body
palatal wide
2
Tongue Body
uvular narrow
3
Tongue Root
advanced
1
150
*COUPLE(nonhigh vowel, self-deact. ATR). This *COUPLE constraint requires a nonhigh ATR
vowel to trigger harmony, regardless of the input specification of its tongue root advancement
gesture. Both IDENT(deactivation)-IO constraints are outranked by *COUPLE and therefore are
unable to preserve any underlying contrast in triggering ability among nonhigh ATR vowels.
MAX(ATR)-IO
Firsts
is fatally violated by candidate (c) [hara-kʊ], demonstrating that in the initial
syllable, violation of *COUPLE is avoided by altering a tongue root advancement gesture’s
deactivation parameter rather than by deleting the offending gesture.
Finally, the tableau in (114) for the word [hʊla-ŋga] ‘crying’ includes the hypothetical
input /hʊlə-ŋga/, which contains an ill-formed word-medial /ə/. In the input, its tongue root
advancement gesture is self-deactivating, but this is an arbitrary decision and does not affect the
outcome of EVAL.
151
(114) Tableau for [hʊla-ŋga] ‘crying’
Input: / h ʊ
1
l ə
2
- ŋ g a
3
/
MAX(gesture)-IO
Firsts
*COUPLE(nonhigh V, self deact. ATR)
IDENT(deactivation)-IO
Firsts
SELFDEACTIVATE
MAX(gesture)-IO
IDENT(deactivation)-IO
PERSIST(ATR)
a. [hʊlə-ŋgə]
*! *
b. [hʊlə-ŋga]
*! *
c. [hʊla-ŋga]
*
Because no tongue root gesture appears in the initial syllable of any of the candidates in
(114), MAX(ATR)-IO
Firsts
and IDENT(deactivation)-IO
Firsts
do no work here. Candidate (a)
[hʊlə-ŋgə] contains a persistent tongue root gesture in a non-initial syllable, fatally violating
SELFDEACTIVATE, as well as low-ranked general IDENT(deactivation)-IO. Candidate (b)
[hʊlə-ŋga], in which the tongue root gesture is self-deactivating, satisfies SELFDEACTIVATE but
Tongue Body
uvular narrow
1
Tongue Body
palatal wide
2
Tongue Body
palatal wide
3
Tongue Root
advanced
2
152
fatally violates *COUPLE, which prohibits self-deactivating tongue root advancement gestures
from co-occurring with nonhigh vowel gestures. Winning candidate [hʊla-ŋga] (c) satisfies both
of these constraints by deleting the tongue root gesture, causing underlying /ə/ to surface as [a].
With the ranking of constraints in (109), this analysis is able to account for the distributions of
both high and nonhigh vowels in Classical Manchu.
The only exception to the generalization that [ə] does not occur outside of an initial
syllable is when it does so as a result of undergoing harmony. The reason for this is apparent in
several of the winning candidates already presented. When a nonhigh vowel follows a triggering
ATR vowel in an initial syllable, it surfaces as ATR not because it is coupled to its own tongue
root advancement gesture, but because it has been overlapped by the tongue root advancement
gesture of an ATR vowel in an initial syllable. The same situation arose in the analyses of
Kyrgyz rounding harmony (section 3.2.1), in which round vowels may only occur in non-initial
syllables as the result of harmony, and Baiyina Oroqen rounding harmony (section 3.4.2), in
which the same positional restriction is placed on nonhigh round vowels.
Classical Manchu tongue root harmony presents an exceptionally complex pattern of
harmony triggering. High ATR vowels display contrastive triggering of tongue root harmony,
while nonhigh ATR vowels are required to always trigger harmony. This is combined with the
restriction of harmony triggers to the privileged position of the initial syllable. The Gestural
Harmony Model successfully accounts for all of this through the interaction of constraints on a
gesture’s deactivation parameter (SELFDEACTIVATE), general and positional faithfulness
constraints (MAX(ATR)-IO and IDENT(deactivation)-IO), and constraints on gestural co-
occurrence (*COUPLE(nonhigh vowel, self-deactivating ATR)). To fully appreciate the
complexity of this system, compare it with one of the simple cases of harmony triggering
153
presented in section 3.2., which rely only on the high ranking of PERSIST(Gest
X
) or
ANTICIPATE(Gest
X
). In Classical Manchu, on the other hand, PERSIST(ATR) occupies the lowest
constraint stratum. This is indicative of the importance of viewing the presence of harmony as
the result of the interaction between markedness and faithfulness constraints such that they shape
phonological inventories to include segments with harmony-triggering gestures, and of
restricting those segments to certain positions. A constraint that simply requires harmony, be it
PERSIST(Gest
X
) in a gestural framework or SPREAD(F) in a featural framework, is unable to
capture the full range of complexity exhibited by patterns of harmony triggering.
3.6 Alternative Accounts of Harmony Triggering
The Gestural Harmony Model’s characterization of the ability to trigger harmony as a
property that is encoded within the representation of a trigger is unique among mainstream
approaches to the analysis of harmony. This approach is crucial to its success in capturing
complex patterns of harmony triggering, in particular those that involve the idiosyncratic
triggering of harmony. In the Gestural Harmony Model, such patterns are analyzed as the result
of a language’s surface phonological inventory being shaped such that it preserves the contrast
between triggering and non-triggering versions of a segment whose gestural makeup may be
otherwise identical. This can be achieved via the high ranking of faithfulness that preserves
gestures’ persistence and/or activation parameters.
In this section I examine several alternative methods of driving harmony, both gesture-
based and feature-based, and show that they are unable to account for patterns of contrastive
triggering without invoking additional theoretical mechanisms. The alternative gestural model I
consider in this section lacks the mechanisms of gestural persistence and anticipation that are
crucial to the Gestural Harmony Model’s success in accounting for cases of contrastive
154
triggering. In the case of feature-based theories, meanwhile, commonly used mechanisms for
dealing with idiosyncratic or exceptional application of phonological processes including (1)
constraint indexation and (2) featural pre- and underspecification make several undesirable
typological predictions, both over- and undergenerating patterns of harmony triggering.
3.6.1 Gestural Spreading via Direct Activation Manipulation
In accounting for patterns of contrastive triggering of harmony, the Gestural Harmony
Model distinguishes itself from the conception of harmony driving in a similar gestural model
proposed by Tejada (2012). In Tejada’s theory of gestural spreading based on direct activation
manipulation, tone spreading across multiple syllables is proposed to be the result of the failure
of a tone gesture to deactivate at a specified phase. This failure of a tone gesture to deactivate
itself is enforced by the constraint *SELF-DEACTIVATE-BY-CLOCK,
21
which penalizes gestural
self-deactivation. In contrast, the constraint *SPAN-SYLL penalizes a gesture that extends its
period of activation beyond a single syllable. Similar, but not identical, constraints are adopted in
the Gestural Harmony Model: PERSIST(Gest
X
) penalizes a self-deactivating gesture of type X,
while SELFDEACTIVATE penalizes any persistent (non-self-deactivating) gesture. Though these
sets of constraints appear quite similar, there are crucial differences in how they are
implemented, how they interact, and the typologies they generate.
Because Tejada’s framework
assumes that all regressive (leftward) harmony is the result of movement, the discussion in this
section of how her model compares to the Gestural Harmony Model focuses only on progressive
harmony.
As discussed throughout this chapter and chapter 2, in the Gestural Harmony Model,
gestures may be either self-deactivating or persistent (non-self-deactivating). Progressive
21
Gestural planning oscillators are often referred to as ‘clocks.’
155
(rightward) harmony occurs in languages in which the language’s surface phonological inventory
includes persistent gestures. A persistent gesture will always surface in a language when
PERSIST(Gest
X
) outranks both SELFDEACTIVATE and the faithfulness constraint
IDENT(deactivation)-IO. Similarly, a self-deactivating gesture will surface in a language with the
ranking of SELFDEACTIVATE over PERSIST(Gest
X
) and IDENT(deactivation)-IO. A third
possibility is that the faithfulness constraint will outrank both SELFDEACTIVATE and
PERSIST(Gest
X
), allowing both types of gestures to surface and potentially to contrast. This third
possible ranking allows the Gestural Harmony Model to account for cases of contrastive
harmony triggering, including the nasal harmony systems of Acehnese and Rejang (section 3.3)
and ATR harmony in Classical Manchu (section 3.5).
In the direct activation manipulation approach to tone spreading, however, tone gestures
are not specified as either persistent or self-deactivating. As a result, the constraints *SELF-
DEACTIVATE-BY-CLOCK and *SPAN-SYLL do not evaluate the type of tone gesture present in a
form, and there are no faithfulness constraints that preserve a gesture’s underlying deactivation
specification. Instead, these constraints directly drive gestures to be active for certain periods of
time. *SPAN-SYLL is satisfied by a tone gesture whose period of activation is equal in length to
that of the vocalic gesture to which it is coupled. It does not matter if this single-syllable
activation of the tone gesture is the result of its self-deactivation or the result of blocking by a
tone gesture in the following syllable. Likewise, *SELF-DEACTIVATE-BY-CLOCK is satisfied by a
tone gesture whose period of activation is equal in length to that of the vowel gesture to which it
is coupled, provided that it is immediately followed by a blocking tone gesture in the next
syllable. Again, it does not matter if this single-syllable activation of the tone gesture is the result
of self-deactivation or of external deactivation; in accordance with *SELF-DEACTIVATE-BY-
156
CLOCK, it has deactivated in a position adjacent to a blocker. While the Gestural Harmony
Model’s constraints PERSIST(Gest
X
) and SELFDEACTIVATE operate to produce gestures of
different types, those either with or without the potential to trigger harmony, the use of *SELF-
DEACTIVATE-BY-CLOCK and *SPAN-SYLL represent an analysis in which there is a more direct
manipulation of the period of gestural activation within output candidates.
The Gestural Harmony Model’s reliance on constraints against certain types of gestures,
either self-deactivating or persistent, rather than constraints that directly manipulate a gesture’s
period of activation, presents significant advantages in analyzing more complex patterns of
triggering, especially contrastive triggering. By viewing progressive harmony as the result of the
presence of a persistent gesture in a phonological form, the presence or absence of harmony in a
language is reduced to the relatively simple matter of inventory shaping via the interaction of
markedness and faithfulness constraints that reference a gesture’s deactivation parameter. The
patterns of harmony triggering that some languages exhibit, then, are the result of constraining
the distributions of the self-deactivating and persistent gestures in a language’s phonological
inventory.
The high ranking of IDENT(deactivation)-IO and/or IDENT(activation)-IO is crucial to
building an accurate account of cases of contrastive triggering, including the nasal harmonies of
Acehnese and Rejang and the ATR harmony of Classical Manchu. In addition, the ability to
relativize this faithfulness constraint to privileged positions accounts for the distinct
distributional properties of triggering and non-triggering segments in these languages. Using
only the markedness constraints *SELF-DEACTIVATE-BY-CLOCK and *SPAN-SYLL, it is unclear
how such attested patterns of triggering could be generated without adopting additional
theoretical mechanisms, such as constraint indexation (Pater 2000, 2009a; Flack 2008; Becker
157
2009; Finley 2010). In doing so, the direct activation manipulation approach to gestural
spreading would essentially be duplicating the results of the Gestural Harmony Model with
respect to cases of idiosyncratic triggering, albeit via the inclusion of an arbitrary specification of
a gesture (whether it bears an index to a harmony-driving constraint) instead of a meaningful
gestural parameter (self-(de)activation). Furthermore, it is unclear whether the positional
restrictions on the contrast between triggers and non-triggers of harmony, exhibited in Acehnese,
Rejang, and Classical Manchu, could be captured by a model that employs constraint indexation.
3.6.2 Constraint Indexation
In a feature-based framework, the use of constraint indexation to capture patterns of
idiosyncratic triggering of harmony encounters several issues. In harmony, exceptional triggering
can be achieved in a feature-based theory via indexation to a harmony driving constraint, such as
SPREAD(F) (Padgett 1995; Walker 1998/2000). However, this approach both over- and
undergenerates patterns of harmony triggering, rendering it an undesirable alternative to the
Gestural Harmony Model. The reasons for this are examined in this section. Though the
discussion focuses on the constraint SPREAD(F), it should be noted that these issues extend to all
analyses that rely on maximal harmony drivers that favor the association of a harmonizing
feature to all segments in a word.
The constraint SPREAD(F), as well as similarly evaluated harmony driving constraints
such as EXTEND(F) and ALIGN(F), drives unbounded harmony by penalizing non-undergoers,
i.e., those segments that are not associated to a harmonizing feature F. It is defined as in (115).
158
(115) SPREAD(F): For each feature F associated to a segment in domain D, assign a violation
mark for every segment S in domain D that is not associated to F. (adapted from Walker
(1998, p. 47))
If a candidate form contains the feature F, it will be subject to SPREAD(F), and any
segments not associated to the feature F in some domain D (often the entire word) will incur a
violation. When SPREAD(F) dominates IDENT(F)-IO, which preserves potential undergoers’
underlying values for the feature F, harmony results.
In order to capture the patterns of harmony triggering in a language such as Classical
Manchu in which only some forms show within- and across-morpheme harmony, constraint
indexation can be used to relativize SPREAD(F) such that it does not apply to all forms. Instead,
an indexed constraint will only be relevant when a word contains a morpheme that shares its
index, and will only be violated if some portion of an indexed morpheme contributes to its
violation. In harmony, patterns in which some forms display harmony while others do not can be
analyzed by indexing triggering morphemes to a version of SPREAD(F) that outranks IDENT(F)-
IO, which in turn outranks a general version of SPREAD(F). This ranking is shown in (116).
(116) Constraint ranking for idiosyncratic triggering of harmony with indexed constraints
SPREAD(F)
i
>> IDENT(F)-IO >> SPREAD(F)
When the domain of SPREAD(F) is specified to refer to the entire word, the constraint
compels both within- and cross-morpheme harmony. This is true of both the general SPREAD(F)
constraint and the version that is indexed only to certain morphemes. Crucially, it is not the case
that the entire structural description of an indexed constraint must be met only by material from
an indexed morpheme. Rather, Pater (2009a) specifies that an indexed constraint is violated if its
structural description contains some part of an indexed morpheme.
159
Tableaux that demonstrate the workings of the ranking in (116) are provided in (117) and
(118). Both include forms in which two segments S
1
and S
2
occupy a root while segment S
3
occupies a suffix attached to that root. In (118), the root morpheme is indexed to high-ranked
SPREAD(F)
i
. In both the input and output, brackets indicate morpheme boundaries.
(117) Tableau with harmony triggered by an indexed morpheme
Input: SPREAD(F)
i
IDENT(F)-IO SPREAD(F)
a.
*!* **
b.
**
In the tableau in (117), the morpheme containing the potential trigger segment S
1
, which
is associated with feature F, is indexed to high-ranking SPREAD(F)
i
. Output candidate (a) violates
both the indexed and non-indexed versions of SPREAD(F) due to the lack of association lines
between the harmonizing feature F and segments S
2
and S
3
. Winning candidate (b) satisfies both
SPREAD constraints but violates IDENT(F)-IO due to the added association lines between feature
F and segments S
2
and S
3
.
The situation is different in the tableau in (118), in which the morpheme containing the
potential trigger segment S
1
bears no index.
160
(118) Tableau with harmony not triggered by an unindexed morpheme
Input: SPREAD(F)
i
IDENT(F)-IO SPREAD(F)
a.
**
b.
*!*
In (118), winning candidate (a), which displays no harmony, violates only low-ranked
SPREAD(F). This is because the morpheme containing the potential trigger segment S
1
bears no
index, and is therefore not subject to high-ranked SPREAD(F)
i
. Harmonizing candidate (b) fatally
violates IDENT(F) due to the added association lines between feature F and segments S
2
and S
3
.
With this ranking, then, a system in which only morphemes indexed to a high-ranked version of
SPREAD(F) display harmony, while other morphemes do not, can be generated straightforwardly.
However, one major drawback of the constraint indexation approach to idiosyncratic
harmony triggering is that it is unable to generate certain aspects of the distributions of triggers
and non-triggers of harmony. In Acehnese, Rejang, and Classical Manchu, the contrast between
triggering and non-triggering segments that bear a harmonizing property is restricted to a
privileged syllable (initial in Classical Manchu, final in Acehnese and Rejang), and neutralized
elsewhere. This sort of preservation of phonological contrast only in privileged positions is
common, and can be accounted for using positional faithfulness constraints (Beckman 1997,
1998; Hyman 1998; Walker 2005, 2011). The Gestural Harmony Model, with its focus on
analyzing harmony triggering patterns as the result of constraints on the positions in which self-
161
deactivating and persistent (or non-anticipatory and anticipatory) gestures may occur in a word,
is well suited to accounting for such patterns.
Indexation of a harmony-driving constraint, on the other hand, encounters difficulty in
generating the different distributional patterns of triggering and non-triggering segments. In a
framework that assumes that harmony-driving constraints may only be indexed to morphemes
(e.g., Pater 2000, 2009a; Becker 2009), this is unsurprising. A segment bearing feature F that
occurs in a morpheme indexed to SPREAD(F)
i
or some other harmony-driving constraint will
trigger harmony no matter its position in the morpheme.
The adoption of segmental rather than morphemic indexation to constraints, as proposed
by Temkin Martínez (2010), does not solve this issue. Markedness and faithfulness constraints,
whether they are relativized to privileged positions or not, can reference properties of
phonological units, such as feature values or gestural parameter settings. For instance, the
analysis of Classical Manchu in section 3.5 utilizes the constraint IDENT(deactivation)-IO
Firsts
to
preserve the contrast between persistent and self-deactivating tongue root advancement gestures
only in the first syllable of a word. However, there is no parallel approach that preserves the
contrast between segments that are indexed to a harmony driving constraint and those that are
not (e.g., /S/ vs. /S
i
/). This is because indices to constraints are not objects that can be referenced
by constraints in the same way that properties of features or gestures can be referenced. That is,
there are no constraints such as IDENT(i)-IO or IDENT(i)-IO
Firsts
, where i is an index to a
harmony-driving constraint. The index i on a morpheme or segment carries no distinctive
information on its own; it is only defined if there is some constraint in the grammar, such as
SPREAD(F)
i
, that shares that indexation. An index conveys information about the relation
between a phonological unit and some element of the phonological grammar, not about the
162
phonological unit itself. Thus, markedness and faithfulness constraints should not be able to
reference any constraint indices a phonological unit might bear, and would therefore be of no use
in determining the distribution of exceptional harmony triggers, i.e., segments bearing an index
to SPREAD(F)
i
. Indexation to standard harmony-driving constraints, therefore, is unable to
generate patterns of harmony triggering in which distributional restrictions are placed on triggers
and non-triggers of harmony.
The constraint indexation approach also encounters an issue of overgeneration of possible
harmony triggering patterns. This issue arises when we consider that it is not only morphemes
that contain the trigger of harmony that can be indexed to the constraint SPREAD(F)
i
. Because the
structural description of the constraint SPREAD(F) may contain segments from two different
morphemes when the domain is specified to be the entire word, there are two morphemes that
might potentially bear an indexation to the constraint. That is, either the trigger or the undergoer
of harmony may be in an indexed morpheme that activates the indexed SPREAD(F)
i
constraint.
This means that the SPREAD(F) constraint depicted graphically in (119) is relevant when either
segment S (either the potential trigger or the potential undergoer) in the structural description is
part of an indexed morpheme.
(119) Structural description of SPREAD(F)
22
22
In the following discussion, I assume that the constraint SPREAD(F) assigns violations each time the structure
illustrated in (119) is present, i.e. whenever a segment S is unassociated with feature F in domain D. Furthermore,
what Pater refers to as the locus of violation I take to mean the structural description. I do not assume, as in
McCarthy (2003a), that the structural description of a constraint should be divided into a locus of violation and its
context.
163
The problem arises when a morpheme containing a potential undergoer of harmony is
indexed to SPREAD(F)
i
while a morpheme containing a potential trigger of harmony is not
indexed. This is demonstrated in the tableau in (120), in which the triggering segment S
1
bearing
feature F is still in the root, but the indexation to SPREAD(F)
i
is now borne by the affix containing
S
3
. This situation results in harmony across the root, despite the fact that the root would surface
as disharmonic if it did not have an indexed affix attached to it, as in the tableau in (118).
(120) Tableau with harmony triggered across an unindexed morpheme due to indexed affix
Input:
SPREAD(F)
i
IDENT(F)-IO SPREAD(F)
a.
*!* **
b.
*!*
In candidate (a) the lack of harmony for the feature F both within and across the
morpheme boundary violates both versions of SPREAD(F). Most crucially, the high-ranked
indexed version of SPREAD(F)
i
is violated because the affix containing S
3
is indexed to it. The
winning candidate (b) satisfies high-ranked SPREAD(F)
i
by spreading the feature F across the root
and onto S
3
in the indexed affix. In candidate (b), the segment S
1
has triggered harmony in order
to satisfy SPREAD(F)
i
despite the fact that S
1
does not even occur in the morpheme bearing the
indexation to that constraint. Therefore, while this hypothetical language usually has no within-
morpheme harmony (due to the ranking of IDENT(F)-IO over the general SPREAD(F) constraint),
within-morpheme harmony has occurred in candidate (b) of (120) so that the feature F can
spread to an indexed morpheme, thereby satisfying high-ranked SPREAD(F)
i
. Such a pattern is
164
unattested according to Finley (2010), who claims that even in the presence of an affix that is an
exceptional undergoer of harmony, harmony is never induced within an otherwise disharmonic
stem. Morpheme indexation to a harmony-driving constraint has therefore generated an
unattested, pathological pattern of harmony.
Also working against a morpheme indexation analysis of so-called exceptional triggering
of harmony is the fact that in Rejang (section 3.3), triggering and non-triggering nasals may
occur in the same morpheme, as in [mĩnae] ‘come here,’ in which the [m] is a triggering nasal
and the [n] is a non-triggering nasal. If a morpheme as a whole were indexed to a harmony
driving constraint, there would be no way to account for the spreading of nasality from one nasal
consonant and not another. Instead, an analysis of Rejang would require some way to index a
harmony-driving constraint to a particular triggering [nasal] feature or segment. Temkin
Martínez (2010) proposes that certain patterns of phonological idiosyncrasy operate on the
segmental rather than the morphemic level, indicating that it is necessary to allow for the
possibility of segmental constraint indexation. A segment indexation solution to nasal harmony
triggering in Rejang essentially captures the same contrast between triggers and non-triggers as
the Gestural Harmony Model. In either case, there is some underlying property of a nasal (either
indexation to a harmony-driving constraint or a gestural deactivation parameter) that determines
whether or not it triggers harmony. However, this approach generates the same pathology seen in
the tableau in (120). If SPREAD(F)
i
were indexed to segment S
3
rather than the morpheme that
contains S
3
, the same pathological pattern of triggering would be generated.
With respect to patterns in which the ability to trigger harmony appears to be a
contrastive property, the Gestural Harmony Model successfully avoids the issues that beset
analyses based upon mechanisms for exceptionality in feature spreading. Because gestural
165
representations include parameters for persistence and anticipation, these parameters can be
referenced by markedness and faithfulness constraints that determine different gestures’
distributions. In particular, these parameters are subject to IDENT(parameter
X
)-IO, which allows
the persistence and anticipation parameters of a gesture to act as contrastive properties. This is a
novel approach to accounting for patterns of harmony triggering that have previously fallen
within the realm of phonological exceptionality. Analyses of such apparent exceptionality in
harmony triggering that are based in featural frameworks fall short of the success of the Gestural
Harmony Model. In particular, analyses that rely on morpheme or segment indexation to
harmony-driving constraints suffer from issues of both over- and undergeneration of patterns of
harmony triggering.
While the Gestural Harmony Model obviates the need for constraint indexation in
accounting for patterns of apparent exceptionality in harmony triggering, I leave open the
question of what the general status of such indexation is in the model. At present, it is not clear
whether contrasts for gestural parameters that are unavailable within featural representations
(such as persistence versus self-deactivation) can wholly replace the results of constraint
indexation. However, there are additional cases of apparent exceptionality that can be reanalyzed
as cases of phonological contrast for a gestural parameter setting. This subject is discussed
further in sections 5.3 and 5.4, in which contrasts based on gestural blending strength are
considered.
3.6.3 Pre- and Underspecification
Another approach to idiosyncrasy in the application of phonological processes relies on
proposing distinctions in the feature specifications of input forms that are neutralized in output
forms. By proposing that segments may be either prespecified or underspecified for a relevant
166
feature in the input, such theories are able to account for the idiosyncratic undergoing of
phonological processes, and also correctly predict that idiosyncratic phonological behavior is
granular at the level of the segment, rather than the morpheme. It is also possible to utilize
distinctions in input specification to account for the idiosyncratic triggering of harmony;
however, such an approach involves the adoption of additional theoretical mechanisms, either
within the phonological grammar or the post-phonological mechanism of phonetic
implementation.
The underlying specification of segments for a relevant spreading feature have long been
used to account for the blocking of feature spreading (see, for example, work in Autosegmental
Phonology by Clements (1976a, 1981) and van der Hulst & N. Smith (1982a, 1982b)). Within
the study of harmony, underlying feature specification has often been used in analyses of
idiosyncratic undergoing and blocking of harmony. Under such analyses, undergoers and
blockers of harmony are distinguished based on whether or not they are already specified for the
harmonizing feature. These include analyses of Turkish backness harmony (Clements & Sezer
1982), the nasal harmonies of Tucano (Noske 1995) and Tuyuca (Barnes 1996), Kalenjin ATR
harmony (Ringen 1988), and others.
Inkelas (1994) and Inkelas, Orgun, & Zoll (1997) propose that this featural pre- and
underspecification approach can be adopted within parallel OT in order to account for patterns of
apparent exceptionality in the application of phonological processes. Under their analysis,
segments may be either pre- or underspecified for a feature in the input, leading to a potential
surface contrast in whether they are targeted by a phonological process. Whether this contrast
arises is due to the interaction of markedness and faithfulness constraints. When faithfulness
constraints referencing some feature specification are ranked high, the featural makeup of
167
prespecified segments will be preserved, while underspecified segments will be permitted to
alternate for that feature specification. Apparently exceptional non-undergoers of phonological
processes, then, are those that are prespecified for a relevant feature in the input, and their failure
to undergo a process is due to the preservation of that feature specification by faithfulness
constraints. The following tableaux demonstrate this approach to apparent exceptionality in
undergoing a harmony process.
(121) Tableau with harmony blocked by a segment prespecified for feature F
Input: / /
FAITH(F)-IO SPREAD(+F)
a.
*
b.
*!
In (121), the input contains two segments that are specified as [+F] and [-F], respectively.
In the winning candidate (a), no feature spreading occurs, and both segments surface with their
underlying feature specifications. This violates SPREAD(+F), but satisfies higher-ranked
FAITH(F)-IO. In candidate (b), the feature [+F] spreads onto segment S
2
, which loses its
underlying [-F] specification in fatal violation of FAITH(F)-IO. When a potential undergoer of
harmony is unspecified for the feature F in the input, however, the outcome is different; this is
illustrated by the tableau in (122).
168
(122) Tableau with harmony undergone by a segment unspecified for feature F
Input: / /
FAITH(F)-IO SPREAD(+F)
a.
* *!
b.
*
In (122), the input contains an S
2
that is unspecified for feature F. In both output
candidates, S
2
is specified for F and as a result both violate FAITH(F)-IO. It then falls to the
constraint SPREAD(+F) to determine the winner. This winner is spreading candidate (b), as its
specification of S
2
satisfies SPREAD(+F). Candidate (a), on the other hand, specifies S
2
for the
feature value [-F], violating SPREAD(+F). The tableaux in (121) and (122), then, demonstrate
how two segments that are potential undergoers of a feature spreading process can be
idiosyncratically targeted by (or can idiosyncratically block) harmony based on whether or not
they are specified for some feature in the input.
However, the picture is more complicated when attempting to use the pre- and
underspecification of input forms to account for the contrastive triggering of harmony. Within
derivational theories of phonology, such as those relying on ordered rules, distinguishing triggers
and non-triggers of harmony in this way produces the correct outcome. In Classical Manchu, for
instance, ATR harmony would be triggered by underlyingly ATR /i/ and /u/, while unspecified
/I/ and /U/ would be unable to trigger ATR harmony. Only after harmony has taken place would
/I/ and /U/ become specified as ATR, too late for them to serve as triggers. This is a case of
counterfeeding opacity with eventual absolute neutralization of an underlying contrast between
169
triggering /i/ and /u/ and non-triggering /I/ and /U/. This approach is illustrated in the derivation
in (123).
(123) Contrastive triggering of ATR harmony via counterfeeding opacity
Input /i-a/ /I-a/
ATR Harmony i-ə —
High Vowels → [ATR] — i-a
Output [i-ə-ə] [i-a-a]
These sorts of crucial rule orderings resulting in neutralization of an input contrast are
incompatible with output-oriented frameworks such as OT or Harmonic Serialism (McCarthy
2000, 2008a, 2008b; McCarthy & Pater 2016). In both of these frameworks, if a segment [i] is
determined by constraint rankings in the grammar to be a trigger of ATR harmony, then that
grammar will select a winning output candidate in which [i] triggers harmony, no matter its
specification in the input (or in any earlier stage of the derivation in the case of Harmonic
Serialism). In their most basic forms, these frameworks do not permit triggering by an [i] that
was previously specified as ATR in the input while prohibiting triggering by an [i] that was
previously /I/. The tableau in (124) demonstrates.
170
(124) Tableau with harmony triggered by underlyingly specified or unspecified segment
Input: / /
FAITH(ATR)-IO SPREAD(+ATR)
a. [ ]
* *!
b. [ ]
*
Input: / I A / FAITH(ATR)-IO SPREAD(+ATR)
a. [ ]
** *!
b. [ ]
**
In (124), the first input contains a high [+ATR] vowel followed by a low vowel
unspecified for [ATR]. Its violation profile parallels that of the tableau in (122). In each output
candidate, the low vowel receives specification for [ATR], incurring a violation of
FAITH(ATR)-IO. Therefore, the constraint SPREAD(+ATR) determines the winner. In candidate
(a), [+ATR] does not spread from the high to the low vowel, violating SPREAD(+ATR). In
winning candidate (b), SPREAD(+ATR) is satisfied. The same outcome results for the second
input in (124), in which the high and low vowel are both unspecified for [ATR]. Both output
candidates, in which these vowels have received specification for [ATR], incur two violations of
FAITH(ATR)-IO, and it again falls to the constraint SPREAD(+ATR) to determine the winner.
Again, it is candidate (b), in which the high vowel’s [+ATR] specification has spread onto the
low vowel. This result demonstrates that even when a constraint such as FAITH(ATR)-IO is
171
ranked high, a surface ATR vowel will be compelled to trigger ATR harmony, regardless of
whether it was specified for this feature in the input.
A possible solution to the incompatibility between the counterfeeding opacity account of
contrastive triggering in (123) and output-oriented, constraint-based grammars is the adoption of
some theoretical mechanism that would prevent harmony that is triggered by a feature that is
inserted into an output form. Under such an approach, harmony would be triggered by segments
specified for a harmonizing feature in an input form, and would not be triggered by segments
unspecified for that feature, even if they came to be specified in the output. One way to achieve
this is through constraint conjunction (Smolensky 1993; Smolensky & Legendre 2006).
Segments that are underspecified for a harmonizing feature in the input could be prevented from
triggering harmony via the conjunction of a constraint that penalizes feature insertion, such as
DEP(F)-IO, with a constraint that is violated when harmony takes place, such as IDENT(F)-IO,
which would be violated by the undergoers of harmony. With this constraint ranked above a
constraint driving harmony, harmony would be triggered only by those segments in an output
candidate that do not insert a feature before spreading it.
Another possibility is the adoption of one of the theoretical mechanisms designed
specifically to capture cases of apparent derivational opacity within constraint-based
frameworks, such as Sympathy Theory (McCarthy 1999) or comparative markedness (McCarthy
2003b). I illustrate this point within comparative markedness theory. Within this theory,
violations marks for markedness constraints are assigned to candidates based on whether their
violations are old (present in both the input and the output candidate) or new (present in the
output candidate only). Markedness constraints come in pairs, with one assigning marks for old
violations and one assigning marks for new violations. According to McCarthy, cases of
172
apparent counterfeeding opacity can be captured by a constraint ranking in which a constraint
against old markedness violations is ranked above a relevant faithfulness constraint, which in
turn outranks a constraint against new markedness violations. The relevant ranking for deriving
harmony that is triggered by a vowel that is underlyingly specified for [ATR] but not an
underlyingly unspecified vowel is provided in (125).
(125) Constraint ranking for idiosyncratic triggering of harmony with comparative markedness
SPREAD(ATR)
old
>> IDENT(ATR)-IO >> SPREAD(ATR)
new
The workings of this constraint ranking are demonstrated in the tableau in (126).
(126) Contrastive triggering of ATR harmony via input underspecification and comparative
markedness
Input: /I-a-a/
SPREAD(ATR)
old
IDENT(ATR)-IO SPREAD(ATR)
new
a. [i-a-a]
**
b. [i-ə-ə]
*!*
Input: /i-a-a/ SPREAD(ATR)
old
IDENT(ATR)-IO SPREAD(ATR)
new
a. [i-a-a] *!*
b. [i-ə-ə] **
In (126), the first input contains a high vowel that is unspecified for the feature [ATR];
both output candidates contain the fully specified vowel [i]. Winning candidate (a), in which [i]
does not trigger harmony, incurs two violations of SPREAD(ATR), as it includes an [ATR] feature
that is not associated with the two [a] vowels. However, this is a new violation of SPREAD(ATR)
that is created by the insertion of [ATR] into the output form; therefore, the violation is incurred
by low-ranked SPREAD(ATR)
new
. Candidate (b), in which harmony does occur, violates higher-
ranked IDENT(ATR)-IO by changing the [ATR] values of the /a/ vowels. In contrast, the second
input contains an /i/ vowel that is specified as [ATR]. Now, non-triggering candidate (a) violates
173
high-ranked SPREAD(ATR)
old
, as the violations of SPREAD(ATR) incurred by the non-undergoing
[a] vowels are present in the input as well. As a result, harmonizing candidate (b) is the winner.
The pre- and underspecification of features, then, can be compelled to account for cases
of idiosyncratic harmony triggering within output-oriented models of phonology such as OT,
provided that additional theoretical architecture is also implemented. This was demonstrated
using comparative markedness; however, it is generalizable to approaches utilizing constraint
conjunction or Sympathy Theory as well. A major drawback of these approaches, however, is
that they are quite powerful and make less constrained typological predictions about patterns of
harmony and the application of phonological processes more generally. The ability of
independently motivated individual constraints to produce unattested patterns when conjoined,
for instance, has been discussed by Itô & Mester (1998), Pater (2009b), and others. Relatedly, the
adoption of comparative markedness is accompanied by a doubling of all of the markedness
constraints in CON; every markedness constraint M in standard markedness theory corresponds
to the constraints M
old
and M
new
in comparative markedness theory. The full range of typological
predictions made by such an explosion of the constraint set assumed within OT has not been
fully explored, though see Łubowicz (2003) and T. Hall (2007) for some discussion of this issue
and arguments against comparative markedness based on typological predictions.
In contrast with these approaches that rely on the surface neutralization of an input
distinction between segments that are specified versus unspecified for a harmonizing feature, the
Gestural Harmony Model’s account of idiosyncratic triggering relies on a contrast in the setting
of a gestural deactivation and/or activation parameter that persists from input to output. Relying
on such a parameter to account for idiosyncratic triggering of harmony is thus entirely
174
compatible with output-oriented frameworks such as OT without invoking additional theoretical
mechanisms such as Sympathy Theory, comparative markedness, or constraint conjunction.
An alternative strategy for deriving idiosyncratic harmony triggering from pre- and
underspecified inputs is to assume that a non-triggering segment is not specified for a
harmonizing feature in either input or output forms. Underspecification in the phonological
output is proposed by, among others, Avery & Rice (1989), Itô, Mester, & Padgett (1995), and
Rice (1996). Under this approach, the contrast in Acehnese and Rejang between a harmony-
triggering nasal consonant and a non-triggering nasal consonant would be preserved between
input and output. Likewise, the contrast in Classical Manchu between a harmony-triggering ATR
[i] and a non-triggering [I], and between a triggering ATR [u] and a non-triggering [U], would be
preserved. In addition, the contrast between non-triggering [U] and non-ATR [ʊ] in Classical
Manchu must also be preserved. This would require that a later grammatical component that
derives fully specified output forms and/or the phonetic implementation of the forms output by
the phonological grammar, either (1) supply a [+F] specification to any segments unspecified for
a harmonizing feature, or (2) automatically produce unspecified segments as [+F]. [+F]
corresponds with [+nasal] in the case of Acehnese and Rejang, and with [+ATR] in the case of
Classical Manchu.
There are a variety of mechanisms that have been proposed to operate after the
phonological grammar has produced an output form. Phonetic implementation models such as
those proposed by Keating (1988, 1990) and Cohn (1990) do not fill in or alter output
phonological forms, but instead determine the states of the vocal tract required to produce
specified segments and interpolate between these specified segments. Other models, such as
those proposed by Zsiga (1993, 1997, 2000) and Rice (1996), fill in underspecified feature
175
values to produce fully specified forms. Zsiga’s model also involves the translation between
feature-based phonological surface forms and gesture-based phonetic forms, as well as language-
specific rules that manipulate the coordination relations between gestures. Rice, meanwhile,
proposes a phonetic implementation mechanism in which individual languages may choose
whether to implement consonants unspecified for place in the phonological output as either
coronal or velar. Both proposals are rooted in the idea that rules apply to phonological outputs to
provide them with default specifications, but that these defaults are determined on a language-
specific basis.
Both Zsiga and Rice propose models of the phonetic implementation mechanism that not
only produce a phonetic form from a phonological form, but do so based on the application of
language-particular rules. In this sense, each of these mechanisms is akin to a second phonetic
grammar that follows the first phonological grammar. Within these theories, the output of
phonology is even more abstract than generally assumed; the phonological grammar does not
actually produce forms that represent the end state of a derivation, but rather intermediate forms.
The adoption of such a stratal framework in order to account for the contrastive triggering of
phonological processes, including harmony, thus introduces an additional level of complexity to
our theory.
This can be contrasted with the Gestural Harmony Model’s account of idiosyncratic
harmony triggering via contrastive settings of the parameters of gestural self-activation and self-
deactivation. The model relies on a contrast between triggering and non-triggering segments that
is present in both input and output phonological forms, obviating the adoption of any sort of
multi-step derivational component within the phonological and/or phonetic grammar. As
discussed in section 1.2.2, the Gestural Harmony Model does adopt the multi-step process of
176
speech production assumed within Articulatory Phonology, by which the Coupled Oscillator
Model calculates a gestural score from a coupling graph and the Task Dynamic Model calculates
articulatory trajectories from a gestural score. However, these later levels of representation are
calculated based on crucially language-universal principles, rather than being chosen by a
language-specific grammatical component. The only grammar assumed within the Gestural
Harmony Model is the Optimality Theoretic phonological grammar that produces the coupling
graph. The Coupled Oscillator Model and the Task Dynamic Model are purely extragrammatical
mechanisms in the sense that their outputs are calculated from the content of a coupling graph
and involve no manipulation of that coupling graph. No additional language-specific,
grammatical transformations may occur between coupling graph and gestural score, nor between
gestural score and articulatory trajectories. The Gestural Harmony Model therefore avoids the
complications that arise when utilizing featural pre- and underspecification to account for
patterns of idiosyncratic triggering of harmony. As a result, the Gestural Harmony Model proves
itself to be a particularly parsimonious option for accounting for cases of idiosyncratic harmony
triggering.
3.7 Summary
This chapter has served as the first introduction to the grammatical component of the
Gestural Harmony Model. The grammar of triggering in this model is one based on the shaping
of surface phonological inventories such that they contain harmony-triggering gestures, as well
as placing distributional restrictions on members of those inventories. This is achieved via the
interaction of markedness and faithfulness constraints, both general and position-specific. In
reconceptualizing the driving of harmony in this way, the Gestural Harmony Model makes
177
significant inroads in addressing some of the issues that arise in the analysis of harmony
triggering.
The Gestural Harmony Model is able to account for harmony systems in which harmony
triggers are restricted to specific positions, as well as those that place conditions on the identities
of triggers of harmony. In particular, the Gestural Harmony Model successfully accounts for
patterns of contrastive triggering in harmony, including those in which the contrast between
triggers and non-triggers is restricted to privileged positions. At the same time, it avoids many of
the issues of over- and undergeneration that arise when accounting for patterns of harmony
triggering within feature-based frameworks.
Appendix A: Constraint Definitions
This appendix contains definitions for all of the constraints used in the analyses presented
in chapter 3.
PERSIST(lip protrusion): Assign a violation mark to a lip protrusion gesture that is self-
deactivating.
PERSIST(velum opening): Assign a violation mark to a velum opening gesture that is self-
deactivating.
PERSIST(ATR): Assign a violation mark to a tongue root advancement gesture that is self-
deactivating.
SELFDEACTIVATE: Assign a violation mark to a gesture that is not self-deactivating.
IDENT(deactivation)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical deactivation specifications.
IDENT(deactivation)-IO
Finalσ
: Assign a violation mark to a gesture in the final syllable of a word
whose input and output correspondents do not have identical deactivation specifications.
IDENT(deactivation)-IO
Firstσ
: Assign a violation mark to a gesture in the first syllable of a word
whose input and output correspondents do not have identical deactivation specifications.
178
ANTICIPATE(velum opening): Assign a violation mark to a velum opening gesture that is self-
activating.
ANTICIPATE(ATR): Assign a violation mark to a tongue root advancement gesture that is self-
activating.
SELFACTIVATE: Assign a violation mark to a gesture that is not self-activating.
IDENT(activation)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical deactivation specifications.
*COUPLE(short nonhigh V, self-deactivating lip protrusion): Assign a violation mark to a short
nonhigh vocalic gesture and a self-deactivating lip protrusion gesture that are coupled to one
another in an output.
*COUPLE(vowel, self-deactivating velum opening): Assign a violation mark to a vocalic gesture
that is coupled to a self-deactivating velum opening gesture in the output.
*COUPLE(nonhigh vowel, self-deactivating ATR): Assign a violation mark to a nonhigh vowel
gesture and a self-deactivating tongue root advancement gesture that are coupled to one another.
LICENSE(lip protrusion, first σ): Assign a violation mark to a lip protrusion gesture that is not in
an initial syllable.
LICENSE(nonhigh round V, first σ): Assign a violation mark to a nonhigh round vowel that is not
in an initial syllable.
LICENSE(ATR, first σ): Assign a violation mark to a tongue root advancement gesture that is not
in an initial syllable.
MAX(lip protrusion)-IO: Assign a violation mark to a segment (set of gestures) that includes a lip
protrusion gesture in the input if its output correspondent does not include that gesture.
MAX(ATR)-IO: Assign a violation mark to a segment (set of gestures) that includes a tongue root
advancement gesture in the input if its output correspondent does not include that gesture.
MAX(ATR)-IO
Firsts
: Assign a violation mark to a segment (set of gestures) in an initial syllable
that includes a tongue root advancement gesture in the input if its output correspondent does not
include that gesture.
179
INTEGRITY-IO: Assign a violation mark to a primary gesture and a non-primary gesture that are
part of the same segment (set of gestures) in the input and are not coupled to one another in the
output.
180
Chapter 4
Patterns of Transparency & Blocking
4.1 Introduction
In many of the harmony systems examined in chapters 2 and 3, harmony proceeds
throughout entire words, causing all segments to take on a harmonizing property. In the Gestural
Harmony Model, this is represented by a gesture that extends to overlap all other segments (sets
of gestures) in a word. In this chapter, I focus on cases of so-called neutrality to harmony, in
which one or more segments in a word surface without a harmonizing property. Neutrality is
often used as a catchall term to refer to transparency, in which segments are seemingly skipped
by a harmony process, and blocking, in which a segment arrests the further spread of a
harmonizing property. However, in this chapter I claim that appealing to this unitary concept of
neutrality in analyzing attested harmony patterns is neither useful nor accurate, and that
transparency and blocking should be treated separately.
One of the principle contributions of the Gestural Harmony Model arises from its
representation of transparency and blocking as the results of two distinct theoretical mechanisms.
In doing away with the concept of neutrality, the model successfully accounts for the distinct
typological patterns that are exhibited by cases of transparency and blocking. C. Smith (2016a)
observes that among rounding and nasal harmony systems, the set of attested transparent
segments is a proper subset of attested blocking segments. This asymmetry is not predicted by
any approach in which segment neutrality to harmony is the result of a single mechanism within
the phonological grammar. However, the Gestural Harmony Model successfully accounts for this
typological asymmetry by regarding transparency and blocking as the results of distinct
181
mechanisms at work within the model. Crucially, the mechanism responsible for transparency is
available to a limited set of segments, as determined by their gestural makeup. The blocking
mechanism, on the other hand, is available to all segment types.
The mechanisms within the Gestural Harmony Model that account for transparency and
blocking are based upon two different possible consequences of gestural overlap. When a
persistent (non-self-deactivating) or an anticipatory (early-activating) gesture extends to overlap
other segments in a word, it will at times result in the concurrent activation of gestures that are
antagonistic to or incompatible with one another. Gestural incompatibility refers to any situation
in which the concurrent activation of two gestures is articulatorily or perceptually marked in
some way. The phonological grammar assumed by the Gestural Harmony Model determines
whether or not the incompatibility that would result from such overlap is permitted to surface.
When the grammar does not allow such incompatibility, the result is the blocking of harmony.
Because there are many different sources of gestural incompatibility, many different classes of
segments are predicted to be able to act as blockers within a given type of harmony.
Gestural antagonism, on the other hand, refers to a specific type of incompatibility in
which two concurrently active gestures are specified for directly opposing target articulatory
states. I propose that transparency to harmony arises when a harmonizing gesture overlaps
another gesture that is antagonistic to it. When this overlap is permitted by the grammar, the
target states of these gestures will compete with one another for realization, and the transparent
gesture will temporarily mask the effect of a harmonizing gesture. In the Gestural Harmony
Model, then, transparent segments are cast as a special type of undergoer of harmony. This
mirrors earlier proposals by Clements (1976b), Piggott (1988), Cole & Kisseberth (1994, 1995),
182
Walker (1998/2000, 2003), Jurgec (2011), and others that at some level of representation,
transparent segments undergo harmony.
By representing transparency as the result of a harmonizing gesture’s overlap of an
antagonistic gesture, the Gestural Harmony Model makes the claim that only those segments that
include an antagonistic gesture are predicted to potentially be transparent to harmony. Because of
this, the model is able to successfully account for crosslinguistic patterns of transparency and
blocking in which attested classes of transparent segments make up a proper subset of attested
classes of blocking segments, as in nasal harmony and in rounding harmony. Analyses of
harmony that analyze transparency and blocking together under the banner of neutrality are
unable to restrict transparency to a subset of segment classes and are therefore unable to generate
these attested typological asymmetries.
By dividing transparency and blocking among two distinct theoretical mechanisms, the
Gestural Harmony Model also allows for these mechanisms to operate independently, and in
some cases concurrently. As a result, the Gestural Harmony Model successfully accounts for
harmony systems in which both transparency and blocking arise. This is illustrated in this
chapter with a case from Halh Mongolian rounding harmony (Svantesson 1985; Svantesson,
Tsendina, Karlsson, & Franzén 2005), in which high front vowels are transparent to harmony,
while high back vowels serve as blockers. Other examples of harmony systems that exhibit both
transparency and blocking include Coatzospan Mixtec nasal harmony (Gerfen 1999, 2001) and
Menominee tongue root harmony (Cole & Trigo 1988; Archangeli & Pulleyblank 1994;
Archangeli & Suzuki 1995; Walker 2009, 2018). The Gestural Harmony Model successfully
accounts for such patterns by allowing the two mechanisms responsible for transparency and
blocking to operate concurrently within the same language. While some feature-based analyses
183
of harmony are able to match the success of the Gestural Harmony Model in this regard, many
others are unable to generate such patterns.
The chapter is organized as follows. Section 4.2 provides an overview of the typological
patterns of transparency and blocking that arise in several different types of harmony. Due to the
typological asymmetries in attested transparent and blocking segments described in this section,
section 4.3 goes on to demonstrate how both transparency and blocking can be represented as the
results of two distinct mechanisms in the Gestural Harmony Model. Sections 4.4, 4.5, and 4.6
show how these mechanisms function in nasal harmony, rounding harmony, and ATR harmony,
respectively, and how the phonological grammar generates attested patterns of transparency and
blocking. Section 4.7 outlines the strengths of the predictions made by the Gestural Harmony
Model with respect to transparency and blocking, and compares these predictions to those of
previous feature-based analyses of harmony. An appendix with the definitions of all of the
constraints used throughout is included at the end of the chapter.
4.2 Typological Patterns of Transparency and Blocking
This section examines the typological patterns of transparency and blocking in various
vowel and vowel-consonant harmonies. For some harmony phenomena, such as those based on
tongue root position, the set of crosslinguistically attested classes of transparent and blocking
segments are identical. However, for other harmony types, such as those based on nasality and
rounding, the set of attested transparent segment types is a proper subset of attested blocking
segment types. This section lays out these typological patterns in detail, providing a clear picture
of the patterns of transparency and blocking that must be matched by the predictions of a
successful model of harmony.
184
4.2.1 Nasal Harmony
Nasal (vowel-consonant) harmony exhibits a well-known crosslinguistic asymmetry
between the sets of attested transparent and blocking segments. Looking across all reported nasal
harmony systems, all consonants are attested blockers. While any consonant may act as a blocker
of nasal harmony, it is not the case that within a given language any arbitrary set of consonants
may be selected as blockers. Rather, the set of blockers of nasal harmony in a language forms an
implicational hierarchy roughly approximating the sonority hierarchy. This pattern is reported by
Schourup (1972), who notes that blocking of nasal harmony by liquids implies blocking by
obstruents in the same language, and that blocking by glides implies blocking by liquids and
obstruents. Similar versions of this implicational pattern of nasal harmony blocking are also
noted by Pulleyblank (1989), Piggott (1992), Cohn (1993a, 1993b), Hume & Odden (1996), and
Walker (1998/2000). Cohn and Hume & Odden consider the likelihood of a segment to block or
to undergo nasal harmony to be roughly a function of its sonority.
23
Pulleyblank and Walker, on
the other hand, analyze the blocking of nasal harmony as a function of certain segments types’
compatibility with nasalization. The implicational hierarchy of blocking effects leads Walker
(1998/2000) to posit the harmony scale in (127) based on segmental incompatibility with
nasalization.
(127) Harmony scale of nasal (in)compatibility proposed by Walker (1998/2000)
nasal
sonorant stop
!
nasal
vowel
!
nasal
glide
!
nasal
liquid
!
nasal
fricative
!
nasal
obstruent stop
According to Walker, languages may mark the cutoff between undergoers and blockers
of nasal harmony at different points along the scale in (127). Harmony may affect only vowels
23
Hume & Odden construct their scale based on segmental ‘impedance,’ which is defined as roughly the converse of
sonority.
185
and glottal consonants, as in Sundanese (Robins 1957); vowels, glottals, and glides, as in
Capanahua (Loos 1967/1969; Safir 1982); vowels, glottals, glides, and liquids, as in Kayan
(Blust 1972); or all segment types, as in Tuyuca (Barnes & Takagi de Silzer 1976; Barnes 1996).
While many different classes of segments may serve as blockers of nasal harmony, it is
only obstruents that may be transparent to it, as observed by Piggott (1992) and Walker
(1998/2000, 2003). Both also observe that obstruent transparency in nasal harmony only occurs
in languages in which no sonorant consonants block harmony. This prompts Walker to propose
that transparent segments pattern as undergoers with respect to the harmony scale in (127). In the
nasal harmony of Moba Yoruba (Niger-Congo; Nigeria), for instance, vowels, glottals, glides,
and liquids all surface as nasalized, while obstruents are transparent to harmony (Ajíbóyè 2001;
Ajíbóyè & Pulleyblank 2008).
It is not uncommon for voiced and voiceless obstruents to pattern distinctly from one
another within a nasal harmony system. For instance, it is possible for voiced obstruents to
undergo nasal harmony while voiceless obstruents are transparent to it, or for voiced obstruents
to undergo nasal harmony while voiceless obstruents block to it. The first of these patterns is
exemplified by Tuyuca (Tucanoan; Colombia, Brazil), in which voiceless stops and fricatives are
transparent to nasal harmony, while voiced oral stops are in complementary distribution with
nasal stops, suggesting that they are targeted by nasal harmony (Barnes & Takagi de Silzer 1976;
Barnes 1996). A similar pattern of complementary distribution between voiced oral stops and
nasal stops is observed in Orejón (Tucanoan; Peru).
24
Pulleyblank (1989), citing an unpublished
manuscript by Arnaiz, claims that in Orejón nasal harmony, voiceless obstruents block the
spread of nasality, while voiced obstruents show the same kind of complementary distribution
24
Velie Gable (1975) and Pulleyblank (1989) report slightly different nasal harmony patterns in their descriptions of
Orejón presumably due to dialectal differences. I adopt Pulleyblank’s description here.
186
with nasal consonants that is seen in Tuyuca. Again, this suggests that voiced obstruents are
targeted by nasal harmony, while voiceless obstruents are not.
In sum, a successful analysis of crosslinguistic patterns of transparency and blocking in
nasal harmony must account for several typological patterns. They include (1) the implicational
hierarchy of blocking behaviors by glides, liquids, and obstruents; (2) the fact that only
obstruents may be transparent to nasal harmony; and (3) the variable susceptibility of voiced
obstruents to nasal harmony.
4.2.2 Rounding Harmony
An asymmetry between attested transparent and blocking segment types is also observed
across rounding harmony systems. Kaun’s (1995, 2004) surveys of rounding harmony show that
all vowels are attested as blockers in some rounding harmony pattern. According to Kaun, the
driving factors behind the various patterns of blocking are the avoidance of cross-height
harmony, non-high round vowels, and front round vowels. Kaun describes a number of rounding
harmony blocking patterns built around these three considerations. In some languages, only high
vowels undergo rounding harmony, while nonhigh vowels serve as blockers. This is the case in a
number of Turkic languages. It is also possible for rounding harmony to hold only among
nonhigh vowels, while high vowels serve as blockers; this is exemplified by Baiyina Oroqen
(sections 3.4.2 and 4.5.3).
In some rounding harmony systems, whether or not a vowel of a certain quality blocks
rounding harmony is dependent upon the quality of the trigger. In many languages, the sets of
triggers, undergoers, and blockers of rounding harmony are determined by a restriction on cross-
height harmony. Combined with height-specific requirements on triggers of rounding harmony,
this cross-height harmony restriction generates a number of blocking patterns. Perhaps the most
187
straightforward example comes from Yokuts: harmony triggered by nonhigh vowels is blocked
by high vowels, and harmony triggered by high vowels is blocked by nonhigh vowels. In other
languages, bans on both cross-height harmony and nonhigh round vowels result in rounding
harmony that holds only among high vowels, as in Kachin Khakhas, Hixkaryana, and Tsou.
Kaun (1995, 2004) also reports that the markedness of front round vowels affects patterns of
triggering and blocking, resulting in rounding harmony systems in which only back vowels may
undergo harmony, while front vowels act as blockers.
While there are many attested patterns of blocking among rounding harmony systems,
transparency in rounding harmony is quite restricted. The only attested transparent segments are
high front vowels. This is observed in the rounding harmony systems of various Mongolic
languages, such as Halh Mongolian and Šuluun Höh (also known as Chakhar Mongolian), in
which [i] and [ɪ] are transparent (Svantesson 1985; Svantesson et al. 2005).
Two apparent exceptions to this generalization come from Buriat (Mongolic; Russia,
Mongolia, China) and Mari (also known as Cheremis; Uralic; Mari Republic, Russia). However,
in both of these cases the transparent vowels in question are reported to be reduced vowels,
rendering their quality indeterminate. Sebeok & Ingemann (1961) and Odden (1991) report that
Mari displays rounding and backness harmonies, and that the reduced vowel [ə] is transparent to
both of them. Likewise, Svantesson, Tsendina, Karlsson, & Franzén (2005) claim that the vowel
transcribed as short [e] in Buriat is actually pronounced as a short, reduced vowel akin to schwa,
and that this vowel is transparent to rounding harmony. However, it is possible that the
apparently transparent vowels in Mari and Buriat are so reduced that any round vowel quality
they would take on as a result of harmony would not be perceivable. It is plausible that neither of
these languages exhibit true transparency, but are instead cases of apparent or perceptual
188
transparency. Such cases are not unheard of; there are other languages in which vowels that were
reported to be transparent to some harmony process were found through instrumental study to
actually be undergoers of harmony. See, for example, work by Gick, Pulleyblank, Campbell, &
Mutaka (2006) on the so-called transparent low vowel in Kinande ATR harmony, Benus &
Gafos (2007) on the transparency of the front vowels /i/, /iː/, and /eː/ in Hungarian backness
harmony, and Ritchart & Rose (2017) on the status of schwa in Moro height harmony.
A final possible counterexample to the generalization that only high front vowels are
transparent to rounding harmony comes from Tutrugbu (also known as Nyangbo; Niger-Congo,
Kwa; Ghana).
25
According to Essegbey & McCollum (2017), Tutrugbu has both rounding and
tongue root harmony processes, though rounding harmony is considerably more limited in its
application. Rounding spreads from word-initial second person pronominal prefixes [o-]~[ɔ-]
(2.sg.) and [no-]~[nɔ-] (2.pl.) onto following prefixes containing nonhigh vowels, but never
affects roots. When a word includes either of the non-initial prefixes [-gi-]~[-gɛ-] (neg. pst.) or
[-ti-]~[-tɛ-] (neg.), they do not undergo harmony but do not prevent following prefixes from
undergoing rounding harmony triggered by word-initial second person pronominal prefixes. The
transparency of [i]-containing prefixes does not affect the generalization that high front vowels
are the only vowels attested as transparent in rounding harmony. However, these prefixes are
also transparent to rounding harmony when they surface as their RTR variants [-gɛ-] and [-tɛ-] as
the result of tongue root harmony with a following root.
While the transparency of [ɛ] may appear to represent a counterexample to the claim that
only high front vowels are transparent to rounding harmony, this transparency is unsurprising
given Essegbey & McCollum’s analysis of the Tutrugbu vowel inventory. They claim that the
25
Thanks to Adam McCollum for bringing this language to my attention, and for his detailed descriptions of its
harmony systems.
189
RTR vowels [ɛ] and [ɔ] are phonologically high vowels, having been derived from a recent
historical process of lowering from *ɪ and *ʊ, respectively. From this insight, there are two
possible explanations for the apparent transparency of [ɛ] in Tutrugbu rounding harmony. The
first possibility is that [ɛ] is transparent to rounding harmony because it is actually a
phonologically high vowel, derived from /i/ via the language’s tongue root harmony process. As
an allophone of /i/, high [ɛ] could be expected to be transparent to rounding harmony just as [i]
would be. A second possibility lies in the historical development of [ɛ] from *ɪ. As a reflex of *ɪ,
[ɛ] has retained its phonological distribution with respect to rounding harmony, at least for the
time being. In any case, the status of [ɛ] as transparent to Tutrugbu rounding harmony does not
provide a strong counterargument to the generalization that high front vowels are the only ones
that are transparent to rounding harmony.
4.2.3 Tongue Root and Postvelar Harmonies
Tongue root harmony does not show the same sort of typological asymmetry in attested
transparent and blocking segment classes that is seen in nasal harmony and rounding harmony.
Instead, the sets of attested blockers and transparent segments overlap entirely. In ATR harmony,
it is not uncommon for the low vowel /a/ to lack an ATR counterpart. When this is the case, it
may be either a blocker, as in Akan (Clements 1981), or transparent, as in Budu (Kutsch Lojenga
1994). There are no cases in which high vowels fail to undergo ATR harmony, and cases in
which mid vowels lack ATR counterparts are very rare. Even in such cases, mid vowels re-pair
with other vowels in the inventory, surfacing as high ATR vowels instead of mid ATR vowels.
This is the case in Fur (Kutsch Lojenga & Waag 2004).
In Casali’s (2008) survey of tongue root harmony systems, transparency of the low RTR
vowel /a/ to ATR harmony is attested in several languages, including Bila, Budu, Ngiti, and
190
Talinga-Bwisi. However, the case of apparent transparency of /a/ in Kinande may cast some
doubt on these cases. Kinande was long reported to have a tongue root harmony system in which
the low vowel /a/ is transparent to the spread of ATR. However, an ultrasound study conducted
by Gick et al. (2006) found that /a/ is actually produced with an advanced tongue root in ATR
words, surfacing as [ə]. In other words, this vowel is an undergoer of ATR harmony, despite the
lack of acoustic cues. This calls the transparent status of /a/ in other languages with ATR
harmony into question; the matter is worthy of further study.
In other tongue root harmony systems in which active retraction of the tongue root is the
harmonizing property, it is the high vowels that are transparent or block. In many languages, /i/
and /u/ block the spread of RTR harmony. A well known case of such a system comes from the
standard variety of Yoruba (Pulleyblank 1988; Archangeli & Pulleyblank 1989, 1994). It is also
possible for /i/ and /u/ to be transparent to RTR harmony. Perhaps the most well-known
examples of high vowel transparency to RTR harmony come from Ife Yoruba (Ọla Orie 2001,
2003) and Wolof (Pulleyblank 1996). However, given the new understanding of vowel
articulation in Kinande, there are also cases in which high vowels are reported as being
transparent to RTR harmony when instrumental study shows that are not. For instance, while /i/
has been described as transparent to RTR harmony in Halh Mongolian, acoustic data collected
by Svantesson et al. (2005) show that /i/ is actually produced as [ɪ] in RTR forms.
Interestingly, a similar pattern of transparency and blocking effects appears to hold across
RTR vowel harmonies and vowel-consonant postvelar harmonies, which also involve retraction
of the tongue body and/or root. In Flathead Montana Salish, /i/ is transparent to a process of
faucal (uvularization or pharyngealization) harmony (Doak 1992; Bessell 1998), which is
triggered by uvular and pharyngeal consonants. The related Coeur d’Alene Salish displays a
191
similar process in which /i/ in some words appears to partially undergo harmony and partially
remain transparent to it (Doak 1992; Bessell 1998); this is discussed in greater detail in section
5.3.
In addition, many varieties of Arabic have a process of emphasis
(uvularization/pharyngealization) harmony that interacts in interesting ways with palatal
consonants and high vowels. In Arabic, coronal consonants may be either plain or emphatic, i.e.
produced with concurrent tongue body retraction. Emphasis harmony spreads this tongue body
retraction from a triggering coronal consonant onto surrounding consonants and vowels (for an
overview, see Watson (2002)). In many varieties of Arabic, palatal consonants and the vowel /i/
block emphasis harmony; this is the case in Palestinian Arabic (Younes 1982, 1993; Herzallah
1990; Davis 1995; Shahin 1997). In other varieties, such as Cairene Arabic, the set of blockers is
expanded to include the high back vowel /u/ (Watson 2002).
4.2.4 Vowel Place Harmonies
The other two properties upon which harmony is commonly based are vowel height and
backness. The most well-known cases of transparency and blocking in backness harmony come
from the Uralic languages, especially Hungarian and Finnish, in which front unround vowels fail
to participate in the progressive spread of backness. Often, these languages show a distinction in
the behavior of front vowels with respect to rounding, with round front vowels frequently
harmonizing while unround front vowels are neutral to harmony. Many analyses focus on
whether these front unround vowels are transparent to harmony, allowing a preceding vowel to
spread its [+back] feature onto a following vowel, or whether they trigger their own harmony
based on the feature [-back]. In Finnish, the front unround vowels /i/ and /e/ lack back
counterparts and are transparent to backness harmony when they occur in suffixes. A similar
192
case comes from Votic backness harmony, in which high /i/ is transparent and /e/ is sometimes
transparent (Blumenfeld & Toivonen 2016). A great deal of research on Hungarian has focused
on the relative likelihood of front unround vowels of different heights and lengths to be
transparent or to block harmony (e.g., Vago 1976; Ringen & Vago 1998; Siptár & Törkenczy
2000; Gafos & Beňuš 2006; Hayes & Londe 2006). Work on this language has established a
continuum of likelihood of a vowel to be transparent to backness harmony: [i] and [iː] are most
likely to be transparent, followed by [eː], followed by [ɛ]. Other sources of the patterns of
backness harmony neutrality in Hungarian include vowel length, as well as statistical patterns
over the lexicon; see Hayes & Londe (2006) for an overview.
Backness harmony is also common among languages with rounding harmony, especially
within the Turkic language family. However, backness harmony in these languages usually
proceeds uninterrupted, with no vowels being either transparent to or blocking harmony.
However, there are reported cases of consonants blocking backness harmony. One example of
this comes from Turkish. In Turkish, all vowels participate in backness harmony; rounding
harmony only holds among high vowels, while nonhigh vowels act as blockers. Interestingly,
while backness harmony is not blocked by any vowels in the Turkish inventory, it is blocked in
some cases by palatalized consonants (Clements & Sezer 1982).
Among height harmonies, it is common for the low vowel /a/ to block processes that
result in either vowel raising or vowel lowering. Parkinson (1996) lists many cases of height
harmony involving stepwise vowel raising. While some of these height harmonies target all
vowels in a language’s inventory, others do not affect the low vowel /a/. One example comes
from Kikuria, in which /a/ blocks raising that is triggered by a high vowel in a suffix from
affecting preceding prefixes. Low /a/ also blocks height harmony in Shona, in which vowel
193
lowering is triggered by mid vowels but not the low vowel /a/; furthermore, /a/ blocks vowel
lowering from spreading further (Beckman 1997, 1998). This is similar to the situation in some
rounding harmony systems, in which non-triggering round vowels also block harmony.
There do not appear to be any instances of transparency in unbounded height harmony. A
classic case of height harmony which has been described as displaying both transparency and
blocking comes from Menominee (Cole & Trigo 1988). However, this harmony system has been
reanalyzed by Archangeli & Pulleyblank (1994) and others as involving tongue root position
rather than height. On the other hand, there are cases of metaphony, a type of bounded harmony,
in which /a/ is transparent to harmony involving raising. In Lena Asturian, high vowels in
inflectional suffixes cause raising of a stressed vowel in the root. When an unstressed [a]
intervenes, it is not affected by harmony (Hualde 1989). However, Walker (2011) suggests that
there is reason to believe that this and other metaphony patterns should not be analyzed as
feature spreading, but as long-distance agreement of some kind. In Lena Asturian, /a/ is a target
of harmony when it is in a stressed syllable, raising to [e] when it precedes a triggering
inflectional suffix with a high vowel. However, /a/ is transparent to harmony, surfacing faithfully
as [a], when it is in a syllable between the stressed syllable and the inflectional suffix.
Harmony systems based on vowel height and backness will be discussed in section 6.2.1.
Currently, there are no direct analogs for vowel height and backness within any theories of
gesture-based phonology. This means that while it is possible within gestural phonology to
represent front and back vowels of all heights, there is at present no way to represent the spread
of height and backness. The representation of such harmonies within the Gestural Harmony
Model requires a substantial modification to the representation of vowel place within gestural
phonology.
194
4.2.5 Summary
For all vowel and vowel-consonant harmonies, with the possible exception of unbounded
height harmony, both transparency and blocking are attested. Of particular interest is the fact that
blocking and transparency show different patterns of attestation in nasal and rounding
harmonies, with the ability to surface as transparent limited to significantly smaller classes of
segments. In nasal harmony, all segment types are attested as blockers, while only obstruents are
attested as transparent. In rounding harmony, all vowels are attested as blockers, while only high
front vowels are attested as transparent. In other types of harmony, including those based on
tongue root position, no much asymmetry exists; the sets of attested blocking and transparent
segments are identical. A successful model of harmony must be able to limit the ability to be
transparent to smaller segment classes in nasal and rounding harmonies, but not in other types of
harmony.
In addition, it is possible for transparent and blocking segments to exist within a single
harmony system. This occurs in some rounding harmony systems (e.g., Halh Mongolian), nasal
harmony systems (e.g., Coatzospan Mixtec), and in tongue root harmony systems (e.g.,
Menominee). A successful model of harmony must be able to generate such systems.
The typological patterns of transparency and blocking in nasal harmony and rounding
harmony are of interest because they are not predicted by standard approached to neutrality in
harmony, in which neutral segments are the result of a single mechanism within the phonological
grammar. The Gestural Harmony Model accounts for both the subset asymmetry among
transparent and blocking segments and the ability of a harmony system to exhibit both
transparency and blocking by analyzing them as the results of distinct theoretical mechanisms.
The following section introduces these two mechanisms in turn.
195
4.3 Gestural Antagonism and Gestural Inhibition
Rather than treating transparency and blocking as two different manifestations of the
singular concept of neutrality, the Gestural Harmony Model provides two distinct mechanisms
with which to drive and represent transparency and blocking. Crucially, the mechanism
responsible for transparency in the Gestural Harmony Model is available to a limited set of
segments, accounting for the fact that transparent segments make up a proper subset of blocking
segments in nasal harmony and rounding harmony. The blocking mechanism, on the other hand,
is available to all segment types. These two mechanisms are based on the notions of gestural
incompatibility and gestural antagonism.
In the Gestural Harmony Model, harmony is the result of a gesture that potentially
overlaps all segments (sets of gestures) in a word, causing them to undergo harmony. However,
at times this sort of overlap may result in the concurrent activation of gestures that are either
antagonistic to or incompatible with one another. These two possible consequences of gestural
overlap are the motivators of the two distinct mechanisms that generate transparency and
blocking within the Gestural Harmony Model.
I propose that in the Gestural Harmony Model, transparent segments are not neutral to
harmony, but instead that they are overlapped by a harmonizing gesture. Transparent segments,
then, are just a special type of undergoer. In this model, transparency to harmony is the result of
the concurrent activation of a harmonizing gesture and a gesture that is antagonistic to it. Two
gestures are antagonistic if they have directly opposing target articulatory states that strive to pull
an active articulator in opposite directions. For example, a velum opening and a velum closure
gesture are antagonistic when active at the same time. During this period of concurrent
activation, the velum receives conflicting instructions from the two active gestures; these target
articulatory states cannot both be achieved. Instead, resolution of this competition is calculated
196
according to the Task Dynamic Model of speech production (Saltzman & Munhall 1989).
According to the Task Dynamic Model, concurrently active gestures with non-identical target
articulatory states undergo gestural blending, or averaging, as discussed in section 1.2.1. During
the period of their concurrent activation, the composite target articulatory state of two or more
blended gestures is the weighted average of their individual target states, with each gesture’s
specified strength parameter contributing the weights. As a result of this blending, one or both of
the antagonistic gestures will not fully achieve its target articulatory state. This is discussed in
further detail in section 5.2.1
The Gestural Harmony Model recruits this mechanism of gestural competition and
blending to account for transparency to harmony. When a harmonizing gesture overlaps another
gesture that is antagonistic to it, competition arises and gestural blending occurs. If the gesture
that is overlapped by the harmonizing gesture is sufficiently strong, its target articulatory state
will be favored by gestural blending, and it will counteract the effect of the harmonizing gesture
throughout its period of activation. The result is transparency to harmony.
This is represented schematically in the gestural score in (128). In this gestural score, a
harmonizing gesture extends to overlap a gesture that is antagonistic to it. During the period of
their concurrent activation, these gestures undergo blending. Because the antagonistic gesture is
specified for a high gestural strength (denoted by ‘S’ for strong) while the harmonizing gesture is
specified for a relatively lower gestural strength (denoted by ‘W’ for weak), blending of these
antagonistic gestures is resolved in favor of the target articulatory state of the antagonistic
gesture. As a result, the active articulator is pulled away from the position specified by the still-
active harmonizing gesture during the production of the antagonistic gesture. When the
197
antagonistic gesture deactivates, the articulator is free to return to the position required by the
harmonizing gesture.
(128) Schematic gestural score showing coactivation transparency due to gestural antagonism
In the Gestural Harmony Model, there is no need for the phonological grammar to be
involved in driving transparency to harmony. Instead, transparency follows directly from the
concurrent activation of antagonistic gestures and the calculations of the Task Dynamic Model of
speech production. The analysis of transparency in the Gestural Harmony Model takes advantage
of the dynamical, goal-based nature of gestures. Cast in this way, transparency is a consequence
not of the work of the phonological grammar, but of the gestural representations themselves.
Therefore, the account of transparency based on intergestural competition predicts that only
certain configurations of gestural overlap can result in transparency. Specifically, only those
segments bearing gestures that are antagonistic to a harmonizing gesture can surface as
transparent to harmony. Because of this, the Gestural Harmony Model is able to account for the
limited sets of possible transparent segments in rounding harmony and nasal harmony.
This representation of transparent segments as a special type of undergoer of harmony is
in keeping with previous work by Clements (1976b), Piggott (1988), Walker (1998/2000, 2003),
Jurgec (2011), and others claiming that transparent segments are more closely related to
undergoers of harmony than blockers of harmony. Walker (2003) refers to transparent segments
198
and undergoers together as permeable segments. Both Clements and Piggott analyze
transparency to harmony as the result of a derivationally opaque process by which both
undergoers and transparent segments are targeted by a feature-spreading rule. Afterward,
transparent segments undergo a repair rule that returns them to their original feature
specifications. Walker (1998/2000, 2003) analyzes transparency to nasal harmony using
Harmonic Sympathy, an extension of Sympathy Theory (McCarthy 1999) in order to mimic this
effect of derivational opacity within OT. In the Gestural Harmony Model, by contrast, neither a
repair rule nor the additional grammatical architecture of Sympathy Theory are necessary in
order to generate transparency to harmony. Instead, transparency will result automatically from
the overlap and subsequent blending of antagonistic gestures, as calculated by the Task Dynamic
Model.
While gestural antagonism refers to a specific case of overlap in which two gestural have
directly opposing target articulatory states, gestural incompatibility refers to any situation in
which the concurrent activation of two or more gestures is marked in some way (including the
special case of gestural antagonism). The overlapping of two incompatible results in the
production of a segment that is either articulatorily or perceptually difficult. For example, the
concurrent activation of a velum opening gesture with the lingual gestures making up a liquid is
marked for perceptual reasons (Walker 1998/2000). A language’s phonological grammar may
either allow the marked structure caused by this gestural overlap, or ban it and prevent the
overlap of these incompatible gestures.
I analyze blocking of harmony as the result of a language banning the concurrent
activation of a harmonizing gesture and the gestures of a blocking segment with which it is
incompatible. Blockers of harmony are distinct from transparent segments in this way; while
199
transparent segments are undergoers of harmony, blocking segments are not. This appeal to
markedness and restrictions on co-occurrence is akin to most feature-based analyses of blocking
in harmony. Where the Gestural Harmony Model differs from these analyses is in the application
of these restrictions on gestural co-occurrence only to cases of blocking and not to both blocking
and transparency. This is the source of the different typological predictions made by the Gestural
Harmony Model and many feature-based analyses of harmony with respect to which segment
types are attested as transparent and blocking segments. This is discussed in detail in section
4.7.2.
A gestural analysis of blocking must account for the fact that in some systems, a
harmony-triggering gesture may extend to overlap some gestures and not others. Furthermore, it
must do so without directly manipulating a gesture’s period of activation, which is not included
within a coupling graph representation. Instead, some relation between a harmonizing gesture
and a blocking gesture (or set of gestures) must be specified in the coupling graph and must
result in no overlap between these gestures. The generalization that seems to be at the heart of
this blocking behavior is that the activation of a blocking gesture prevents the activation of a
harmonizing gesture. In the case of progressive harmony, a blocking gesture causes deactivation
of a persistent (non-self-deactivating) gesture, while in the case of regressive harmony, a
blocking gesture prevents an anticipatory (early-activating) gesture from activating even earlier.
I propose that this inhibitory relation between incompatible gestures is what must be
represented in the coupling graph. Inhibition refers to a relation between two units in which the
activation level of one unit detracts from the activation level of the other. Within the Gestural
Harmony Model, I assume this inhibition relation to be unidirectional, with the activation of the
inhibiting gesture determining the activation of the inhibited gesture. In a coupling graph, an
200
inhibitory relation between two gestures can be represented as in (129) for a case of progressive
harmony. In this figure, the inhibitory relation is represented by the dotted line between the two
gestures, and the X indicates the gesture which is being inhibited.
(129) Intergestural inhibition relation between incompatible gestures
In the case of progressive harmony, the potential for a persistent gesture to extend its
period of activation is represented by the gesture’s setting of its deactivation parameter such that
it will not self-deactivate upon reaching its target articulatory state. Its precise end time is not
specified, but is rather calculated by the augmented version of the Coupled Oscillator Model
described in section 2.2. The inclusion of an inhibition relation between a persistent gesture and
a following blocking gesture specifies that whenever a blocking gesture is activated, the
harmonizing gesture will deactivate. In (129), the activation periods of the inhibited and
inhibiting gestures are not represented; however, this figure still represents the fact that the
inhibited gesture will deactivate whenever the inhibiting gesture activates. When a coupling
graph that includes an inhibition relation between a harmonizing gesture and a blocking gesture
is input to the Coupled Oscillator Model, the result will be a gestural score in which the
harmonizing gesture extends only partially through a word. This is illustrated in the schematic
gestural score in (130).
(130) Schematic gestural score showing intergestural inhibition due to gestural incompatibility
201
Early work in Articulatory Phonology (e.g., Browman & Goldstein 1986, 1989) assumes
that gestures are either active (with an activation level equal to one) or inactive (with an
activation level equal to zero). In this view, the transition from inactive to active and vice versa
is instantaneous. If this is the case, at the moment when a blocking gesture goes from fully
inactive to fully active, an inhibited persistent gesture will go from fully active to fully inactive.
More recent work in Articulatory Phonology, such as that reported by Byrd & Saltzman (2003),
assumes a ramped representation of gestural activation and deactivation, in which a gesture’s
transition between activity and inactivity is rapid but not instantaneous, as in (131).
(131) Timecourse of ramped activation and deactivation of a gesture
Intergestural inhibition can be implemented in a gestural framework that assumes ramped
activation as well. As a working hypothesis I assume that the calculation of the activation levels
of gestures that are in an inhibitory relationship is fairly simple (though see section 6.2.2 for
further discussion). When two gestures are in an inhibitory relationship and are simultaneously
active, the activation of the inhibited gesture is calculated as the difference between 1 and the
activation level of an inhibitory gesture, as in (132). When an inhibitory gesture is not activated,
it has no effect on the activation of the inhibited gesture and the equation in (132) does not hold.
202
(132) Equation for gestural inhibition
Activation Level of
Inhibited Gesture
= 1 –
Activation Level of
Inhibitory Gesture
That the activation level of the blocking gesture determines the sum total activation of the
two gestures is captured by the asymmetric nature of the inhibitory relation between them. As the
activation of the inhibitory gesture approaches 1, the activation of the inhibited gesture must fall
to zero. I assume that once a gesture’s activation reaches 0, it remains there; gestures cannot be
reactivated at a later time. The figure in (133) shows the time course of activation of two gestures
in an inhibitory relationship.
(133) Persistent gesture (dashed) is inhibited and deactivated by a following blocking gesture
(solid)
This inhibition mechanism can also be used to represent blocking in regressive harmony
systems, in which a blocker precedes a harmony triggering anticipatory gesture. In this case,
rather than deactivating a harmonizing gesture, inhibition by a blocking gesture prevents the
Coupled Oscillator Model from activating an anticipatory gesture even earlier and extending that
gesture’s activation even further in the regressive (leftward) direction. This is illustrated in the
schematic gestural score in (134).
203
(134) Schematic gestural score showing intergestural inhibition due to gestural incompatibility
The presence of an inhibition relation between a harmonizing gesture and an
incompatible blocking gesture can be attributed to the need to satisfy a constraint within the
phonological grammar. Note that this contrasts with the analysis of transparency via the
concurrent activation of antagonistic gestures, in which transparency results directly from the
representational units themselves and not as the result of any grammatical mechanism. I propose
that blocking via intergestural inhibition is enforced by constraints from the *OVERLAP family,
the schematic definitions of which are repeated in (135) from (19) in section 1.2.2.
(135) Schemas for *OVERLAP constraints
a. *OVERLAP(Gest
X
, Gest
Y
): Assign a violation mark for a pair of gestures of type X and
type Y that are concurrently active.
b. *OVERLAP(Gest
X
, Gest
Y
, Gest
Z
): Assign a violation mark for a gesture of type X that
is concurrently active with a gesture of type Y and with a gesture of type Z.
The marked temporal overlap between a harmonizing gesture and a gesture or set of
gestures with which it is incompatible is prevented when these gestures enter into an inhibition
relation with one another, such that the blocking gesture deactivates the harmonizing gesture.
The overlap of incompatible gestures is not prevented by simply stating the starting point or end
point of the harmonizing gesture within the gestural score, but by altering the coupling graph (via
the inclusion of an inhibition relation) such that the desired gestural score is produced from it by
the Coupled Oscillator Model.
This alteration of the coupling graph involves not only the inclusion of an inhibition
relation, but also the proper setting of the directionality of the asymmetric inhibition relation.
204
*OVERLAP constraints do not specify the direction of intergestural inhibition. Instead, the desired
directionality of inhibition, by which a blocking gesture inhibits a harmonizing gesture (and not
vice versa), will arise automatically for both persistent and anticipatory gestures. Looking first at
progressive (rightward) harmony, consider the two possible directions of asymmetric inhibition
between a persistent (non-self-deactivating) gesture and a following blocking gesture in (136).
(136) Inhibition relations between harmonizing persistent gesture and blocking gesture
a.
b.
The relation shown in (136a), in which the intended blocking gesture inhibits the
persistent harmonizing gesture, produces the desired time course of gestural activation shown in
(133) above. With this inhibition relation in place, the harmonizing gesture activates itself, and
deactivates when the inhibitory blocking gesture activates itself. The relation shown in (136b),
on the other hand, produces the undesired time course of gestural activation in (137).
(137) Persistent gesture (dashed) inhibits following blocking gesture (solid), preventing its
activation (not produced by GEN)
Again, the persistent harmonizing gesture activates itself, but now it is this harmonizing
gesture that inhibits the intended blocking gesture. Because the summed activations of two
205
gestures in an inhibition relation must equal one, and the fully active velum opening gesture
already has an activation of one, the blocking gesture will be prevented from activating at the 0º
phase of its planning oscillator as specified in the coupling graph. Meanwhile, the harmonizing
gesture will remain active. This directionality of inhibition has produced two undesirable
outcomes: the persistent gesture has not been deactivated by its intended blocker, and the
blocking gesture has failed to activate at all.
The inhibition relation in (136b) produces a scenario in which a gesture that is present in
the coupling graph fails to ever activate. The Gestural Harmony Model must ban such a
configuration from ever being generated by the phonological grammar. This can be achieved by
stating that it is a property of GEN that it will not produce candidate output forms that do not
allow for the activation all gestures present in the coupling graph. By ruling out the inhibition
relation in (136b) in which the harmonizing persistent gesture inhibits its intended blocking
gesture, only one possibility remains. When an inhibition relation is added to a coupling graph
between a harmonizing gesture and a blocking gesture, it will always be the case that the
blocking gesture inhibits the harmonizing gesture. There is no need to stipulate this directionality
in the definition of a constraint like *OVERLAP.
The same reasoning holds true of regressive (leftward) harmony, in which a blocking
gesture precedes a harmonizing anticipatory (early-activating) gesture. The desired directionality
of inhibition, by which a blocking gesture inhibits a following harmonizing anticipatory gesture
as in (136a), will arise automatically in candidate coupling graphs. If a harmonizing anticipatory
gesture were to instead inhibit a preceding blocking gesture, it would be permitted to activate as
early as possible in a word and would extend to overlap the blocking gesture, thus preventing it
from ever activating. The time course of activation for such a configuration would be identical to
206
that in (137). This configuration is also assumed to be universally banned from being generated
by the phonological grammar.
In order to account for patterns of blocking across harmony systems, the phonological
grammar must include some constraint that conflicts with the family of *OVERLAP constraints.
Otherwise, harmony will be predicted never to proceed past a segment that is subject to some
*OVERLAP constraint, i.e. any segment that is attested as a blocker of harmony in any language.
In many analyses of harmony, this is achieved by ranking a harmony-driving constraint over the
constraint that is responsible for blocking. However, the Gestural Harmony Model does not
make use of any explicit harmony driver, instead relying on a property of the harmonizing
gesture, its persistence and/or early activation, to motivate the temporal extension of a gesture.
Rather than using a harmony driver to counter *OVERLAP, I propose the addition of a constraint
that penalizes inhibition relations generally, as defined in (138).
(138) *INHIBIT: Assign a violation mark to an inhibition relation between gestures in a coupling
graph.
When a *OVERLAP constraint is ranked above *INHIBIT, the gestures that are named as
the arguments of the *OVERLAP constraint are allowed to enter into an inhibition relation in order
to prevent the overlap of incompatible gestures. This results in the blocking of harmony. When
*INHIBIT is ranked above *OVERLAP, the overlap of incompatible gestures will be permitted, and
harmony will not be blocked. This interaction between *OVERLAP and *INHIBIT constraints is
illustrated by the tableau in (139). In candidate (a), two incompatible gestures overlap one
another, violating *OVERLAP(Gest
X
, Gest
Y
) but satisfying *INHIBIT. In candidate (b), Gesture Y
inhibits Gesture X, violating *INHIBIT but satisfying *OVERLAP(Gest
X
, Gest
Y
). The candidate
that surfaces as the winner is determined by the relative ranking of these two constraints.
207
(139) Tableau illustrating violation profiles for *OVERLAP and *INHIBIT
Input:
*OVERLAP
(Gest
X
, Gest
Y
)
*INHIBIT
a.
*
b.
*
In closing, it should be reiterated that while gestural antagonism and gestural
incompatibility are distinct concepts whose effects are manifested in the Gestural Harmony
Model in different ways, gestural antagonism is a form of gestural incompatibility. When two
concurrently active gestures have conflicting target articulatory states, they are both antagonistic
to and incompatible with one another. However, it is not the case that incompatibility between
gestures entails gestural antagonism. Gestures may be incompatible with one another for a
variety of reasons, both articulatory and perceptual, that do not necessarily have to do with
gestural antagonism. Because of this, the ability to induce transparency to a given type of
harmony is restricted to the specific set of segments that include a gesture that is antagonistic to a
harmonizing gesture, while any incompatible gesture is capable of blocking harmony. By
splitting the motivations and theoretical mechanisms responsible for transparency and blocking
within the Gestural Harmony Model, distinct sets of predictions are made with respect to which
types of segments can be transparent to harmony, and which can block it. This is essential to
accurately capturing the fact that in nasal and rounding harmonies the sets of attested transparent
and blocking segments are in a subset-superset relation.
Gest
X
Gest
Y
208
Now that the theoretical mechanisms responsible for transparency and blocking in the
Gestural Harmony Model have been introduced, the following sections examine how they can be
implemented within nasal harmony, rounding harmony, and tongue root harmony. For each type
of harmony, it is necessary to determine what gestural representations are necessary to produce
transparency via intergestural competition and blending, and which constraints must be present
in the phonological grammar to induce intergestural inhibition relations and produce attested
patterns of blocking.
4.4 Transparency and Blocking in Nasal Harmony
This section focuses on transparency and blocking in nasal harmony, in which all
consonant types are attested as blockers, while only obstruents are attested as being transparent
to harmony. This section begins by examining the gestural representations and constraints
necessary to successfully produce the attested patterns of transparency and blocking in nasal
harmony. The proposals made here are then illustrated by analyses of transparency in Tuyuca, as
well as blocking in several languages, including Orejón and Capanahua.
4.4.1 Sources of Antagonism and Incompatibility
As discussed in section 4.2.1, Piggott (1992) and Walker (1998/2000, 2003) claim that
obstruents are the only consonants attested to be transparent to nasal harmony. The particularly
strong ban on the co-occurrence of obstruency and nasality is well established within both
phonetics and phonology. An oral stop is characterized by a buildup of air pressure during its
closure phase, which then leads to an audible burst upon release. Fricatives depend on sufficient
airflow through a narrow channel formed at some point along the vocal tract in order to produce
the turbulence necessary for frication. Both of these acoustic events are dependent upon precise
aerodynamic conditions of the vocal tract, conditions that are significantly disrupted by the
209
opening of the velum. These aerodynamic factors lead Ohala & Ohala (1993) to posit the
theorem in (140) with respect to obstruents and velum position.
(140) Theorem A: The velic valve must be closed (i.e., the soft palate must be elevated) for an
obstruent articulated further forward than the point where the velic valve joins the nasal
cavity and the oral cavity. (p. 227)
Following the lead of Theorem A, some representations of obstruents in Articulatory
Phonology include both an oral constriction gesture (full closure for stops and critical
constriction for fricatives) and a velum closure gesture. I adopt this representation of obstruents.
The inclusion of a velum closure gesture in the representation of an obstruent will ensure that
there is a tight seal of the velopharyngeal port that will prevent the escape of air through the
nasal cavity. While the neutral position of the velum (the position it assumes when not engaged
by an active gesture) is assumed to be high enough to close the velopharyngeal port, the seal is
presumably not sufficiently tight for the production of obstruents. Thus, tighter closure between
the velum and pharyngeal wall must be achieved by an active velum closure gesture that raises
the velum relative to its neutral position. These velum positions are illustrated in the figure in
(141).
(141) Vocal tract with velum in neutral, loosely sealed position (dashed line) and in actively
raised, tightly sealed position (solid line)
210
Articulatory data supports the claim that the gestural representation of an obstruent
should include a velum closure gesture. Numerous studies, including those by Lubker (1968),
Bell-Berti & Hirose (1975), and Bell-Berti (1976) report raising of the velum during the
production of oral stops relative to the height achieved during the production of oral vowels
(which are assumed to represent the lower neutral position of the velum), as well as during the
production of nasal consonants.
The presence of a velum closure gesture in the representation of obstruents is crucial to
the analysis of their transparency to nasal harmony within the Gestural Harmony Model. A
velum closure gesture is directly antagonistic to a harmonizing velum opening gesture, and is the
source of obstruents’ resistance to nasalization. When an obstruent is overlapped by a velum
opening gesture, its antagonistic velum closure gesture will ensure that the obstruent is not
nasalized. In contrast, a loose seal of the velopharyngeal port is sufficient to achieve the less
stringent oral airflow goals of liquids, glides, and vowels. Thus, these sounds are not represented
with an accompanying velum closure gesture. Because of this lack of a velum closure gesture,
the Gestural Harmony Model accurately predicts that these segments are unable to surface as
transparent to nasal harmony, and will always surface as nasalized when they are overlapped by
a harmonizing velum opening gesture.
Complicating the picture slightly is the status of fricatives and of voiced obstruents,
which appear to vary across languages with respect to whether they are classified as obstruents
or as sonorants. In some nasal harmony systems, both fricatives and oral stops are transparent to
nasal harmony, suggesting that they are specified for active velum closure in that language. An
example of such a nasal harmony system comes from Moba Yoruba (Ajíbóyè 2001; Ajíbóyè &
Pulleyblank 2008). In other nasal harmony systems, however, fricatives appear to pattern with
211
sonorants in surfacing as nasalized. An example of such a system comes from Applecross
Scottish Gaelic (Ternes 1973; Warner, Brenner, Schertz, Fisher, & Hammond 2015). There also
appears to be variability across languages as to whether the voicing of a fricative or oral stop
determines its status as a sonorant or an obstruent. In Tuyuca nasal harmony, for instance,
voiceless fricatives and stops are transparent to nasal harmony, while voiced stops surface as
nasalized (Barnes & Takagi de Silzer 1976; Barnes 1996).
26
The Gestural Harmony Model’s account of the variable behavior of voiced and voiceless
obstruents in nasal harmony adopts the proposals by Rice & Avery (1989), Piggott (1992), Rice
(1993), Botma (2004), and Botma & N. Smith (2007) that voiced obstruents that undergo
nasalization are represented in a way that is similar to the representation of sonorants. Rice
(1993) uses the term ‘sonorant obstruents’ to refer to voiced fricatives and stops that surface as
oral in oral environments but are susceptible to nasalization in the same way that sonorants are.
Within gestural phonology, I propose that the class of obstruents be defined as those fricatives
and stops that are accompanied by a velum closure gesture. The class of sonorants, then,
encompasses any segment that does not include a velum closure gesture. Languages appear to
vary as to whether their fricatives and voiced oral stops are accompanied by a velum closure
gesture, rendering them obstruents, or whether they lack a velum closure, rendering them
sonorants. As a result of this lack of a velum closure gesture, both true sonorants and ‘sonorant
obstruents’ (fricatives and oral stops without a velum closure gesture) are predicted to be unable
to surface as transparent to nasal harmony.
While transparency to nasal harmony is limited to the class of obstruents, all classes of
consonants (and in rarer cases vowels) are able to block harmony on a language-specific basis.
The propensity of different consonant types to block harmony follows the implicational
26
Tuyuca has no voiced fricatives.
212
hierarchy observed by Schourup (1972), Pulleyblank (1989), Piggott (1992), Cohn (1993a,
1993b), Hume & Odden (1996), and Walker (1998/2000). In particular, I follow the proposals of
Pulleyblank and Walker that this implicational hierarchy is based on different types of segments’
incompatibility with nasalization. However, I propose that this implicational hierarchy need not
distinguish between fricatives and oral stops in terms of their incompatibility with nasalization.
Instead, whether fricatives pattern with liquids or with obstruents is determined by the presence
or absence of a velum closure gesture in their representations; the same holds true of voiced
fricatives and stops. The revised scale of nasal incompatibility that I assume is provided in (123).
(123) Proposed harmony scale of nasal (in)compatibility
nasal
sonorant stop
!
nasal
vowel
!
nasal
glide
!
nasal
liquid
!
nasal
obstruent
The blocking of nasal harmony can be captured within the Gestural Harmony Model by
the interaction of constraints from the *OVERLAP and *INHIBIT families. In order to account for
the patterning of different types of consonants as either undergoers or blockers of nasal harmony,
it is necessary to first explicitly define consonantal types in gestural terms. I assume the
definitions of different segment types provided in (124), and the use of these terms in the
constraint definitions that follow can be considered shorthand for these definitions.
(124) Gestural definitions of segment types
a. Vowel: a segment whose primary gesture is a low-stiffness vocalic gesture
b. Glide: a segment whose primary gesture is a high-stiffness vocalic gesture with
narrow constriction degree
c. Liquid: a segment that includes a consonantal tongue tip gesture and not a velum
gesture
d. Consonant: a segment whose primary gesture is a consonantal gesture or a high-
stiffness vocalic gesture with narrow constriction degree
213
e. Sonorant: a segment that does not include a velum closure gesture
f. Obstruent: a segment that includes a velum closure gesture
g. Oral consonant: a consonantal segment that does not include a velum opening gesture
Having defined these terms for referring to different types of consonants, the constraints
that make reference to them can now be defined as well. First, a *OVERLAP constraint penalizes
the concurrent activation of any oral consonant and a velum opening gesture. This constraint is
defined in (125).
(125) *OVERLAP(oral C, velum opening): Assign a violation mark to the gesture(s) of an oral
consonant that is/are active concurrently with a velum opening gesture.
This constraint is violated by any oral consonant (as defined in (124g)) that undergoes
nasal harmony. It conflicts with constraints from the *INHIBIT family by motivating forms to
surface with an inhibition relation between the gestures of an oral consonant and a velum
opening gesture.
The implicational hierarchy of nasal harmony blockers can be captured by a set of
stringent constraints (de Lacy 2002) requiring that a constraint that penalizes a marked structure
must also penalize any other structure that is considered more marked by some harmony scale.
The constraint *INHIBIT is defined stringently in order to capture the implicational hierarchy of
nasal harmony blockers. The set of *INHIBIT constraints will capture this hierarchy by most
harshly penalizing the poorest blockers of nasal harmony. In order to do this, the stringency
relation must run from the least to the most incompatible with nasality: glides are the least likely
to block nasal harmony, followed by liquids, followed finally by obstruents. The set of
stringently defined *INHIBIT constraints that reflect this implicational hierarchy are defined in
(126). These *INHIBIT constraints can be ranked relative to the general *OVERLAP(oral C, velum
opening) constraint in order to capture attested patterns of blocking of nasal harmony.
214
(126) Constraints against velum gesture inhibition by consonant type
a. *INHIBIT(glide, velum opening): Assign a violation mark for an inhibition relation
between a glide and a velum opening gesture.
b. *INHIBIT(sonorant C, velum opening): Assign a violation mark for an inhibition
relation between a sonorant consonant and a velum opening gesture.
c. *INHIBIT(oral C, velum opening): Assign a violation mark for an inhibition relation
between an oral consonant and a velum opening gesture.
Many analyses of harmony focus on defining the class of blockers of a certain harmony
process. In the Gestural Harmony Model’s analysis of nasal harmony, the *INHIBIT constraints
achieve this by determining which class or classes of segments may not block harmony in some
language. The family of *INHIBIT constraints can be thought of as favoring the spreading of
nasalization through various classes of segments. *INHIBIT(glide, velum opening) prevents
blocking of nasal harmony by glides; in other words, it drives glides to be permeable to nasal
harmony. *INHIBIT(sonorant C, velum opening) similarly drives glides and liquids to be
permeable to nasal harmony, and *INHIBIT(oral C, velum opening) drives all oral consonants to
be permeable to nasal harmony rather than blocking it (with obstruents surfacing as transparent
due to their gestural makeup). It should be noted that harmony is still the result of the presence of
a persistent or an anticipatory gesture in a coupling graph and resulting gestural score, and not of
a constraint explicitly requiring that a harmonizing gesture remain active for a longer period of
time. The *INHIBIT constraints only serve to ensure that when a persistent or an anticipatory
gesture does extend its period of activation, that it is not prevented from extending as far as
possible by a weak blocker.
Typology calculation using OT-Help 2 (Potts, Pater, Jesney, Bhatt, & Becker 2010)
verifies that the relative ranking of the set of *INHIBIT constraints ranked relative to
*OVERLAP(oral C, velum opening) generates the attested patterns of transparency and blocking
215
in nasal harmony listed in (127). The variable patterning of fricatives and voiced obstruents
within each of these generated blocking patterns is determined by whether they include a velum
closure gesture in their gestural representations.
(127) Predicted nasal harmony blocking patterns
Ranking Result Attested in
*OVERLAP(oral C) >>
*INHIBIT(glide)
*INHIBIT(sonorant C)
*INHIBIT(oral C)
Glides, liquids,
fricatives, and stops
block nasal harmony.
Sundanese (Robins
1957)
*INHIBIT(glide) >> *OVERLAP(oral C) Glides undergo nasal
harmony; liquids,
fricatives, and stops
block.
Capanahua (Loos
1967; Safir 1982)
*INHIBIT(sonorant C) >> *OVERLAP(oral C) Sonorants undergo nasal
harmony; obstruents
block.
All fricatives and stops
pattern as obstruents.
Kayan (Blust 1972)
Voiced fricatives and
stops pattern as
sonorants; voiceless
fricatives and stops
pattern as obstruents.
Orejón (Pulleyblank
1989, citing Arnaiz
1988)
Fricatives pattern as
sonorants; stops pattern
as obstruents.
Applecross Scottish
Gaelic (Ternes 1973)
216
Ranking Result Attested in
*INHIBIT(oral C) >> *OVERLAP(oral C) All segments undergo
nasal harmony;
obstruents surface as
transparent.
All fricatives and stops
pattern as obstruents.
Moba Yoruba
(Ajíbóyè 2001;
Ajíbóyè & Pulleyblank
2008)
Voiced fricatives and
stops pattern as
sonorants; voiceless
fricatives and stops
pattern as obstruents.
Guaraní (Gregores &
Suárez 1967)
Fricatives pattern as
sonorants; stops pattern
as obstruents.
It should be noted that while most of the patterns of transparency and blocking in nasal
harmony generated by the constraint set are attested, there is one gap in the table in (127). As yet
I have been unable to find a clear-cut case of a nasal harmony system in which fricatives surface
as nasalized while stops surface as transparent to harmony. This is likely due to the relative rarity
of distinctions between the patterning of stops and fricatives within nasal harmony systems
generally. Based on the database of nasal harmony systems provided by Walker (1998/2000),
nasal harmonies are much more likely to distinguish between voiced and voiceless obstruents
than they are to distinguish between stops and fricatives.
The remainder of this section presents analyses of several of these nasal harmony
systems, from Tuyuca, Orejón, and Capanahua. First to be examined is Tuyuca, exemplifying a
process of nasal harmony in which all segments undergo nasal harmony and voiceless obstruents
surface as transparent.
217
4.4.2 Tuyuca: Voiceless Obstruent Transparency
The nasal harmony of Tuyuca (Tucanoan; Colombia, Brazil; Barnes & Takagi de Silzer
(1976), Barnes (1996), Walker (1998/2000)) represents a system in which all segments may
undergo harmony, and in which voiceless obstruents surface as transparent. The phonological
inventory in (128) is reported by Barnes & Takagi de Silzer (1976).
(128) Tuyuca phonological inventory
Consonants Vowels
p t k ʔ i ɨ u
b d g e o
s a
ɾ
j w
h
In Tuyuca, nasality is a property of an entire morpheme; morphemes may be either oral
or nasal. In oral morphemes, both voiced and voiceless stops and fricatives may appear, as well
as sonorant consonants, as in (129). All data come from Barnes & Takagi de Silzer (1976) and
Barnes (1996).
(129) a. [japa] ‘point’
b. [ete] ‘parakeet’
c. [juka] ‘falcon’
d. [sobo] ‘foam’
e. [bueda] ‘rainbow’
f. [hooga] ‘banana flower’
In nasal morphemes, nasality is expressed on all segments except voiceless obstruents,
which neither undergo nasal harmony nor block it from spreading further (130a-d). Voiced stops,
on the other hand, do not appear in nasal morphemes, instead surfacing as their nasal stop
counterparts (130e-h). (There are no voiced fricatives in Tuyuca.)
218
(130)
a. [mĩpĩ] ‘badger’ e. [ȷ
̃ ãmĩ] ‘night’
b. [w
̃ ãtĩ] ‘demon’ f. [w
̃ ĩnõ] ‘wind’
c. [ȷ
̃ ũkã] ‘yucca soup’ g. [t2
̃ ŋõ] ‘Yapara rapids’
d. [ȷ
̃ õsõ] ‘bird’ h. [ȷ
̃ õɾ
̃ẽ] ‘small hen’
Across morpheme boundaries, the picture is slightly different. Some Tuyuca suffixes
undergo nasal harmony that is triggered by a nasal root, as in [h̃2
̃ 2
̃ -ɾ
̃ĩ] ‘burn (imperative of
warning)’ (cf. [tuti-ɾi] ‘scold (imperative of warning)’). Other suffixes may idiosyncratically
block nasal harmony, and will surface as oral even when followed by a nasal root, as in
[mãmã-wi] ‘new (vehicle classifier).’ When a suffix begins with an obstruent, either voiced or
voiceless, the suffix will always block nasal harmony from a root, as in [w
̃ ãkã-go] ‘wake up
(evidential).’ I focus in this section on an analysis of within-morpheme nasal harmony in
Tuyuca, setting aside a possible avenue for analysis of across-morpheme harmony until section
6.2.2.
Nasal harmony in Tuyuca can be attributed to the presence of a root-initial persistent
(non-self-deactivating) velum opening gesture that extends to overlap the gestures of following
segments. As proposed in section 3.2, persistent gestures will surface when a constraint of the
type PERSIST(Gest
X
) is ranked high; in Tuyuca, this constraint is PERSIST(velum opening). The
positioning of the persistent velum opening gesture at the beginning of a word in Tuyuca can be
attributed to the high ranking of a constraint from the LICENSE family, paralleling the analysis of
initial-syllable triggering of rounding harmony in Kyrgyz (section 3.2.1). I will not focus on an
analysis of triggering in Tuyuca nasal harmony, and will instead only consider output forms in
which a persistent velum opening gesture is coupled to the first segment in a word.
The absence of within-morpheme blocking of nasal harmony can be captured within the
Gestural Harmony Model by the ranking of *INHIBIT(oral C) over *OVERLAP(oral C). This will
ensure that any oral consonant overlapped by a harmonizing velum opening gesture will be
219
prevented from inhibiting it, despite the resulting incompatibility. For any segments that do not
include a velum closure gesture, this will result in their surfacing as nasalized. This is the case
for the Tuyuca word [ȷ
̃ õɾ
̃ ẽ] ‘small hen,’ in which all consonants and vowels surface as nasalized.
The gestural score and resulting velum aperture time course in (131) illustrates. The dashed line
in the portion of the figure showing resulting velum aperture represents the neutral position of
the velum.
(131) Gestural score for Tuyuca [ȷ
̃ õɾ
̃ ẽ] ‘small hen’
In (131), the persistent velum opening gesture remains active throughout the word,
resulting in the nasalization of all consonants and vowels that it overlaps. Because none of these
sonorant segments includes a velum closure gesture, they surface as nasalized rather than as
transparent when they are overlapped by the velum opening gesture.
A voiceless obstruent, on the other hand, will surface as transparent to nasal harmony
when it is overlapped by a velum opening gesture due to the inclusion in its gestural
representation of an antagonistic velum closure gesture. This voiceless obstruent transparency is
demonstrated in the following coupling graph and resulting gestural score for the Tuyuca word
220
[mĩpĩ] ‘badger,’ in which the voiceless [p] surfaces in a nasal morpheme. The coupling graph for
this form is shown in (132).
(132) Coupling graph for [mĩpĩ] ‘badger’
The transparency of obstruent [p] will obtain directly from the content of this coupling
graph when it is input to the Coupled Oscillator Model and the Task Dynamic Model. Unlike in
many other analyses of transparency in harmony, no additional repair mechanism is necessary to
ensure that the [p] of [mĩpĩ] is produced as oral rather than nasal. The coupling graph is specified
such that when the gestural score and articulatory trajectories for this word are calculated, nasal
harmony with a transparent obstruent will result. This is illustrated in the gestural score in (133),
which is output by the Coupled Oscillator Model for the coupling graph in (132).
221
(133) Gestural score and resulting velum aperture for [mĩpĩ] ‘badger’
In (133), a persistent velum opening gesture overlaps all other gestures in the word. The
initial consonant and the vowels surface as nasalized, but the voiceless obstruent [p], also
overlapped by the velum opening gesture, surfaces as oral. This is due to its antagonistic velum
closure gesture. This velum closure gesture is specified for a high gestural strength, allowing it to
overpower the effect of the velum opening gesture during the period of their concurrent
activation. This overpowering is formalized within the Task Dynamic Model of speech
production as the result of weighted averaging of the target articulatory states of the two velum
gestures such that the target state of the strong velum closure gesture is favored over that of the
velum opening gesture. Once the production of the obstruent has concluded and the velum
closure gesture deactivates, there is no longer an active gesture that is antagonistic to the velum
opening gesture, and the velum opens once again. The validity of this analysis of obstruent
transparency in Tuyuca nasal harmony is tested via computational modeling within the Task
Dynamic Model of speech production in section 5.2.2.
This representation of transparency as the result of intergestural competition in the
Gestural Harmony Model successfully accounts for obstruent transparency in Tuyuca nasal
222
harmony while maintaining a local representation of the harmonizing element. It also
successfully accounts for the ability of voiceless obstruents to surface as transparent to harmony
while other segments that are overlapped by a harmonizing velum opening gesture surface as
nasalized. The distinction lies in the gestural makeup of these different segment classes, and the
analysis of obstruents as consonants that are accompanied by velum closure gestures.
Unlike their voiceless counterparts, voiced obstruents surface as nasalized in the domain
of nasal harmony rather than transparent. I analyze this as being due to the surface
representations of these segments lacking a velum closure gesture that is necessary to generate
transparency. In the form [ȷ
̃ ãmĩ] ‘night,’ for instance, the underlying segment /b/ is proposed to
surface as [m] in a nasal morpheme, having undergone nasalization rather than surfacing as
transparent. The /b/ in this form surfaces without a velum closure gesture, whether it was
accompanied by such a gesture in the input or not. If this form were to surface with a velum
opening gesture accompanying /b/, it would be incorrectly produced as *[ȷ
̃ ãbĩ], with a transparent
voiced obstruent.
A successful analysis of nasal harmony in Tuyuca and other languages in which only
voiceless obstruents are transparent must account for the fact that retention of an obstruent’s
velum closure gesture is enforced only when that obstruent is voiceless. That is, a consonant may
only retain a velum closure gesture when it is also accompanied by a glottal opening gesture.
This can be accomplished by a constraint from the LICENSE family (section 1.2.2) that licenses a
velum closure gesture only when it is accompanied by a glottal opening gesture. The constraint is
defined as in (134).
223
(134) LICENSE(velum closure, glottis open): Assign a violation mark to a velum closure gesture
that is not active concurrently with glottal opening gesture.
This licensing constraint is defined negatively, such that an obstruent that is not
accompanied by a glottal opening gesture (i.e., a voiced obstruent) violates it, but an obstruent
that is accompanied by a glottal opening gesture (i.e., a voiceless obstruent) satisfies it. It can be
viewed as the gestural equivalent of the well-established constraint *VOICEDOBSTRUENT, whose
phonetic grounding lies in the aerodynamic factors necessary for voicing.
In order to satisfy this LICENSE constraint, an underlying voiced obstruent will de-
obstruentize by deleting its velum closure gesture. The deletion of the velum closure gesture of a
voiceless consonant, on the other hand, is not motivated by this LICENSE constraint, as the
presence of the velum closure gesture is licensed by its accompanying glottal opening gesture.
Therefore, when LICENSE outranks MAX(velum closure)-IO, voiced obstruency is penalized,
while voiceless obstruency is not.
Note that the definition of LICENSE does not penalize voiced obstruency only in a
nasalizing context; voiced obstruency is penalized in all contexts. Even in forms without nasal
harmony, the velum closure gesture of an underlying voiced obstruent will delete when
LICENSE(velum closure, glottis open) outranks MAX(velum closure)-IO. In oral environments,
these consonants will still surface as oral, though with only a loose seal between the velum and
pharyngeal wall rather than the tight seal that is characteristic of obstruents that include a velum
closure gesture. When overlapped with a velum opening gesture, these consonants will surface as
nasalized. A possible alternative analysis of the status of voiced obstruents in Tuyuca is
discussed in section 5.2.2.
The result of ranking LICENSE(velum closure, glottis open) over MAX(velum closure)-IO
is illustrated in the tableau in (135) for the Tuyuca form [mĩpĩ] ‘badger,’ in which the
224
hypothetical word-initial consonant /b/ has surfaced as [m] (despite the inclusion of a velum
closure gesture in its underlying representation) while the word-medial voiceless /p/ surfaces as
transparent to nasal harmony.
225
(135) Tableau for Tuyuca [mĩpĩ] ‘badger’
Input: / [nasal]
1
b
1
i
2
p
3
i
4
/
LICENSE(velum closure)
MAX(velum closure)-IO
a.
*!
b.
*
c.
**!
Lip
closure
1
Tongue Body
palatal narrow
2
Lip
closure
3
Tongue Body
palatal narrow
4
Velum
closure
1
Velum
closure
3
Velum
opening
1
Glottis
open
3
226
In (135), candidate (a) [bĩpĩ] retains the underlying velum closure gestures for both the
voiced and voiceless underlying obstruents, satisfying MAX(velum closure)-IO and resulting in
both obstruents surfacing as transparent to nasal harmony. However, the unlicensed velum
closure gesture accompanying voiced [b] fatally violates high-ranked LICENSE(velum closure,
glottis open). Winning candidate (b) [mĩpĩ] deletes the unlicensed velum closure gesture of
voiced /b/ while retaining the velum closure gesture of voiceless /p/, violating MAX(gesture)-IO
once but fully satisfying LICENSE(velum closure, glottis open). Candidate (c) [mĩm ̥ ĩ] goes a step
further by deleting both velum closure gestures, resulting in both consonants surfacing as
nasalized. This candidate incurs a second violation of MAX(velum closure)-IO while not
improving upon its performance with respect to LICENSE(velum closure, glottis open); it is
harmonically bounded.
The ranking of LICENSE(velum closure, glottis open) over MAX(velum closure)-IO, then,
ensures that in Tuyuca voiced obstruents nasalize when overlapped by a harmonizing velum
opening gesture while voiceless obstruents surface as transparent to it. The reverse ranking, in
which MAX(gesture)-IO dominates LICENSE(velum closure, glottis open), results in a pattern in
which velum closure gestures are retained for both voiced and voiceless obstruents, and both are
transparent to nasal harmony. This pattern is less common than the Tuyuca-type pattern, but is
attested in Moba Yoruba (Ajíbóyè 2001; Ajíbóyè & Pulleyblank 2008).
The constraint LICENSE(velum closure, glottis open) causes underlying voiced obstruents
to de-obstruentize, rendering them susceptible to nasalization. This constraint can be used to
straightforwardly account for patterns of voiced and voiceless obstruent transparency in nasal
harmony, as demonstrated by the case of transparency in Tuyuca nasal harmony. The disparate
227
patterning of voiced and voiceless obstruents also arises in the blocking of nasal harmony in
Orejón, to be taken up in the next section.
4.4.3 Orejón: Obstruent Voicing and Blocking
Orejón presents a case similar to Tuyuca, in which voiced and voiceless obstruents differ
in how they pattern with respect to nasal harmony. However, unlike in Tuyuca, the nasal
harmony system in Orejón exhibits blocking, making it an interesting case in which the
implicational hierarchy of blockers of nasal harmony (section 4.4.1) interacts with the
classification of stops and fricatives as either sonorants or obstruents based on their voicing
specifications. Orejón nasal harmony exemplifies a system in which voiced obstruents undergo
nasal harmony while voiceless obstruents block it. This is illustrated by the data in (136),
provided by Pulleyblank (1989).
27
In (136a-c), nasal harmony does not extend beyond a
voiceless obstruent. In (136d-f), all sonorant stops are nasalized, and voiced obstruents are
disallowed in such forms.
(136) a. [nãsoʔ] ‘crab’ d. [
ʔ
mõn2
̃ ] ‘come’ *[
ʔ
bõd2
̃ ]
b. [ȷ
̃ ãkoaʔ] ‘eye’ e. [ȷ
̃ ẽnĩ] ‘flow’ *[ȷ
̃ ẽdĩ]
c. [mĩteʔ] ‘mosquito’ f. [ŋãnã] ‘fly’ *[gãdã]
Following Pulleyblank (1989), I analyze morphemes in Orejón as being specified as
either oral or nasal. As such, nasal morphemes are analyzed as having a morpheme-initial
persistent (non-self-deactivating) velum opening gesture, as in Tuyuca (section 4.4.2). This
persistent velum opening gesture extends to overlap following vowels, glottals, glides, and
liquids. As discussed in section 4.4.1, this pattern of blocking can be generated in the Gestural
Harmony Model by the ranking in (137).
27
Velie Gable (1975) and Cole & Kisseberth (1995) provide a different description of nasal harmony in what is
presumably another dialect of Orejón in which both voiced and voiceless obstruents act as blockers of nasal
harmony. In this section I follow Pulleyblank’s description of Orejón nasal harmony, which is in turn based on an
unpublished manuscript by Arnaiz.
228
(137) Constraint ranking for blocking of Orejón nasal harmony
*INHIBIT(sonorant C, velum opening) >> *OVERLAP(oral C, velum opening) >>
*INHIBIT(oral C, velum opening)
In addition, it appears that voiced stops also undergo nasal harmony; they never occur in
nasal morphemes, while their nasal counterparts do. This necessitates the inclusion of the
constraint ranking LICENSE(velum closure, glottis open) >> MAX(gesture)-IO. As demonstrated
by the case of Tuyuca nasal harmony (section 4.4.2), this ranking will ensure that an underlying
voiced obstruent will delete its unlicensed velum opening gesture, rendering it a sonorant
according to the definition provided in section 4.4.1. As a result, this de-obstruentized consonant
will be susceptible to nasalization when the ranking *INHIBIT(sonorant C) >> *OVERLAP(oral C)
holds.
The undergoing, rather than blocking, of nasal harmony by voiced obstruents in Orejón is
demonstrated in the tableau in (138) for the form [ȷ
̃ ẽnĩ] ‘flow.’
229
(138) Tableau for Orejón [ȷ
̃ ẽnĩ] ‘flow’
Input: / ȷ
̃ 1
e
2
d
3
i
4
/
LICENSE(velum,glottis)
MAX(gesture)-IO
*INHIBIT(sonorant C)
*OVERLAP(oral C)
*INHIBIT(oral C)
a.
*! * *
b.
*! **
c.
* *! * *
d.
* **
Tongue Body
pal nar
1
Tongue Body
palatal mid
2
Tongue Tip
alv clo
3
Tongue Body
pal nar
4
Velum
open
1
Velum
closure
3
230
In (138), candidates (a) [ȷ
̃ ẽdi], with blocking of nasal harmony by [d], and (b) [ȷ
̃ ẽdĩ], with
transparent obstruent [d], each preserve the velum closure gesture of the underlying voiced
obstruent, fatally violating high-ranked LICENSE(velum closure, glottis open). Candidates (c) and
(d) delete this velum closure gesture, effectively becoming be-obstruentized. This violates
MAX(gesture)-IO in favor of satisfying higher-ranked LICENSE(velum closure, glottis open).
Candidates (c) [ȷ
̃ ẽdi] and (d) [ȷ
̃ ẽnĩ] differ only in whether this sonorant obstruent blocks nasal
harmony. In (c), harmony is blocked via an inhibition relation between the tongue tip gesture of
[d] and the velum opening gesture. Because this [d] is de-obstruentized, its inhibition of the
velum opening gesture violates *INHIBIT(sonorant C). Winning candidate (d), however, includes
no such inhibition relation, and the sonorant obstruent undergoes harmony, surfacing as [n].
Blocking by a voiceless obstruent is demonstrated in the tableau in (139) for the Orejón
form [mĩteʔ] ‘mosquito.’
231
(139) Tableau for Orejón [mĩteʔ] ‘mosquito’
Input: / m
1
i
2
t
3
e
4
ʔ
5
/
LICENSE(velum, glottis)
MAX(gesture)-IO
*INHIBIT(sonorant C)
*OVERLAP(oral C)
*INHIBIT(oral C)
a.
*
b.
*!*
c.
*! * *
Lip
closure
1
TB
pal nar
2
TT
alv clo
3
TB
pal mid
4
Glottis
clo
5
Velum
open
1
Velum
clo
3
Glottis
open
3
232
LICENSE(velum, glottis)
MAX(gesture)-IO
*INHIBIT(sonorant C)
*OVERLAP(oral C)
*INHIBIT(oral C)
d.
*! **
In (139), none of the candidates violate LICENSE(velum closure, glottis open), as they all
include a glottal opening gesture accompanying the medial consonant. Therefore, this constraint
will not favor any de-obstruentized candidates. As a result, candidates (c) [mĩteʔ] and (d)
[mĩn ̥ eʔ], which both delete the velum closure gesture of the medial consonant, are ruled out due
to their extraneous violations of MAX(gesture)-IO. Candidates (a) [mĩteʔ] and (b) [mĩtẽʔ̃] both
contain an obstruent [t]; as a result, *INHIBIT(sonorant C) is irrelevant and it falls to
*OVERLAP(oral C) to decide between the candidates. The winning candidate (a) [mĩteʔ], in which
the voiceless obstruent [t] blocks nasal harmony by inhibiting the velum opening gesture,
eliminates a violation of *OVERLAP(oral C) at the expense of violating lower-ranked
*INHIBIT(oral C). Candidate (b) allows [t] and [ʔ] to be overlapped, fatally violating
*OVERLAP(oral C).
The tableau in (138) demonstrates that even when a voiced obstruent includes a velum
closure gesture underlyingly, it is lost when LICENSE is ranked high, resulting in de-
obstruentization. The same result is seen in the analysis of Tuyuca in section 4.4.2. The only
233
difference between the nasal harmony systems of Tuyuca and Orejón lies in the behavior of
voiceless obstruents. In Tuyuca, they are transparent due to the high ranking of *INHIBIT(oral C),
which prevents the blocking of nasal harmony entirely. In Orejón, on the other hand, it is the less
stringent *INHIBIT(sonorant C) that outranks *OVERLAP(oral C), leading to blocking by voiceless
obstruents, as demonstrated by the tableau in (139). In both languages, voiceless obstruents are
the only true obstruents, while voiced obstruents are what Rice & Avery (1989), Piggott (1992),
and Rice (1993) call sonorant obstruents. In the Gestural Harmony Model, this distinction
between obstruent and sonorant is represented by the presence or absence of a velum closure
gesture in the representation of a segment. This determines whether a consonant is able to
surface as transparent to nasal harmony due to antagonism with a harmonizing velum opening
gesture, as exemplified by Tuyuca nasal harmony. It can also determine whether a consonant
undergoes or blocks harmony by determining whether that consonant is under the purview of
*INHIBIT(sonorant C), as exemplified by nasal harmony in Orejón.
4.4.4 Revisiting Capanahua: Blocking by Obstruents and Liquids
The final case of nasal harmony examined here comes from the familiar case of
Capanahua, whose harmony system was previously discussed in sections 2.2.2 and 3.4.3. In
these earlier discussions of nasal harmony in Capanahua, it is mentioned that when a liquid or
obstruent occurs in a word with a harmony-triggering nasal, it blocks the spread of nasality. This
section provides an analysis of this pattern of blocking, and demonstrates how the use of
increasingly less stringent *INHIBIT constraints generates patterns of harmony with fewer types
of segments designated as undergoers. For the sake of simplicity, I focus only on regressive
harmony in Capanahua, though the pattern of blocking is identical for both regressive and
bidirectional nasal harmony.
234
Recall that regressive nasal harmony in Capanahua is triggered by nasal consonants and
affects surrounding vowels, glides, and glottals, as in (140) (repeated from (34) in section 2.2.2).
Nasal harmony is blocked by obstruents and the liquid [ɾ].
(140) a. [h̃ãmawɯ] ‘step on it’
b. [põȷ
̃ ãn] ‘arm’
c. [bãw
̃ĩn] ‘catfish’
d. [cĩʔ̃ĩn] ‘by fire’
e. [cipõnki] ‘downriver’
f. [wɯɾãnwɯ] ‘push it’
g. [wɯɾãnjasãʔ̃nwɯ] ‘push it sometime’
h. [bãnawɯ] ‘plant it’
This blocking by liquids and obstruents can now be accounted for within the Gestural
Harmony Model by assuming an inhibition relation between the blocking gestures of liquids and
obstruents and the anticipatory velum opening gesture responsible for regressive nasal harmony.
In the case of [wɯɾãnwɯ] ‘push it,’ for example, the velum opening gesture of [n] and the
gesture of the blocking tap [ɾ] must be in an inhibitory relationship. The result of this
intergestural inhibition is illustrated in the gestural score in (141).
(141) Gestural score for Capanahua [wɯɾãnwɯ] ‘push it’
The pattern of blocking in Capanahua nasal harmony, in which vowels, glides, and
glottals are undergoers while liquids and obstruents are blockers, can be generated by the ranking
in (142).
235
(142) Constraint ranking for blocking in Capanahua nasal harmony
*INHIBIT(glide) >>*OVERLAP(oral C) >>*INHIBIT(sonorant C), *INHIBIT(oral C)
This ranking ensures that vowels, glottals, and glides will not inhibit an anticipatory
velum opening gesture and will therefore not block harmony, while liquids and obstruents will
act as blockers. The tableau in (143) demonstrates this for the Capanahua form [wɯɾãnwɯ]
‘push it.’ In order to focus solely on the issue of blocking, only candidates with anticipatory,
self-deactivating gestures are considered here.
236
(143) Tableau for Capanahua [wɯɾãnwɯ] ‘push it’
Input: / w
1
ɯ
2
ɾ
3
a
4
n
5
w
6
ɯ
7
/
*INHIBIT(glide)
*OVERLAP(oral C)
*INHIBIT(sonorant C)
*INHIBIT(oral C)
a.
*!*
b.
* *
In (143), candidate (a) [w
̃ɯ
̃ɾ
̃ ãnwɯ] includes a velum opening gesture that has extended to
the beginning of the word, satisfying all of the *INHIBIT constraints but fatally violating high-
ranked *OVERLAP(oral C) twice. The winning candidate (b) [wɯɾ
̃ ãnwɯ] violates the low-ranked
*INHIBIT(sonorant C) and *INHIBIT(oral C) and avoids violation of *OVERLAP(oral C) by
including an inhibition relation between the tap gesture of [ɾ] and the velum opening gesture. The
result of this inhibition relation is the blocking of regressive nasal harmony.
Lip
protr
1
TB
uvu nar
2
TT
alv tap
3
TB
phar wide
4
TT
alv clo
5
Lip
protr
6
TB
uvu nar
7
Velum
open
5
237
This ranking of the *INHIBIT constraints relative to *OVERLAP(oral C, velum opening)
also accounts for the status of obstruents as blockers and of glides as undergoers of nasal
harmony in Capanahua. This is illustrated by the tableau in (144) for the form [põȷ
̃ ãn] ‘arm.’
(144) Tableau for blocking of nasal harmony in Capanahua [põȷ
̃ ãn] ‘arm’
Input: / p
1
o
2
j
3
a
4
n
5
/
*INHIBIT(glide)
*OVERLAP(oral C)
*INHIBIT(sonorant C)
*INHIBIT(oral C)
a.
*! * *
b.
* *
In (144), candidate (a) [pojãn] includes an inhibition relation between the glide [j] and the
anticipatory velum opening gesture of word-final [n], blocking the spread of nasality beyond the
vowel of the second syllable. While this inhibition relation results in satisfaction of
Lip
clo
1
TB
uvu-phar nar
2
TB
pal nar
3
TB
phar wide
4
TT
alv clo
5
Velum
open
5
238
*OVERLAP(oral C), it incurs violations of all three *INHIBIT constraints, including a fatal
violation of high-ranked *INHIBIT(glide). In winning candidate (b), [põȷ
̃ ãn] ‘arm’ the anticipatory
velum opening gesture extends to overlap the glide [j], resulting in a violation of *OVERLAP(oral
C). However, it satisfies higher-ranked *INHIBIT(glide). Instead, this form includes an inhibition
relation between the gestures of [p] and the velum opening gesture of [n], resulting in a violation
of low-ranked *INHIBIT(oral C).
This ranking of constraints is thus able to capture the undergoing of nasal harmony by
glides and the blocking of harmony by liquids and obstruents in Capanahua. It also demonstrates
that the mechanism of blocking via intergestural inhibition in nasal harmony is available not just
to obstruents, but to sonorants as well.
4.4.5 Summary
This section has demonstrated the Gestural Harmony Model’s ability to use the dual
mechanisms of transparency via intergestural competition and blending and blocking via
intergestural inhibition to account for crosslinguistic patterns of transparency and blocking in
nasal harmony. The constraint set outlined for nasal harmony in section 4.4.1, which includes a
set of stringently defined *INHIBIT constraints, accurately captures the implicational hierarchy of
blocking observed across nasal harmony systems.
In addition, the gestural framework develops a classification of obstruent segments as
those whose gestural representations include a velum closure gesture. In the Gestural Harmony
Model, this representation of obstruents accurately predicts that they are the only class of
segments that possess the gestural makeup necessary to surface as transparent to harmony, in
keeping with previous typological observations by Piggott (1992) and Walker (1998/2000). This
is due to the model’s representation of a transparent obstruent as an undergoer of harmony that
239
includes a velum closure gesture that is antagonistic to the harmonizing velum opening gesture.
The variable representation of fricatives and voiced obstruents with respect to the inclusion of a
velum closure gesture also correctly predicts that these segment classes should vary as to
whether they pattern as obstruents or as sonorants in a given harmony system.
The adoption of both gestural antagonism and gestural inhibition by the Gestural
Harmony Model is crucial to its success in capturing typological asymmetries that arise among
nasal harmony systems. Such asymmetries are also attested in rounding harmony, and thus also
benefit from analysis within the Gestural Harmony Model. This is the subject of section 4.5
4.5 Transparency and Blocking in Rounding Harmony
4.5.1 Sources of Antagonism and Incompatibility
As discussed in section 4.2.2, rounding harmony displays a typological asymmetry in
attested patterns of transparency and blocking. In rounding harmony, the set of attested
transparent segment types is smaller than the set of attested blocking segment types. As in the
case of nasal harmony, I propose that within the Gestural Harmony Model this asymmetry is due
to the existence of two distinct theoretical mechanisms that are driven by related but distinct
consequences of gestural overlap. While the mechanism of transparency to rounding harmony
via intergestural competition is available to a small set of vowels due to their gestural makeup, a
wide range of vowels may block rounding harmony on the basis of several phonetically
grounded conditions proposed by Hong (1994) and Kaun (1995, 2004).
Recall from section 4.2.2 that while all vowels are attested as blockers in some rounding
harmony system, the only vowels that are attested as being transparent to rounding harmony are
the high front vowels /i/ and /ɪ/. Again, this restriction of the ability to surface as transparent to a
small class of segments is made possible by modeling transparency as a direct consequence of
240
overlap by a harmonizing gesture. I claim that in the Gestural Harmony Model, the transparency
of high front vowels in rounding harmony is a direct consequence of their gestural makeup. A
high front vowel is represented by a palatal constriction gesture of the tongue body; this alone is
insufficient to account for its transparency to rounding harmony. However, there is evidence that
the production of high front vowels is also characterized by active lip spreading, posited as a
means of raising the second formant of high front vowels in order to maximize their perceptual
distance from back vowels. The presence of active lip spreading during the production of high
front vowels is supported by the findings of several articulatory studies, including those
conducted by Hadding, Hirose, & Harris (1976), Sussman & Westbury (1981), and Goldstein
(1991). These findings suggest that the representation of high front vowels should include a lip
spreading gesture in addition to a palatal tongue body gesture.
The articulatory effects of active lip protrusion and active lip spreading are illustrated in
(145). A gesture for lip protrusion is responsible for rounding harmony, as proposed in section
2.2.1; its effect is illustrated in (145a). In addition, I propose that the set of assumed gestural
representations should include a gesture for lip spreading, as in (145c), characteristic of the
production of some high front vowels.
(145) Active and neutral lip positions
a. Active Lip Protrusion
b. Neutral Lip Position
c. Active Lip Spreading
In comparing the position of the lips in (145a) and (145c), it is apparent that the lip
spreading gesture of a high front vowel is antagonistic to the lip protrusion gesture that is
241
responsible for rounding harmony. As a result, high front vowels whose representations include a
lip spreading gesture are predicted to be transparent to rounding harmony. No other vowels
include this gesture in their representations, and thus are unable to surface as transparent to
rounding harmony when overlapped by a lip protrusion gesture.
In many languages there is evidence that high front vowels are produced with active lip
spreading and that a lip spreading gesture should therefore be included in their phonological
representations. However, there are also languages in which high front vowels surface as
rounded in the domain of rounding harmony, suggesting that they are not accompanied by a lip
spreading gesture. Kyrgyz rounding harmony (sections 2.2.1 and 3.2.1) represents one such
case. It appears, then, that high front vowels vary across languages as to whether they are
accompanied by lip spreading gestures that result in transparency to rounding harmony. This
parallels the crosslinguistic variability in the inclusion of a velum closure gesture in the proposed
representations of fricatives and voiced obstruents in section 4.4.1.
With respect to blocking patterns in rounding harmony, Kaun (1995, 2004) proposes a set
of three markedness restrictions on round vowels whose interactions generate the rather complex
typology of blocking in rounding harmony. The first is a restriction on nonhigh round vowels,
motivated primarily by the findings of Linker (1982) that nonhigh vowels tend to be produced
with less rounding and are therefore marked for perceptual reasons. Hong (1994) proposes a
similar restriction on vowel height and rounding, drawing on additional evidence relating jaw
height to lip rounding presented by Lindblom & Sundberg (1971). While few languages ban
nonhigh round vowels altogether, many languages place restrictions on where these vowels are
allowed to occur in a word, and whether they may be derived as a result of vowel harmony.
242
As a precursor to defining the constraints necessary for an account of blocking in
rounding harmony, which must reference vowels according to their height and backness, I
provide the gestural definitions of vowel height and backness in (146). These definitions are
revisited in section 6.2.1.
(146) Gestural definitions of vowel place
a. High vowel: a segment whose primary vocalic tongue body gesture is specified for
narrow constriction degree in the palatal or uvular region
b. Nonhigh vowel: a segment whose primary vocalic tongue body gesture is not
specified for narrow constriction degree in the palatal or uvular region
c. Front vowel: a segment whose primary vocalic tongue body gesture is specified for
constriction in the palatal region
d. Back vowel: a segment whose primary vocalic tongue body gesture is specified for
constriction in the uvular or pharyngeal region
The restriction on the co-occurrence of a nonhigh vowel with lip rounding can be
accounted for within the Gestural Harmony Model using a constraint from the *OVERLAP family.
It is defined in (147).
(147) *OVERLAP(nonhigh vowel, lip protrusion): Assign a violation mark to a nonhigh vowel
gesture and a lip protrusion gesture that are concurrently active.
When this constraint outranks the anti-blocking constraint *INHIBIT, nonhigh vowels will
block rounding harmony rather than undergoing it. A similar restriction proposed by Kaun
(1995, 2004) that affects the typology of blocking in rounding harmony is a restriction on front
round vowels. Front vowels are also perceptually disadvantaged bearers of rounding, as rounding
renders them less acoustically distinct from back vowels. In this case, some languages do impose
a total ban on front round vowels in surface forms. Similar to (147), the avoidance of front round
vowels is accounted for by a *OVERLAP constraint, defined in (148).
243
(148) *OVERLAP(front vowel, lip protrusion): Assign a violation mark to a front vowel gesture
and a lip protrusion gesture that are concurrently active.
Finally, many rounding harmony patterns are governed by a principle which Kaun (1995,
2004) refers to as Gestural Uniformity. According to this principle, the span of a single [round]
autosegment must be realized uniformly with respect to jaw height; sequence such as [u-o] or
[o-u] in which both vowels are linked to the same instance of the feature [round] are ill-formed.
Complicating matters somewhat is the fact that Gestural Uniformity is often used by Kaun in
tandem with constraints on harmony trigger conditions in order to account for cases in which
[o-u] sequences are permitted while [u-o] sequences are not. This strategy for accounting for the
directional asymmetry in the effect of Gestural Uniformity is not available to the Gestural
Harmony Model; the reason for this and the modifications necessary to adapt Gestural
Uniformity for use within the Gestural Harmony Model is discussed in greater detail in the
examination of Yakut rounding harmony in section 4.5.5.
The remainder of this section demonstrates the workings of each of these markedness
constraints in shaping the patterns of blocking of rounding harmony by gestures that are
incompatible with rounding, as well as the representation of transparency resulting from the
antagonism of lip protrusion and lip spreading gestures. These are illustrated with examinations
of the rounding harmony systems of Halh Mongolian, Baiyina Oroqen (first introduced in section
3.4.2), Tuvan, and Yakut.
4.5.2 Halh Mongolian: Transparency and Blocking by High Vowels
Halh Mongolian (Mongolic; Mongolia; Svantesson (1985), Steriade (1987), van der Hulst
& N. Smith (1987), Svantesson et al. (2005)) presents an especially interesting case of rounding
harmony that exhibits both transparency and blocking. While high back round vowels block
harmony, high front vowels are transparent to it. In Gestural Harmony Model, blocking and
244
transparency are the products of two distinct theoretical mechanisms that are able to operate
independently of one another, and can even operate concurrently. As a result, the Gestural
Harmony Model is particularly well suited to analyzing the pattern of transparency and blocking
exhibited by Halh Mongolian rounding harmony.
Halh Mongolian has the vowel inventory in (149), as reported by Svantesson et al.
(2005).
28
(149) Halh Mongolian vowel inventory
Non-Pharyngeal Pharyngeal
High i u ɪ ʊ
Nonhigh (e) ɵ a ɔ
Halh has two processes of vowel harmony, one based on tongue root position and one
based on rounding. Tongue root harmony divides the vowel inventory into two classes,
pharyngeal (RTR) and non-pharyngeal (non-RTR). Words may contain vowels from only one of
these classes, as in (150).
29
(150) a. [xeeɮ-uɮ-ɮe] ‘to decorate (caus. past)’ e. [jaw-ʊɮ-ɮa] ‘to go (caus. past)’
b. [piːr-ig-e] ‘brush (acc. refl.)’ f. [mʊːr-ɪg-a] ‘cat (acc. refl.)’
c. [suːɮ-ig-e] ‘tail (acc. refl.)’ g. [xʊn
j
-ʊɮ-ɮa] ‘to pleat (caus. past)’
d. [pɵːr-ɵ] ‘kidney (refl.)’ h. [xɔːɮ-ɔ] ‘food (refl.)’
Halh also has a process of rounding harmony that holds only among nonhigh vowels,
with [e] alternating with [ɵ] and [a] alternating with [ɔ]. It is triggered by a nonhigh round vowel
in the initial syllable of a word. The data in (151) illustrate; compare the rounded vowels of the
past suffix in (151) to their unrounded variants in (150a,e,g).
28
Svantesson et al. (2005) transcribe the nonhigh non-pharyngeal round vowel as [o] but state that its pronunciation
is closer to central [ɵ]. Their acoustic data confirms that this vowel is central, and it is transcribed as such here. The
vowels in parentheses occur only in non-initial syllables.
29
The high front vowel is often claimed to be transparent to tongue root harmony and transcribed as [i] even in RTR
(pharyngeal) words. However, the acoustic data reported by Svantesson et al. indicate that this vowel undergoes
tongue root harmony just as the other vowels in the inventory do, surfacing as [ɪ] in RTR words. Therefore, this
vowel will be transcribed as [ɪ] here.
245
(151) a. [ɵg-ɮɵ] ‘to give (past)’
b. [c
h
ɵːr- ɮɵ] ‘to decrease (past)’
c. [ɔr-ɮɔ] ‘to enter (past)’
d. [c
h
ɔːr- ɮɔ] ‘to be pierced (past)’
The high vowels do not participate in rounding harmony. High front [i] and [ɪ] are
transparent to rounding harmony, neither undergoing it nor blocking it from spreading further, as
in the data in (152).
(152) a. [pɵːr-ig-ɵ] ‘kidney (acc. refl.)’
b. [ɵːr-ig-ɵ] ‘self (acc. refl.)’
c. [xɔːɮ-ɪg-ɔ] ‘food (refl.)’
d. [ɔɮ
j
ɪ-ɮɔ] ‘to squint (past)’
As in Baiyina Oroqen (sections 3.4.2 and 4.5.3), the high round vowels /u/ and /ʊ/ pattern
quite differently from the nonhigh round vowels. The vowels /u/ and /ʊ/ do not trigger rounding
harmony (153a-d), and are unrestricted in their distribution (153e-f), surfacing in initial and non-
initial syllables. In contrast, the nonhigh round vowels /ɵ/ and /ɔ/ may only occur in an initial
syllable, or in a non-initial syllable as the product of rounding harmony. When the high round
vowels occur after triggering nonhigh vowels, they block rounding harmony despite being round
themselves (153g-h).
(153) a. [uc-ɮe] ‘jump (past)’ e. [it-uɮ-ɮe] ‘to eat (caus. past)’
b. [suːɮ-e] ‘tail (refl.)’ f. [jaw-ʊɮ-ɮa] ‘to go (caus. past)’
c. [xʊn
j
-ɮa] ‘to pleat past)’ g. [ɵg-uɮ-ɮe] ‘to give (caus. past)’
d. [mʊːr-a] ‘cat (refl.)’ h. [ɔr-ʊɮ-ɮa] ‘to enter (caus. past)’
Halh Mongolian is an example of a harmony system that exhibits both transparency and
blocking; while high front vowels are transparent to rounding harmony, high back vowels serve
as blockers. It is also a case in which non-triggering bearers of a harmonizing property, the high
back round vowels, block harmony from spreading further. The Gestural Harmony Model is able
to account for all of this straightforwardly due to its reliance on two distinct mechanisms for
analyzing transparency and blocking: competition between antagonistic gestures and inhibition
246
between incompatible gestures. In Halh Mongolian, both mechanisms play an active role in
shaping the patterns of transparency and blocking of rounding harmony among the high vowels.
The transparency of high front /i/ and /ɪ/ can be attributed to their inclusion of both a
tongue body gesture specified for narrow palatal constriction as well as a lip spreading gesture.
(RTR /ɪ/ also includes a tongue root retraction gesture.) These vowels’ lip spreading gestures are
antagonistic to the lip protrusion gesture responsible for rounding harmony in this language.
When a persistent (non-self-deactivating) lip protrusion gesture overlaps a following high front
vowel, it results not in rounding of that vowel, but in transparency to the rounding harmony
process.
The coupling graph in (154) for [pɵːr-ig-ɵ] ‘kidney (acc. refl.)’ includes a lip spreading
gesture in the representation of /i/, which is responsible for its transparency in rounding
harmony. When this coupling graph is input to the Coupled Oscillator Model and the Task
Dynamic Model to calculate the gestural score and the articulatory trajectories for this form,
transparency will result directly from the gestural makeup of transparent /i/. There is no need for
the phonological grammar to independently enforce transparency.
(154) Coupling graph for [pɵːr-ig-ɵ] ‘kidney (acc. refl.)’
The gestural score in (155) depicts the vocalic portion of the Halh word [pɵːr-ig-ɵ]
‘kidney (acc. refl.)’ that is constructed from the coupling graph in (154). The persistent lip
247
protrusion gesture extends throughout the word, resulting in rounding harmony. However, here
its overlap with the gestures of /i/ results not in rounding of the vowel, but in transparency. The
lip spreading gesture included in the representation of /i/ is antagonistic to the lip protrusion
gesture that overlaps it. Because the lip spreading gesture is specified for a high gestural strength
while the harmonizing lip protrusion gesture is specified for a relatively lower strength, blending
of their antagonistic target articulatory states during the period of their concurrent activation is
resolved in favor of the lip spreading gesture of /i/. When the production of /i/ ceases, the lips are
free to return to the protruded position required by the lip protrusion gesture. The dashed line
represents the neutral value for lip protrusion.
(155) Gestural score for vocalic portion of Halh Mongolian [pɵːr-ig-ɵ] ‘kidney (acc. refl.)’
Despite the temporary lack of rounding during the production of the vowel [i], the
harmonizing lip protrusion gesture is active throughout the entire word. It is simply prevented
from achieving its target articulatory state during the production of the high front vowel as a
result of that vowel’s inclusion of an antagonistic lip spreading gesture in its gestural
representation. The phonological grammar’s only role here is in allowing the transparent vowel
to be overlapped by the harmonizing lip protrusion gesture; the Coupled Oscillator Model and
the Task Dynamic Model of speech production do the rest.
248
However, the phonological grammar does play a role in enforcing the blocking of
rounding harmony in Halh Mongolian. In this rounding harmony system, the blockers of
harmony are precisely those vowels that do not serve as triggers, despite being round vowels.
Within the Gestural Harmony Model, this can be accounted for with a single *OVERLAP
constraint that penalizes the overlap of a high back vowel with a persistent lip protrusion gesture.
This constraint will account for both the non-triggering and blocking properties of high back
round vowels /u/ and /ʊ/.
In the analysis of trigger conditions in Baiyina Oroqen rounding harmony (section 3.4.2),
the status of nonhigh round vowels as triggers of harmony is accounted for with a *COUPLE
constraint that penalizes the coupling of a nonhigh vowel gesture to a self-deactivating lip
protrusion gesture. A similar tactic can be used to account for non-triggers of harmony in Halh
Mongolian by penalizing the co-occurrence of a high back vowel gesture with a persistent (non-
self-deactivating) lip protrusion gesture. However, rather than using a constraint from the
*COUPLE family, I employ a *OVERLAP constraint in order to simultaneously account for both
the non-triggering and blocking exhibited by high back vowels. This constraint is defined in
(156).
(156) *OVERLAP(high back vowel, persistent lip protrusion): Assign a violation mark to a high
back vowel gesture that is active concurrently with a persistent lip protrusion gesture.
As discussed in section 1.2.2, *OVERLAP constraints are stricter than *COUPLE constraints
in that they penalize any co-occurrence of two gestures, whether they are coupled to one another
or whether they are concurrently active due to one of the gestures extending to overlap the other.
In the case of Halh Mongolian rounding harmony, the stricter of these two constraints is
necessary to explain both the non-triggering and blocking of high back round vowels. The effect
of the *OVERLAP constraint in (156) is twofold. First, it will prevent high back vowels /u/ and /ʊ/
249
from surfacing with a persistent lip protrusion gesture, rendering them non-triggers of rounding
harmony. Second, it will prevent them from being overlapped by a persistent gesture that is
coupled to another vowel that triggers harmony, rendering them blockers of rounding harmony.
In Halh Mongolian, /u/ and /ʊ/ never trigger rounding harmony, indicating that
*OVERLAP(high back vowel, persistent lip protrusion) is never violated. It must therefore be
ranked above IDENT(deactivation)-IO and PERSIST(lip protrusion). It must also be ranked above
*INHIBIT in order to capture the fact that /u/ and /ʊ/ block rounding harmony. The triggering of
rounding harmony by nonhigh vowels is achieved by ranking PERSIST(lip protrusion) above
IDENT(deactivation)-IO and SELFDEACTIVATE. This ranking is summarized in the Hasse diagram
in (157).
(157) Constraint ranking for triggering and blocking of Halh Mongolian rounding harmony
With the constraint ranking in (157), the status of high back vowels as non-triggers of
rounding harmony is demonstrated in the tableau in (158) for the Halh Mongolian word [uc-ɮe]
‘jump (past).’ It includes a hypothetical input with a persistent lip protrusion gesture to
demonstrate the effect of high-ranked *OVERLAP. For reasons of space and clarity, only the
vocalic portions of output candidates are included, and candidates are represented in gestural
score form.
250
(158) Tableau for Halh Mongolian [uc-ɮe] ‘jump (past)’
Input: / u
1
c - ɮ e
2
/
*OVERLAP(high back V, persist. LP)
PERSIST(lip protrusion)
IDENT(deactivation)-IO
SELFDEACTIVATE
*INHIBIT
a. [uc-ɮɵ]
*! *
b. [uc-ɮe]
* *
In candidate (a) [uc-ɮɵ], the [u] in the initial syllable triggers harmony, satisfying
PERSIST(lip protrusion) but violating higher-ranked *OVERLAP(high back vowel, persistent lip
protrusion). In the winning candidate (b) [uc-ɮe], the initial /u/ surfaces with a self-deactivating
lip protrusion gesture, satisfying *OVERLAP(high back vowel, persistent lip protrusion) at the
expense of lower-ranked PERSIST(lip protrusion) and IDENT(deactivation)-IO.
The effect of the *OVERLAP constraint in casting the high back vowels as both non-
triggers and blockers of rounding harmony is demonstrated in the tableau in (159) for [og-uɮ-ɮe]
‘to give (caus. past).’ In this form, nonhigh /ɵ/ in the initial syllable triggers rounding harmony,
Tongue Body
uvular narrow
1
Tongue Body
palatal wide
2
Lip
protrusion
1
251
and high /u/ blocks it. A hypothetical input in which /ɵ/ is self-deactivating is considered here in
order to demonstrate that PERSIST(lip protrusion) must dominate IDENT(deactivation)-IO.
(159) Tableau for [ɵg-uɮ-ɮe] ‘to give (caus. past)’
Input: / ɵ
1
g - u
2
ɮ - ɮ e
3
/
*OVERLAP(high back V, persist. LP)
PERSIST(lip protrusion)
IDENT(deactivation)-IO
SELFDEACTIVATE
*INHIBIT
a.
*! * * *
b.
**!
c.
* * * *
In (159), the lip protrusion gesture of initial /ɵ/ is self-deactivating in the input. All
candidates incur one violation of PERSIST(lip protrusion) for the self-deactivating lip protrusion
gesture of [u]. Candidate (a) [ɵg-uɮ-ɮɵ] surfaces with a persistent lip protrusion gesture
Tongue Body
palatal wide
1
Tongue Body
uvular narrow
2
Tongue Body
palatal wide
3
Lip
protrusion
1
Lip
protrusion
2
252
accompanying [ɵ], violating low-ranked IDENT(deactivation)-IO and SELFDEACTIVATE but
avoiding additional violation of higher-ranked PERSIST(lip protrusion). However, this candidate
fatally violates highest-ranked *OVERLAP(high back vowel, persistent lip protrusion) because the
harmonizing lip protrusion gesture overlaps the following [u]. In candidate (b) [ɵg-uɮ-ɮe], /ɵ/
surfaces faithfully with a self-deactivating lip protrusion gesture, failing to trigger rounding
harmony and incurring a second violation of PERSIST(lip protrusion). The winning candidate (c)
[ɵg-uɮ-ɮe] surfaces with a persistent lip protrusion gesture accompanying [ɵ]; however,
harmony is blocked by the following [u] via an inhibition relation between the tongue body
gesture of [u] and the lip protrusion gesture of triggering [ɵ]. While this candidate violates low-
ranked constraints IDENT(deactivation)-IO, SELFDEACTIVATE, and *INHIBIT, it satisfies the high-
ranked constraints *OVERLAP(high back vowel, persistent lip protrusion) (via blocking by [u])
and PERSIST(lip protrusion) (via triggering).
In the Gestural Harmony Model’s analysis of Halh Mongolian rounding harmony, the
blocking and non-triggering of round harmony exhibited by the back round vowels /u/ and /ʊ/ is
given a unified explanation. The constraint *OVERLAP(high back vowel, persistent lip protrusion)
prevents the co-occurrence of high back vowels and persistent lip protrusion gestures, whether
they are part of the same segment in the input or whether they come into contact with one
another due to rounding harmony. This *OVERLAP constraint is able to shape the surface
phonological inventory of Halh Mongolian such that a triggering /u/ or /ʊ/ will never surface. It
can also motivate the inclusion of an inhibition relation between a high back vowel gesture and a
preceding persistent lip protrusion gesture, resulting in the blocking of rounding harmony.
The transparency of high front vowels, on the other hand, is due not to any constraints in
the grammar but rather to their gestural makeup. High front /i/ and /ɪ/ are accompanied by lip
253
spreading gestures, which are antagonistic to harmonizing lip protrusion gestures. Because
transparency to rounding harmony is restricted only to those vowels that include a lip protrusion
gesture in their gestural representations, the Gestural Harmony Model predicts that the high front
vowels /i/ and /ɪ/ are the only vowels that may be transparent to rounding harmony. This is a
major advantage of the Gestural Harmony Model, as this prediction matches the crosslinguistic
typology of attested transparency in rounding harmony outlined in section 4.2.2.
Another important aspect of the Gestural Harmony Model’s analysis of Halh Mongolian
rounding harmony is that the two mechanisms responsible for transparency and blocking may
operate concurrently. Transparency of the high front vowels is the result of competition and
blending of antagonistic gestures, and has no basis in constraint interaction. The constraints
necessary to account for the blocking and non-triggering of rounding harmony exhibited by back
vowels will apply only to those back vowels, and will not affect the transparency of high front
vowels in any way. This separation of the analyses of transparency and blocking is crucial to the
Gestural Harmony Model’s success in accounting for Halh Mongolian rounding harmony, in
which both transparent and blocking segments occur.
The use of a *OVERLAP constraint to account for both blocking and non-triggering by
bearers of a harmonizing property can be applied to several other cases of Tungusic rounding
harmony as well. A similar pattern is found in Baiyina Oroqen. An analysis of nonhigh vowels’
participation in this rounding harmony system was presented in section 3.4.2; the following
section revisits Baiyina Oroqen and completes the analysis of its rounding harmony system.
4.5.3 Revisiting Baiyina Oroqen: Non-Triggers as Blockers and Undergoers
This section returns to the analysis of rounding harmony in Baiyina Oroqen (Li 1996;
Kaun 2004; Walker 2014; Dresher & Nevins 2017), previously taken up in section 3.4.2. As in
254
the case of Halh Mongolian rounding harmony discussed in the previous section, in Baiyina
Oroqen only a subset of round vowels trigger rounding harmony. What distinguishes Baiyina
Oroqen rounding harmony as a particularly interesting case is that some non-triggering round
vowels propagate rounding harmony, while others block it. The Gestural Harmony Model is able
to account for these complex patterns of triggering, non-triggering, undergoing, and blocking
straightforwardly via the use of constraints from both the *COUPLE and *OVERLAP families.
Recall that Baiyina Oroqen exhibits harmony for both tongue root position and rounding.
The vowel inventory in (160), reported by Li (1996), is repeated from (76) in section 3.4.2.
(160) Baiyina Oroqen vowel inventory
ATR Vowels RTR Vowels
High i iː u uː ɪ ɪː ʊ ʊː
Nonhigh ie ə əː o oː ɪɛ a aː ɔ ɔː
While there are a number of round vowels in the language’s inventory, rounding harmony
in Baiyina Oroqen holds only among the set of non-high back vowels. Harmony is triggered only
by the nonhigh short vowels /o/ and /ɔ/ in an initial syllable, as in (161).
(161) a. [tʃolpon] ‘morning star’
b. [somsok-jo] ‘pasture (indef. acc.)’
c. [ɔrɔktɔ] ‘hay’
d. [ɔlɔ-jɔ] ‘fish (indef. acc.)’
Long vowels /oː/ and /ɔː/ never trigger rounding harmony in an initial syllable (162a-d),
but may surface as the products of harmony and propagate that harmony onto following nonhigh
vowels (162e-h).
(162) a. [koːməxə] ‘windpipe’ e. [ɲoɲo-xoːn-mo] ‘bear (dim. def. acc.)’
b. [boːl-jə] ‘slave (indef. acc.)’ f. [oloː-wkoːn-no] ‘to cook (caus. pres.)’
c. [kɔːŋakta] ‘handbell’ g. [dʒɔlɔ-xɔːn-mɔ] ‘stone (dim. def. acc.)’
d. [gɔːl-ja] ‘policy (indef. acc.)’ h. [bɔdɔ-kwɔːn-nɔ] ‘to think (caus. pres.)’
255
The patterning of nonhigh vowels in Baiyina Oroqen rounding harmony is accounted for
in section 3.4.2 by proposing an inventory in which only underlyingly short nonhigh round
vowels are accompanied by persistent lip protrusion gestures. This is accomplished by ranking
the constraint *COUPLE(short nonhigh vowel, self-deactivating lip protrusion), which penalizes
non-triggering /o/ and /ɔ/, above the constraint SELFDEACTIVATE. Under this ranking, short /o/
and /ɔ/ will be compelled to surface with persistent lip protrusion gestures, and will therefore
trigger harmony. Meanwhile, the ranking of SELFDEACTIVATE over PERSIST(lip protrusion)
compels long /oː/ and /ɔː/ to surface with self-deactivating lip protrusion gestures, rendering
them non-triggers of harmony. However, none of these constraints prevents long nonhigh vowels
from undergoing harmony, or from propagating it to following nonhigh vowels. In a non-initial
syllable, /oː/ and /ɔː/ are round due to overlap by the lip protrusion gesture of a triggering vowel
in an initial syllable. They are not coupled to their own lip protrusion gestures, and they do not
deactivate the lip protrusion gestures of other round vowels. The surface phonological inventory
generated by this constraint ranking is provided in (163), repeated from (79) in section 3.4.2.
256
(163) Baiyina Oroqen round vowel inventory represented gesturally
/u/ /ʊ/ /o/ /ɔ/
/uː/ /ʊː/ /oː/ /ɔː/
Not yet accounted for is the distribution of the high round vowels /u/, /uː/, /ʊ/, and /ʊː/,
and their patterning with respect to rounding harmony. In both respects, the high round vowels
are markedly different from the nonhigh round vowels. While nonhigh round vowels are
restricted to the initial syllable (unless they are derived by rounding harmony), high round
vowels may occur in any position in the word, as in (164).
(164) a. [bəjun] ‘elk’
b. [pəntuː] ‘pilose antler’
c. [talʊ] ‘birch bark’
d. [akkʊː] ‘filled, solid’
257
Also unlike nonhigh round vowels /o/ and /ɔ/, the high round vowels never act as triggers
of rounding harmony. The data in (165) illustrates this.
(165) a. [urə-jə] ‘mountain (indef. acc.)’
b. [buː-wkəːn-nə-] ‘to give (caus. pres.)’
c. [luxi-xəːn-mə] ‘arrow (dim. def. acc.)’
d. [ʊnta] ‘leather shoe’
e. [tʃʊːxa] ‘grass’
f. [mʊrɪn-a] ‘horse (indef. acc.)’
Finally, when high round vowels occur following triggering /o/ or /ɔ/, these high round
vowels block rounding harmony rather than propagating it, as in (166).
(166) a. [owon-duləː] ‘pancake (destin.)’
b. [ɔrɔn-dʊlaː] ‘reindeer (destin.)’
This blocking of rounding harmony extends to high vowels more generally. Following
triggering /o/ or /ɔ/, the high front vowels /i/ and /ɪ/, as well as the diphthongs /ie/ and /ɪɛ/, also
block rounding harmony, as in (167).
(167) a. [moliktə] ‘kind of wild fruit’
b. [bolboxi-wə] ‘wild duck (def. acc.)’
c. [bomboŋkie-wə] ‘Shaman’s hat (def. acc.)’
d. [ɔxɪxan] ‘flame’
e. [ɔmɔlɪɛ-xal] ‘grandson (pl.)’
(c.f. [bɔdɔ-xɔl] ‘to think (int. imp. 2sg.)’)
f. [tʃɔlɪk-pa] ‘cloud-shaped design (def. acc.)’
A successful account of Baiyina Oroqen rounding harmony must account for the
following observations. (1) A condition is placed on which round vowels may be triggers of
harmony. (2) There are two disparate distributional patterns for non-triggers. On the one hand,
long nonhigh /oː/ and /ɔː/ fail to trigger harmony, but will propagate it; on the other hand, high
/u/, /uː/, /ʊ/, and /ʊː/ fail to trigger harmony and block it from spreading any further. The
Gestural Harmony Model is able to account for such a pattern due to having two different types
of gestural co-occurrence constraints at its disposal: *COUPLE and *OVERLAP.
258
I interpret the patterns of non-triggering of rounding harmony in Baiyina Oroqen as
follows. Non-triggering long nonhigh /oː/ and /ɔː/ may not be coupled to a persistent lip
protrusion gesture, though they are not prohibited from being overlapped by the lip protrusion
gesture of another vowel. The result of this coupling-based restriction is that these vowels will
neither trigger nor block harmony. However, high vowels are entirely prohibited from being
concurrently active with a persistent lip protrusion gesture. The result of this stricter co-
occurrence restriction is that a high vowel may neither be coupled to a persistent lip protrusion
gesture, nor may it be overlapped by a lip protrusion gesture as a result of harmony.
The patterning of the high back vowels as non-triggers and blockers of harmony in
Baiyina Oroqen parallels the role of the high back vowels in the rounding harmony of Halh
Mongolian (section 4.5.2). The only difference is that in Baiyina Oroqen the high front vowels
also block rounding harmony. From this perspective, the non-triggering and blocking behavior of
high vowels in Baiyina Oroqen can be analyzed in a similar way, via the high ranking of a
constraint from the *OVERLAP family.
30
The constraint necessary for the analysis of Baiyina
Oroqen is defined in (168).
(168) *OVERLAP(high vowel, persistent lip protrusion): Assign a violation mark to a high vowel
gesture and a persistent lip protrusion gesture that are concurrently active.
The constraint in (168) disallows the concurrent activation of a high vowel gesture (either
back or front) and a persistent lip protrusion gesture. When this constraint is ranked above
*INHIBIT, an inhibition relation will be included in a coupling graph such that the concurrent
30
It is also possible that two separate *OVERLAP constraints are responsible for blocking by high vowels in Baiyina
Oroqen, one for front vowels and one for back vowels. While the high back vowels are prohibited from co-occurring
with persistent lip protrusion gestures, it appears that the high front vowels are prohibited from co-occurring with lip
protrusion gestures of any kind, as evidenced by the complete lack of high front round vowels in the language.
While the use of one versus two *OVERLAP constraints has consequences for shaping the phonological inventory of
the language, it is not crucial to the analysis of blocking laid out here. To simplify the analysis, I assume a single
*OVERLAP constraint.
259
activation of a persistent lip protrusion gesture and a high vowel gesture is avoided. This applies
whether the high vowel is round or unround. Because the nonhigh front vowels are
diphthongized and include a high vowel portion, they are also rendered blockers by
*OVERLAP(high vowel, persistent lip protrusion). In addition, this constraint ensures that any
high round vowels do not trigger rounding harmony, as they are prohibited from surfacing with a
persistent lip protrusion gesture.
The full ranking of constraints necessary to generate the pattern of triggering, non-
triggering, blocking, and undergoing in Baiyina Oroqen is provided in the Hasse diagram in
(169).
(169) Constraint ranking for Baiyina Oroqen rounding harmony
The ranking *COUPLE(nonhigh vowel, self-deactivating lip protrusion) >>
SELFDEACTIVATE >> IDENT(deactivation)-IO, PERSIST(LP) has already been shown to account
for the role of nonhigh vowels in Baiyina Oroqen rounding harmony in section 3.4.2. The
constraint *OVERLAP(high vowel, persistent lip protrusion) can now be added to the ranking to
account for the non-triggering and blocking behavior of high vowels as well.
The tableau in (170) illustrates the selection of [ʊnta] ‘leather shoe,’ in which the high
vowel /ʊ/ does not trigger rounding harmony, as the optimal candidate, even with a hypothetical
input in which /ʊ/ is accompanied by a persistent lip protrusion gesture. Again, for reasons of
260
space and clarity, I only include vowel gestures, and output candidates are displayed as gestural
scores.
(170) Tableau for Baiyina Oroqen [ʊnta] ‘leather shoe’
Input: / ʊ
1
n t a
2
/
*COUPLE(nonhigh short V)
SELFDEACTIVATE
*OVERLAP(high V)
IDENT(deactivation)-IO
PERSIST(lip protrusion)
*INHIBIT
a. [ʊntɔ]
*! *!
b. [ʊnta]
* *
In candidate (a) [ʊntɔ], high back round [ʊ] surfaces faithfully with a persistent lip
protrusion gesture, triggering harmony and overlapping the following nonhigh vowel. This
violates both SELFDEACTIVATE and *OVERLAP(high vowel, persistent lip protrusion); either of
these violations is fatal. The winner is candidate (b) [ʊnta], in which the lip protrusion gesture of
[ʊ] is self-deactivating and harmony is not triggered. This candidate violates low-ranked
IDENT(deactivation)-IO and PERSIST(lip protrusion), but satisfies higher-ranked
SELFDEACTIVATE and *OVERLAP(high vowel, persistent lip protrusion).
Tongue Body
uvular narrow
1
Tongue Body
pharyngeal wide
2
Lip
protrusion
1
261
The same high ranking of *OVERLAP(high vowel, persistent lip protrusion) that renders
high back /u/ and /ʊ/ non-triggers of harmony is responsible for their status as blockers. The
tableau in (171) illustrates this for the Baiyina Oroqen form [ɔrɔn-dʊlaː] ‘reindeer (destin.).’ For
reasons of space, I do not represent the extended length of the vowel [aː] in the candidate
gestural scores.
(171) Tableau for Baiyina Oroqen [ɔrɔn-dʊlaː] ‘reindeer (destin.)’
Input: / ɔ
1
r ɔ
2
n – d ʊ
3
l aː
4
/
*COUPLE(nonhigh short V)
SELFDEACTIVATE
*OVERLAP(high V)
IDENT(deactivation)-IO
PERSIST(lip protrusion)
*INHIBIT
a. [ɔran-dʊlaː]
*! **
b. [ɔrɔn-dʊlɔː]
* *! * *
c. [ɔrɔn-dʊlaː]
* * * *
TB
phar wide
1
TB
phar wide
2
TB
uvu nar
3
TB
phar wide
4
Lip
protrusion
1
Lip
protrusion
3
262
In (171), candidate (a) [ɔran-dʊlaː] is ruled out immediately by undominated
*COUPLE(nonhigh short vowel, self-deactivating lip protrusion), as it contains a non-triggering
short nonhigh [ɔ]. The two remaining candidates both contain harmony-triggering [ɔ], which is
accompanied by a persistent lip protrusion gesture. This satisfies *COUPLE(nonhigh short vowel,
self-deactivating lip protrusion), at the expense of lower-ranked SELFDEACTIVATE (as well as
IDENT(deactivation)-IO, due to the hypothetical input with a self-deactivating lip protrusion
gesture). In candidate (b) [ɔrɔn-dʊlɔː], the high vowel [ʊ] is overlapped by the harmonizing lip
protrusion gesture, resulting in full harmony throughout the word. This satisfies low-ranked
*INHIBIT but fatally violates *OVERLAP(high vowel, persistent lip protrusion). The winning
candidate is (c) [ɔrɔn-dʊlaː], in which an inhibition relation between the vocalic gesture of [ʊ]
and the lip protrusion gesture of [ɔ] results in the blocking of harmony. This violates *INHIBIT
but satisfies higher-ranked *OVERLAP(high vowel, persistent lip protrusion).
The pattern of blocking in Baiyina Oroqen is distinguished from that of Halh Mongolian
by the fact that high front vowels also block harmony, rather than being transparent to it.
*OVERLAP(high vowel, persistent lip protrusion) will also account for this, as illustrated by the
tableau for [ɔxɪxan] ‘flame’ in (172). Having already established that a nonhigh round vowel is a
trigger of vowel harmony in Baiyina Oroqen regardless of the input deactivation parameter
specification of its lip protrusion gesture, the lip protrusion gestures in the input and both output
candidates in (172) are all persistent.
263
(172) Tableau for Baiyina Oroqen [ɔxɪxan] ‘flame’
Input: / ɔ
1
x ɪ
2
x a
3
n /
*COUPLE(nonhigh short V)
SELFDEACTIVATE
*OVERLAP(high V)
IDENT(deactivation)-IO
PERSIST(lip protrusion)
*INHIBIT
a. [ɔxʏxɔn]
* *
b. [ɔxɪxan]
* *
In candidate (a) [ɔxʏxɔn], the high front round vowel undergoes rounding harmony,
incurring a fatal violation of *OVERLAP(high V, persistent lip protrusion). Winning candidate (b)
[ɔxɪxan] eliminates the violation of this *OVERLAP constraint by including an inhibition relation
between [ɪ] and the harmonizing lip protrusion gesture. This blocking configuration violates only
low-ranked *INHIBIT.
31
31
A conceivable third candidate would look identical to candidate (a) but would include a lip spreading gesture in
the representation of [ɪ], rendering it transparent to rounding harmony. It would also be ruled out due to its violation
of *OVERLAP(high vowel, persistent lip protrusion), which requires all high vowels to block rounding harmony,
whether they include a lip spreading gesture or not. Because of this, it is impossible to tell without phonetic evidence
whether the production of high front vowels in Baiyina Oroqen involves active lip spreading that should be reflected
in their gestural makeup.
Tongue Body
pharyngeal wide
1
Tongue Body
pal nar
2
Tongue Body
pharyngeal wide
3
Lip
protrusion
1
264
The Gestural Harmony Model has proven itself successful in analyzing the complex
patterns of triggering, propagating, blocking, and undergoing of rounding harmony in Baiyina
Oroqen. This analysis accounts for the two distinct types of non-triggers in this system: the long
nonhigh round vowels that propagate harmony and the high round vowels that block it. More
generally, the Gestural Harmony Model provides a unified account of patterns of blocking
involving non-triggers of harmony, as exemplified by the cases of rounding harmony in Halh
Mongolian and Baiyina Oroqen. The non-triggering and blocking behaviors of high vowels are
both accounted for by constraints from the *OVERLAP family. The somewhat rarer case in which
a non-trigger of harmony propagates that harmony, rather than blocking it, also receives
explanation in this analysis as the result of a constraint from the *COUPLE family. The following
subsections focus on patterns of blocking in rounding harmony that are motivated by the
avoidance of gestural incompatibility, based on the markedness conditions discussed in section
4.5.1.
4.5.4 Tuvan: Blocking by Nonhigh Vowels
Tuvan rounding harmony represents a case in which blocking of harmony is necessary to
prevent the concurrent activation of incompatible gestures. As in many Turkic languages,
including Kyrgyz (sections 2.2.1 and 3.2.1), Tuvan (Northern Turkic; Republic of Tuva, Russia)
words harmonize for both backness and rounding. This section focuses on a gestural analysis of
Tuvan rounding harmony, which shows the effect of a restriction on nonhigh round vowels. The
language has a symmetrical inventory that contrasts vowels according to height, backness,
rounding, and length. The inventory in (173) is reported by Harrison (2000).
265
(173) Tuvan vowel inventory
Front Back
Unround Round Unround Round
High i iː y yː ɯ ɯː u uː
Nonhigh e eː ø øː a aː o oː
As in Kyrgyz, in Tuvan vowels harmonize with a vowel in the initial syllable of a word
for backness and rounding. Backness harmony proceeds;\ both within and across morpheme
boundaries, with all of the vowels in a word either front (174a-c) or back (174d-f). Data and
description are from Harrison (2000).
(174) a. [is-ter-im-den] ‘footprint (pl. 1p abl.)’ d. [at-tar-ɯm-dan] ‘name (pl. 1p abl.)’
b. [esker-be-di-m] ‘notice (neg. past 1p)’ e. [udu-va-dɯ-m] ‘sleep (neg. past 1p)’
c. [xøl-y] ‘lake (3p poss.)’ f. [toːl-u] ‘story (3p poss.)’
Rounding harmony, on the other hand, does not apply throughout entire words without
exception. As in many other Turkic languages, rounding harmony in Tuvan is triggered by a
round vowel in an initial syllable. While both high and nonhigh round vowels trigger harmony,
only high vowels are undergoers, surfacing as [y] and [u]. This is illustrated in (175).
(175) a. [ulu-zu] ‘dragon (3p poss.)’
b. [xol-u] ‘hand (3p poss.)’
c. [byry-zy] ‘wolf (3p poss.)’
d. [xøːr-y] ‘cemetery (3p poss.)’
cf. [ɯr-ɯ] ‘song (3p poss.)’
[is-i] ‘footprint (3p poss.)’
The nonhigh vowels do not undergo rounding harmony, and surface as unround [e] and
[a] after a triggering round vowel. These nonhigh vowels block rounding from spreading any
further, as in (176).
(176) a. [udu-va-dɯ-m] ‘sleep (neg. past 1p)’
b. [nom-nar-ɯm] ‘book (pl. 1p)’ (cf. [nom-um] ‘book (1p)’)
c. [byry-ler-niŋ] ‘leaf (pl. gen.)’ (cf. [byry-nyŋ] ‘book (gen.)’)
d. [xøl-der] ‘lake (pl.)’
266
From this data it is apparent that Tuvan does not exhibit any sort of trigger asymmetry;
both high and nonhigh round vowels are triggers of rounding harmony. However, high vowels
are allowed to undergo harmony while nonhigh vowels are not. This can be explained as the
result of the prohibition on nonhigh round vowels proposed by Kirchner (1993), Hong (1994)
and Kaun (1995, 2004). As discussed in section 4.5.1, Kirchner’s and Kaun’s constraints against
nonhigh round vowels (referred to by Kaun as *ROLO) are redefined within the Gestural
Harmony Model as a constraint from the *OVERLAP family. It penalizes the concurrent activation
of a lip protrusion gesture and a nonhigh vowel gesture and is defined in (177), repeated from
(147) in section 4.5.1.
(177) *OVERLAP(nonhigh vowel, lip protrusion): Assign a violation mark to a nonhigh vowel
gesture and a lip protrusion gesture that are concurrently active.
In order for this *OVERLAP constraint to enforce blocking of rounding harmony in Tuvan,
it must outrank *INHIBIT, which penalizes the deactivation of a harmony-triggering gesture.
However, *OVERLAP must itself be outranked by MAX(lip protrusion)-IO, which will prevent the
deletion of an input lip protrusion gesture. This ranking will preserve any underlying /o/ or /ø/
vowels while preventing a persistent lip protrusion gesture from deriving additional nonhigh
round vowels. Finally, MAX(lip protrusion)-IO must be outranked by LICENSE(lip protrusion,
first σ) (first proposed in section 3.2.1 for Kyrgyz rounding harmony) in order to account for the
restriction of round vowels to the initial syllable of the word in Tuvan (as reported by Harrison
(2000)). The full ranking is provided in (178):
(178) Constraint ranking for Tuvan rounding harmony:
LICENSE(lip protrusion, first σ) >> MAX(lip protrusion)-IO >>
*OVERLAP(nonhigh vowel, lip protrusion) >> *INHIBIT
267
The licensing constraint will not be included in the following analysis as it is
undominated in Tuvan, and any candidates that violate it will not be considered. The workings of
the rest of this constraint ranking are demonstrated in the following tableaux. The first, for the
Tuvan form [xol-u] ‘hand (3p poss.),’ is provided in (179). In this form, rounding harmony is
triggered by a nonhigh round vowel and targets a high vowel.
(179) Tableau for Tuvan [xol-u] ‘hand (3p poss.)’
Input: / x o
1
l – u
2
/
MAX(LP)-IO
*OVERLAP
( nonhigh V, LP)
*INHIBIT
a. [xol-u]
*
b. [xol-ɯ]
* *!
c. [xal-ɯ]
*!
In the winning candidate (a) [xol-u], the persistent lip protrusion gesture of the [o] in the
initial syllable overlaps the following vowel gesture. Because it is a high vowel gesture, this
overlap does not incur a violation of *OVERLAP(nonhigh vowel, lip protrusion). In candidate (b)
Tongue Body
pharyngeal wide
1
Tongue Body
uvular narrow
2
Lip
protrusion
1
268
[xol-ɯ] the overlap of the second vowel gesture by the lip protrusion gesture is prevented by an
inhibition relation that causes the lip protrusion gesture to deactivate when the following high
back vowel activates. This incurs a fatal violation of *INHIBIT. In candidate (c) [xal-ɯ], the lip
protrusion gesture of underlying /o/ has been deleted, incurring a fatal violation of MAX(lip
protrusion)-IO.
While full harmony is favored when a high vowel follows a triggering round vowel, the
situation is different when a potential undergoer of harmony is a nonhigh vowel. In this case,
harmony is blocked in order to prevent violation of *OVERLAP(nonhigh vowel, lip protrusion).
This is demonstrated in the tableau for [nom-nar-ɯm] ‘book (pl. 1p)’ in (180). In this form,
harmony is blocked by [a] in the plural suffix.
269
(180) Tableau for Tuvan [nom-nar-ɯm] ‘book (pl. 1p)’
Input: / n o
1
m – n a
2
r – ɯ
3
m /
MAX(LP)-IO
*OVERLAP
(nonhigh V, LP)
*INHIBIT
a. [nom-nor-um]
**!
b. [nom-nar-ɯm]
* *
c. [nam-nar-ɯm]
*!
Again, candidates (a) [nom-nor-um] and (b) [nom-nar-ɯm] both incur violations of
*OVERLAP(nonhigh vowel, lip protrusion) due to the preservation of underlying nonhigh round
/o/ in their respective initial syllables. Candidate (c) [nam-nar-ɯm], in which all vowels surface
as unround, eliminates this violation of *OVERLAP(nonhigh V, lip protrusion) by deleting the
harmonizing lip protrusion gesture entirely, fatally violating MAX(lip protrusion)-IO. Candidate
(a) [nom-nor-um] incurs an additional *OVERLAP(nonhigh V, lip protrusion) violation due to the
overlap of the nonhigh vowel in the second syllable by the persistent lip protrusion gesture in the
first syllable. This additional violation of the *OVERLAP constraint is fatal to candidate (a). As a
Tongue Body
pharyngeal wide
1
Tongue Body
pharyngeal wide
2
Tongue Body
uvular narrow
3
Lip
protrusion
1
270
result, candidate (b), [nom-nar-ɯm] in which this overlap is prevented by an inhibition relation
between the second vowel gesture and the lip protrusion gesture, emerges as the winner.
This brief sketch of Tuvan rounding harmony has demonstrated that *OVERLAP(nonhigh
vowel, lip protrusion) is able to account for the same general pattern as Kaun’s *ROLO. Many
rounding harmony systems, particularly among the Turkic languages, do not target nonhigh
vowels, despite the fact that nonhigh round vowels may exist underlyingly in the language and
may even trigger rounding harmony. The constraint ranking in (178) correctly generates the
rounding harmony pattern for Tuvan and other languages in which underlying nonhigh round
vowels are preserved, but are not created by rounding harmony.
4.5.5 Yakut: Blocking of Cross-Height Harmony
Rounding harmony in Yakut (also known as Sakha/Saxa; Turkic; Sakha Republic,
Russia; Krueger (1962), Kaun (1995), G. Anderson (1998), Sasa (2001), Walker (2017b)) has
received attention as a harmony system that appears to exhibit local, iterative patterns of
harmony triggering and propagation based on vowel height. Walker (2017b) points out that this
places Yakut rounding harmony in contrast with Baiyina Oroqen rounding harmony (sections
3.4.2 and 4.5.3), which provides evidence for one-to-many, potentially non-local trigger-target
relations. This section demonstrates that the Gestural Harmony Model is able to account both for
harmony systems that exhibit non-local trigger-target relations, such as Baiyina Oroqen, and
those that exhibit apparent local iterativity, such as Yakut.
Yakut has a vowel inventory that is almost perfectly symmetrical for height, backness,
and rounding, and length, save for the absence of a long version of the nonhigh front round
vowel /ø/. The inventory reported by Krueger (1962) and G. Anderson (1998) is provided in
(181).
271
(181) Yakut vowel inventory
Front Back
Unround Round Unround Round
High i iː y yː ɯ ɯː u uː
Nonhigh e eː ø a aː o oː
Diphthongs ie yø ɯa uo
Yakut has full backness harmony that holds throughout roots and suffixes. Words are
either composed of all front vowels (182a-c) or all back vowels (182d-f). All data are from
Krueger (1962) and G. Anderson (1998).
(182) a. [kinige-ler] ‘book (pl.)’ d. [aɣa-lar] ‘father (pl.)’
b. [et-ter] ‘meat (pl.)’ e. [balɯk-lar] ‘fish (pl.)’
c. [øj-y] ‘reason (acc.)’ f. [oɣo-nu] ‘child (acc.)’
The focus of this section is on rounding harmony in Yakut, which shows interesting
asymmetries based on the heights of both triggers and targets. As in other Turkic languages,
round vowels must either be in an initial syllable, or be the products of rounding harmony. A
high vowel undergoes rounding harmony when it follows either a high or nonhigh round vowel,
as in (183).
(183) a. [oχ-u] ‘arrow (acc.)’
b. [børø-ny] ‘wolf (acc.)’
c. [murum-u] ‘nose (acc.)’
d. [tynnyk-y] ‘window (acc.)’
Harmony also proceeds from a nonhigh vowel to another nonhigh vowel, as in (184).
(184) a. [son-ton] ‘jacket (abl.)’
b. [oɣo-lor] ‘child (pl.)’
c. [øj-tøn] ‘reason (abl.)’
d. [børø-lør] ‘wolf (pl.)’
However, when a nonhigh vowel follows a high round vowel, harmony is blocked, as in
(185). This blocking occurs even if the original trigger of harmony (the round vowel in the initial
syllable) is nonhigh, as in (185c-d).
272
(185) a. [kus-tar] ‘duck (pl.)’ c. [tobuk-ka] ‘knee (dat.)’
b. [tynnyk-ler] ‘window (pl.)’ d. [ørys-ter] ‘river (pl.)’
Unlike in Tuvan rounding harmony (section 4.5.4), rounding harmony in Yakut can
derive nonhigh round vowels; however, this is only possible when that nonhigh round vowel is
not preceded by a high round vowel. There have been various analyses proposed in order to
account for this pattern of blocking and non-propagation in Yakut rounding harmony. Kaun
(1995) analyses the blocking of rounding harmony in a high-nonhigh sequence as the result of a
restriction on cross-height rounding harmony, enforced by the constraint UNIFORM[round]. It is
defined as in (186).
(186) UNIFORM[round]: The autosegment [+round] may not be multiply linked to slots bearing
distinct feature specifications. (Kaun 1995, p. 142)
In Yakut, UNIFORM[round] is proposed to prevent rounding from spreading from a high
vowel to a nonhigh vowel, but to be violated in order to spread rounding from a nonhigh vowel
to a high vowel. Kaun analyses this as the effect of the special harmony driving constraint
EXTEND(round)IF[-high] being ranked above UNIFORM[round], motivating its violation only
when a trigger is a nonhigh vowel. Meanwhile, the general harmony driving constraint
EXTEND(round) is ranked below UNIFORM[round], resulting in the blocking of rounding harmony
triggered by a high vowel.
However, such an analysis cannot be directly translated into a gestural analysis of the
pattern of rounding harmony in Yakut. The issue lies in the differences in how harmony is driven
by a maximal harmony-driving constraint such as EXTEND(F) (Kaun 1995) or SPREAD(F)
(Padgett 1995; Walker 1998/2000) in featural phonology versus PERSIST(Gest
X
) and
ANTICIPATE(Gest
X
) in the Gestural Harmony Model. EXTEND(F) and SPREAD(F) are only
satisfied when additional segments in a word are associated to a harmonizing feature. Because of
273
this, high-ranked EXTEND(F) and SPREAD(F) can be used to drive harmony in violation of
UNIFORM[round]. In contrast, as discussed in section 3.2, the Gestural Harmony Model’s
PERSIST(Gest
X
) and ANTICIPATE(Gest
X
) are satisfied if a gesture is specified for a parameter
setting that renders it a trigger of harmony, whether or not it actually extends to overlap any
other gestures. As a result, constraints calling for gestures to surface as persistent or anticipatory
cannot be used to drive violation of UNIFORM[round]. Instead, such violation is motivated by
*INHIBIT. Because of this, a rounding harmony system may either permit cross-height harmony
(if *INHIBIT outranks UNIFORM[round]) or ban cross-height harmony (if UNIFORM[round]
outranks *INHIBIT), regardless of the height of the harmony trigger.
This inability to account for asymmetric cross-height harmony restrictions using
UNIFORM[round] in the Gestural Harmony Model is demonstrated below. The constraint
PERSIST(lip protrusion) is ranked above SELFDEACTIVATE and IDENT(deactivation)-IO in order to
capture the fact that all round vowels in Yakut are attested as triggers of rounding harmony. To
account for the restriction on cross-height harmony, a gestural version of UNIFORM[round] is
included and is ranked above *INHIBIT. It is implemented as a *OVERLAP constraint and defined
as in (187):
(187) *OVERLAP(lip protrusion; high V, nonhigh V): Assign a violation mark to a lip protrusion
gesture that is active concurrently with both a high vowel gesture and a nonhigh vowel
gesture. (abbreviated GESTUNI(lip protrusion))
In the Gestural Harmony Model, it is not necessary to rank a markedness constraint
responsible for blocking over PERSIST. The ranking of GESTUNI(lip protrusion) over *INHIBIT
ensures that cross-height harmony will be banned, even if PERSIST(lip protrusion) is ranked high.
This is because PERSIST(lip protrusion) is not a maximal harmony driver; it does not explicitly
require spreading. Instead, it is satisfied when a gesture of a certain type surfaces as persistent,
274
whether or not that triggering gesture actually extends its period of activation. This is
demonstrated in the tableau in (188) for the Yakut form [kus-tar] ‘duck (pl.).’
(188) Tableau for [kus-tar] ‘duck (pl.)’
Input: /k u
1
s – t a
2
r /
PERSIST(lip protrusion)
*GESTUNI(lip protrusion)
*INHIBIT
a. [kus-tar]
*!
b. [kus-tar]
*
c. [kus-tor]
*!
Candidate (a) [kus-tar] is disharmonic due to the presence of a self-deactivating lip
protrusion gesture, violating PERSIST(lip protrusion). The winning candidate (b), which also
surfaces as [kus-tar], is disharmonic due to the blocking of the persistent lip protrusion gesture of
[u] by the nonhigh vowel gesture of the following [a]. This blocking via gestural inhibition
Tongue Body
uvular narrow
1
Tongue Body
pharyngeal wide
2
Lip
protrusion
1
275
prevents a violation of *GESTUNI(lip protrusion) at the expense of a violation of low-ranked
*INHIBIT. However, even though the lip protrusion gesture has not extended beyond the span of
time it would be active if it were self-deactivating, it does not violate PERSIST(lip protrusion).
This constraint is satisfied even when a persistent gesture is immediately blocked and does not
extend its period of activation. The fully harmonic candidate (c) [kus-tor] contains a persistent
lip protrusion gesture that overlaps both a high and nonhigh vowel gesture, resulting in a fatal
violation of *GESTUNI(lip protrusion).
In (188), cross-height harmony is prevented even though the constraint that enforces this
prevention, GESTUNI(lip protrusion), is outranked by the constraint that is responsible for
harmony, PERSIST(lip protrusion). This demonstrates that a constraint that requires a gesture to
trigger harmony cannot motivate the violation of a cross-height harmony restriction in the same
way that a constraint like SPREAD(F) or EXTEND(F) can. Because of this, it is not possible to
replicate Kaun’s analysis of Yakut’s asymmetric cross-height harmony ban using a special
harmony driving constraint. The addition of a high-ranked constraint that specifically requires
nonhigh vowels to trigger rounding harmony (such as *COUPLE(nonhigh vowel, self-deactivating
lip protrusion), introduced in the analysis of Baiyina Oroqen in section 3.4.2) does not generate
[o-u] or [ø-y] vowel sequences. This is demonstrated in the tableau for the Yakut word [oχ-u]
‘arrow (acc.)’ in (189).
276
(189) Tableau for [oχ-u] ‘arrow (acc.),’ with incorrect candidate selected
Input: / o
1
χ – ɯ
2
/
*COUPLE
(nonhigh V, self-deact. LP)
PERSIST(lip protrusion)
GESTUNI(lip prorusion)
*INHIBIT
a. [oχ-ɯ]
*! *
M b. [oχ-ɯ]
*
c. [oχ-u]
*!
The violation profile for the tableau in (189) is nearly identical to that of (188); the only
difference is seen in candidate (a) [oχ-ɯ], in which a nonhigh round vowel is accompanied by a
self-deactivating lip protrusion gesture, violating both *COUPLE and PERSIST(lip protrusion). The
winner is again candidate (b), [oχ-ɯ], in which harmony is blocked in order to avoid a violation
of GESTUNI(lip protrusion). The bomb icon next to this candidate indicates the erroneous
selection of this candidate as the winner.
Tongue Body
pharyngeal wide
1
Tongue Body
uvular narrow
2
Lip
protrusion
1
277
The tableau in (189) demonstrates that the addition of a special constraint that requires
nonhigh vowels to trigger harmony cannot motivate the violation of the ban on cross-height
harmony in Yakut. Because *COUPLE and PERSIST(lip protrusion) are not satisfied when
segments are overlapped by a harmonizing gesture but rather when a gesture is specified as
persistent (non-self-deactivating), these constraints cannot be used to overrule the effects of
markedness constraints that prevent harmony, such as UNIFORM[round] or its gestural
counterpart, *GESTUNI(lip protrusion).
A solution to this problem comes from a proposal by Sasa (2001, 2009), who argues that
UNIFORM[round] should be split into two constraints: *H-L[round], which penalizes a high-
nonhigh vowel sequence associated to the same [round] feature, and *L-H[round], which
penalizes a nonhigh-high vowel sequence associated to the same [round] feature. By splitting this
constraint, he is able to analyze the Yakut pattern as the result of the high ranking of
*H-L[round] and the low ranking of its counterpart, *L-H[round]. This approach eliminates the
need to use the constraint EXTEND(round)IF[-high] to counteract the effect of UNIFORM[round]
only in cases in which a nonhigh vowel precedes a high vowel. This is precisely what is needed
in order for the analysis within the Gestural Harmony Model to be successful, as there is no way
to replicate the effect of EXTEND(round)IF[-high] on patterns of blocking.
In the gestural analysis of Yakut rounding harmony, the ranking of the constraint
*H-L[round] above *INHIBIT will generate the correct pattern of asymmetric cross-height
harmony restriction seen in Yakut. A sequence like [u-o] is still penalized, but [o-u] sequences
are not. A gestural version of *H-L[round] is generated by modifying the *OVERLAP constraint
in (187) slightly to specify the sequencing of the two vowel gestures that are prohibited from
being overlapped by a lip protrusion gesture. This new constraint is defined in (190).
278
(190) *OVERLAP(lip protrusion; high V, nonhigh V): Assign a violation mark to a lip protrusion
gesture that is active concurrently with a high vowel gesture followed by a nonhigh
vowel gesture. (abbreviated *H-L[round])
With the gestural version of *H-L[round] ranked above *INHIBIT, [u-o] sequences will
still be penalized as they were by GESTUNI(lip protrusion) (see the tableau in (188)), but [o-u]
sequences will not incur violations of this constraint. A form such as [oχ-u] ‘arrow (acc.)’ can
now be correctly selected as the winning candidate. This is demonstrated in the tableau in (191),
in which both output candidates contain persistent lip protrusion gestures.
(191) Tableau for Yakut [oχ-u] ‘arrow (acc.),’ with correct candidate selected
Input: / o
1
χ – ɯ
2
/
*H-L[round]
*INHIBIT
a. [oχ-ɯ]
*
b. [oχ-u]
Candidate (a) [oχ-ɯ] includes an inhibition relation between the high back vowel and the
lip protrusion gesture of the [o]. This incurs a violation of *INHIBIT; however, blocking of
harmony in this candidate does nothing to improve its performance with respect to *H-L[round],
Tongue Body
pharyngeal wide
1
Tongue Body
uvular narrow
2
Lip
protrusion
1
279
as neither candidate violates this constraint. The winner in this tableau is candidate (b) [oχ-u],
which incurs no violations. *H-L[round] is satisfied because the lip protrusion gesture in this
candidate overlaps a nonhigh-high sequence and not a high-nonhigh sequence.
Sasa (2001, 2009) also argues that splitting UNIFORM[round] into two constraints is
necessary in order to accurately account for the pattern of rounding harmony in Yakut words
with three or more syllables. As illustrated by the data in (185) above, high vowels undergo
vowel harmony, but do not allow it to spread further. For instance, the Yakut word for ‘knee
(dat.)’ is [tobuk-ka], not *[tobuk-ko]. It is not the quality of the originator of harmony (i.e., the
vowel in the initial syllable) that determines whether a given segment undergoes harmony, but
the quality of the vowel immediately adjacent to that potential undergoer. This can be contrasted
with Baiyina Oroqen rounding harmony (sections 3.4.2 and 4.5.3), in which long nonhigh round
vowels do not trigger rounding harmony but do undergo and propagate it.
Sasa points out that Kaun’s analysis, which uses the maximal harmony-driving constraint
EXTEND(round)IF[-high] ranked above UNIFORM[round], actually predicts that all vowels should
undergo rounding harmony that is triggered by a nonhigh vowel in an initial syllable, even if a
high vowel intervenes. The fact that [u] does not trigger rounding harmony in Yakut does not
affect its ability to propagate rounding onto a following nonhigh vowel, because in a vowel
sequence such as [o
1
-u
2
-o
3
] the initial [o
1
] is the trigger, not [u
2
]. Walker (2014) shows that this
casting of initial [o] as a global trigger of harmony is a property of maximal harmony drivers that
require all segments in a harmony domain to associate to a harmonizing feature, even if they are
not adjacent to the trigger. Kaun’s constraint EXTEND(round)IF[-high] is such a global harmony
driver, and therefore it predicts harmony patterns that are characteristic of nonlocal trigger-target
relations as defined by Walker.
280
However, adoption of the constraint *H-L[round] in place of UNIFORM[round] and the
elimination of EXTEND(round)IF[-high] from the analysis of Yakut solves this issue. The gestural
analysis of Yakut rounding harmony is able to generate forms in which harmony is triggered by a
nonhigh vowel but arrested by an undergoing high vowel. The tableau in (192) for the form
Yakut [tobuk-ka] ‘knee (dat.)’ demonstrates.
(192) Tableau for Yakut [tobuk-ka] ‘knee (dat.)’
Input: /t o
1
b ɯ
2
k – k a
3
/
*H-L[round]
*INHIBIT
a. [tobuk-ka]
*
b. [tobuk-ko]
*!
In winning candidate (a) [tobuk-ka], the lip protrusion gesture that accompanies the [o] of
the initial syllable is inhibited by nonhigh [a] in order to avoid violation of higher-ranked
*H-L[round]. In candidate (b) [tobuk-ko], the lip protrusion gesture is allowed to extend
throughout the entire word, resulting in an illicit [u-o] sequence in violation of *H-L[round].
Note that both candidates contain the cross-height round vowel sequence [o-u], which is not
Tongue Body
pharyngeal wide
1
Tongue Body
uvular narrow
2
Tongue Body
pharyngeal wide
3
Lip
protrusion
1
281
penalized. The Gestural Harmony Model has thus proven itself able to account both for patterns
of apparent local triggering, as in Yakut, and for patterns of nonlocal trigger-target relations, as
in Baiyina Oroqen, provided the proper constraints on cross-height harmony are adopted.
Walker (2017b) argues against the split of UNIFORM[round] into the two constraints
*H-L[round] and *L-H[round] on the grounds that the freely rankable *L-H[round] proposed by
Sasa (2001, 2009) predicts patterns of rounding harmony that are unattested according to Kaun’s
(1995, 2004) typological surveys. Within these typologies, many rounding harmony systems
allow nonhigh vowels to spread rounding onto high vowels. However, few languages allow high
vowels to spread rounding onto nonhigh vowels, and they only do so if nonhigh-to-high
spreading of rounding is also permitted. This implicational relationship between types of cross-
height harmony is lost if *L-H[round] is admitted into the constraint set. Walker instead
proposes an analysis of Yakut rounding harmony that is situated within the framework of
Harmonic Grammar (Legendre, Miyata, & Smolensky 1990; Smolensky & Legendre 2006) and
relies on cumulative constraint interaction between UNIFORM[round] and *ROLO, which
penalizes nonhigh round vowels.
However, within the Gestural Harmony Model it is impossible to generate the asymmetry
in cross-height harmony seen in Yakut using a single UNIFORM[round] constraint for the reasons
stated earlier in this section. It is necessary to use Sasa’s *H-L[round] constraint in order to
capture the Yakut pattern within the Gestural Harmony Model. His other constraint,
*L-H[round], on the other hand, appears to be unnecessary, in addition to being the potential
source of a number of pathological predictions about rounding harmony. One option for
eliminating pathological *L-H[round] is to claim that there are two constraints on cross-height
harmony that are in a specific-general relationship with one another. While specific *H-L[round]
282
penalizes only cross-height rounding harmony that involves a high vowel followed by a non-high
vowel, general UNIFORM[round] and its gestural counterpart GESTUNI(lip protrusion) penalize all
cross-height rounding harmony, as attested in Yowlumne Yokuts. The special-general
relationship between these constraints will capture Kaun’s typological observation that if a
language prohibits nonhigh-to-high cross-height rounding harmony, it also prohibits high-to-
nonhigh harmony.
A possible additional reason for splitting UNIFORM[round] into two constraints comes
from vowel harmony in Arapaho (Algonquian; Midwestern United States).
32
Cowell & Moss
(2008) report that Arapaho has two vowel harmony processes that affect backness and rounding,
one progressive and one regressive. Progressive harmony is triggered by /o/ and causes a
following /i/ to surface as [u], as in [hoːw-u-seː] ‘to walk downward’ (cf. [ceb-i-seː] ‘to walk
(along)’). Crucially, this harmony process results in the vowel sequence [o-u]. Regressive
harmony is also triggered by /o/; however, it only targets nonhigh /e/ rather than high /i/. This
process produces the vowel sequence [o-o], as in [boː’-oːwu] ‘the water flows red’ (cf. [beː’-eː]
‘it is red’), but apparently avoids producing the sequence [u-o]. In Yakut, the avoidance of the
sequence [u-o] is analyzed by Walker (2017b) to be the result of an avoidance of cross-height
harmony when it would also derive a marked nonhigh round vowel. However, vowel harmony in
Arapaho exhibits the same avoidance of the sequence [u-o] despite it being the [u] that is derived
by vowel harmony, and not the [o]. Cumulative interaction of UNIFORM[round] and *ROLO does
not account for this pattern of vowel co-occurrence in Arapaho, while *H-L[round] does.
This study of Yakut rounding harmony has demonstrated that the Gestural Harmony
Model is able to account for a case of apparent local iterative harmony, provided the constraint
on cross-height harmony is properly defined. This is an important point in favor of the Gestural
32
Thanks to Ksenia Bogomolets for bringing this phenomenon to my attention.
283
Harmony Model, as it demonstrates that it is capable of accounting for patterns that appear to be
cases of locally triggered, iterative harmony in the case of Yakut, as well as global, one-to-many
triggering in the case of Baiyina Oroqen (sections 3.4.2 and 4.5.3). Many other analyses of
harmony are capable of generating one or the other, but not both. A notable exception is
Walker’s (2017b) model of harmony in Harmonic Grammar, though it remains to be seen
whether that approach is robust across different sets of markedness constraints that are active in
harmony systems.
4.5.6 Summary
This section has illustrated the Gestural Harmony Model’s ability to account for patterns
of transparency and blocking in rounding harmony, particularly with respect to the typological
asymmetry observed between the sets of attested transparent and blocking segments. While all
vowels are attested as blockers of rounding harmony in some system, the ability to surface as
transparent is limited only to high front vowels. The Gestural Harmony Model predicts this
asymmetry, and analyzes it as the result of the representation of transparent segments as a special
type of undergoers of harmony. The special status of high front vowels comes from their
inclusion of a gesture for active lip spreading, which is antagonistic to the lip protrusion gesture
responsible for rounding harmony.
In addition to accurately predicting the asymmetry between attested transparent and
blocking segments, the Gestural Harmony Model is also able to account for harmony systems
that exhibit both transparency and blocking. This is exemplified by Halh Mongolian in section
4.5.2. By splitting the sources of transparency and blocking among two distinct theoretical
mechanisms, the Gestural Harmony Model correctly predicts that the effects of both of these
284
mechanisms, transparency via competition between antagonistic gestures and blocking via
inhibition between incompatible gestures, should arise in some languages.
Across rounding harmony systems, there are many sources of gestural incompatibility
that drive patterns of blocking that can be quite complex. The Gestural Harmony Model has
proven itself capable of accounting for them, given the right constraint set. A particular asset to
the theory is its ability to account for patterns of blocking that have been described as involving a
local, iterative interpretation of harmony triggering (e.g. Yakut, section 4.5.5) as well as those
with a global interpretation of triggering (e.g. Baiyina Oroqen, sections 3.4.2 and 4.5.3).
4.6 Transparency and Blocking in Tongue Root Harmony
This section focuses on ATR and RTR harmonies and their attested patterns of
transparency and blocking. While nasal harmony and rounding harmony show asymmetries in
the sets of attested transparent and blocking segments, no such asymmetry exists across ATR and
RTR systems. The Gestural Harmony Model predicts this based on differences in the sources of
gestural antagonism and incompatibility across these three types of harmony. How this
prediction is borne out with respect to tongue root harmony is the subject of this section, which
examines cases of RTR harmony from three dialects of Yoruba.
4.6.1 Sources of Antagonism and Incompatibility
As discussed in section 4.2.3, there is no asymmetry in the crosslinguistically attested sets
of transparent and blocking segments in tongue root harmony. In ATR harmony, the low vowel
/a/ is the only segment that appears not to participate in some harmony processes; it is
transparent in some languages, and a blocker in others. In RTR harmony, the high vowels /i/ and
/u/ are attested as both transparent and blocking segments. Despite the distinct mechanisms of
neutrality responsible for transparency and blocking in the Gestural Harmony Model, the lack of
285
an asymmetry within the sets of attested transparent and blocking segments is predicted for
tongue root harmony. If the set of segments that are antagonistic to a harmonizing gesture is
identical to the set of segments that are incompatible with that harmonizing gesture, no
asymmetry between attested transparent and blocking segments is predicted. This is the case in
ATR and RTR harmonies.
Archangeli & Pulleyblank (1994) provide a detailed account of restrictions on the co-
occurrence of tongue root advancement and retraction with vowels of different heights. They
propose a number of phonetically grounded conditions on the co-occurrence of vowel height and
tongue root position, which, when active in a language, affect both phonological inventories and
processes of ATR and RTR harmony. These conditions are stated in (193).
(193) Conditions on vowel height and tongue root position (Archangeli & Pulleyblank 1994,
pp. 174, 176)
33
a. ATR/HI Condition: If [ATR] then [+high] / If [ATR] then not [-high]
b. ATR/LO Condition: If [ATR] then [-low] / If [ATR] then not [+low]
c. RTR/HI Condition: If [RTR] then [-high] / If [RTR] then not [+high]
d. RTR/LO Condition: If [RTR] then [+low] / If [RTR] then not [-low]
e. HI/ATR Condition: If [+high] then [ATR] / If [+high] then not [RTR]
f. LO/RTR Condition: If [+low] then [RTR] / If [+low] then not [ATR]
All of these conditions that render different combinations of tongue root gestures and
vocalic tongue body gestures marked are based upon articulatory incompatibility. In many
languages, high vowels require tongue root advancement in order to successfully shift the mass
of the tongue body upward; a non-advanced tongue root, whether due to the tongue root being in
a neutral position or being actively retracted, interferes with raising of the tongue body.
Similarly, low vowels, which are often back vowels, require tongue root retraction in order to
successfully achieve the necessary degree of retraction of the tongue body. A non-retracted
33
Archangeli & Pulleyblank refer to RTR vowels as [-ATR], and to ATR vowels as [+ATR].
286
tongue root interferes with tongue body retraction for /a/. As a result, high ATR vowels in RTR
harmony languages often have no RTR counterparts, while low RTR vowels in ATR harmony
languages often have no ATR counterparts. The involvement of tongue root position in the
production of vowels of different heights is illustrated in the figure in (194).
(194) High vowel with advanced tongue root, low vowel with retracted tongue root
Because the incompatibility of vowel height and tongue root position is based in
articulation, the sets of possible transparent and blocking vowels in ATR harmony systems
overlap entirely. This is different from the cases of nasal harmony and rounding harmony, in
which gestural incompatibility is rooted in both articulatory and perceptual difficulty. As a result,
the Gestural Harmony Model makes different predictions for nasal and rounding harmonies
versus tongue root harmonies with respect to which segments can be transparent to harmony and
which can block it. Because gestural incompatibility for nasality and rounding is not solely based
on articulatory factors, asymmetries in possible transparent and blocking segments are predicted
for these types of harmony. However, no such asymmetry is predicted for tongue root harmony.
The effects of Archangeli & Pulleyblank’s conditions on vowel height and tongue root
position can be replicated in the Gestural Harmony Model with constraints from the *OVERLAP
family. The *OVERLAP constraints will penalize a high vowel that is concurrently active with a
tongue root retraction gesture (195) and a low vowel that is concurrently active with a tongue
287
root advancement gesture (196). These constraints will prevent such vowels from arising in a
language’s surface phonological inventory, and will prevent them from being created via tongue
root harmony. For the purposes of defining these constraints, low vowels refer to vocalic
segments including a tongue body gesture with constriction in the pharyngeal region.
(195) *OVERLAP(high vowel, tongue root retraction): Assign a violation mark to a high vowel
gesture and a tongue root retraction gesture that are concurrently active.
(196) *OVERLAP(low vowel, tongue root advancement): Assign a violation mark to a low
vowel gesture and a tongue root advancement gesture that are concurrently active.
The relative ranking of these constraints along with *INHIBIT and constraints on
faithfulness determine languages’ surface phonological inventories, as well as the patterns of
transparency and blocking in tongue root harmony systems. As in the analyses of nasal harmony
and rounding harmony presented in sections 4.4 and 4.5, blocking of harmony results when a
*OVERLAP constraint outranks *INHIBIT. Blocking of ATR harmony by low vowels is the result
of *OVERLAP(low vowel, ATR) ranked above *INHIBIT. Likewise, blocking of RTR harmony by
high vowels is the result of *OVERLAP(high vowel, RTR) ranked above *INHIBIT. Mid vowels,
which are subject to neither *OVERLAP constraint, are correctly predicted not to block tongue
root harmony.
When *INHIBIT is ranked above *OVERLAP, blocking of harmony is prohibited, and a
harmonizing gesture will be allowed to extend to overlap vowels with which it is incompatible.
There are two possible outcomes of this overlap. When a harmonizing tongue root gesture is
active concurrently with a vowel that includes a gesture that is antagonistic to it, that vowel will
surface as transparent to harmony. For instance, a low vowel gesture may be accompanied by a
tongue root retraction gesture. Due to the ranking of *INHIBIT over *OVERLAP(low vowel, ATR),
a tongue root advancement gesture will be allowed to overlap the low RTR vowel, resulting in
288
the concurrent activation of a tongue root advancement gesture and a tongue root retraction
gesture. The result of this configuration is transparency via intergestural competition and
blending, with the tongue root retraction gesture responsible for transparency overpowering the
harmonizing tongue root advancement gesture when it is specified for greater blending strength.
Similarly, the presence of a tongue root advancement gesture accompanying a high vowel will
induce high vowel transparency to RTR harmony. A high vowel whose representation does not
include a tongue root advancement gesture will surface as RTR when overlapped by a
harmonizing tongue root retraction gesture.
This typological sketch of tongue root harmony, along with the in-depth examinations of
various nasal and rounding harmonies already reported in sections 4.4 and 4.5, demonstrates that
it is possible for the Gestural Harmony Model to capture crosslinguistic generalizations about
transparency and blocking in various types of harmony, whether they show asymmetries in the
sets of transparent and blocking segments or not. Typological asymmetries in attested transparent
and blocking segments are accurately reflected in the Gestural Harmony Model’s distinct
mechanisms of transparency, based on concurrent activation of antagonistic gestures, and
blocking, based on intergestural inhibition motivated by gestural incompatibility. However, the
prediction of an asymmetry disappears when the class of segments that include antagonistic
gestures is not a proper subset of the class of segments that are incompatible with a harmonizing
gesture, as is the case in tongue root harmony. These strong typological predictions about the
types of segments that can be transparent to harmony versus those that can block it are among
the greatest strengths of the Gestural Harmony Model.
The typological patterns predicted by the interaction of the constraint *OVERLAP(high
vowel, RTR), *OVERLAP(low vowel, ATR), and *INHIBIT, as well as the gestural makeup of the
289
vowels in a language’s phonological inventory, are summarized in the following tables. The
table in (197) contains the constraint rankings and resulting patterns of transparency and
blocking for harmony based on tongue root advancement.
(197) Predicted typology of transparency and blocking in ATR harmony
Ranking Result Attested in
*OVERLAP(low V, ATR) >> *INHIBIT Low vowels block ATR
harmony.
Akan (Clements
1981)
*INHIBIT >> *OVERLAP(low V, ATR) All vowels undergo ATR
harmony.
Low vowels surface as
ATR.
Nandi (Creider &
Creider 1989)
Low vowels surface as
transparent.
Budu (Kutsch
Lojenga 1994)
A parallel typology emerges from the interaction of constraints on RTR harmony. This
typology is provided in (198).
(198) Predicted typology of transparency and blocking in RTR harmony
Ranking Result Attested in
*OVERLAP(high V, RTR) >> *INHIBIT High vowels block RTR
harmony.
Standard Yoruba
(Pulleyblank 1988,
1996; Archangeli &
Pulleyblank 1989; Ọla
Orie 2003)
*INHIBIT >> *OVERLAP(high V, RTR) All vowels undergo RTR
harmony.
High vowels surface as
RTR.
Ekiti Yoruba (Ọla
Orie 2003)
High vowels surface as
transparent.
Ife Yoruba (Ọla Orie
2001, 2003)
290
As in nasal harmony and rounding harmony, the constraints shape the typology by
determining which types of segments will be overlapped by a harmonizing gesture, and which
will block harmony by inhibiting a harmonizing gesture. From there, the representation of an
overlapped segment (i.e., whether it includes a high-strength gesture that is antagonistic to the
harmonizing gesture) will determine whether it surfaces with the harmonizing property, or
whether it is transparent to harmony. To illustrate this, the remainder of this section analyses
several cases of RTR harmony in different varieties of Yoruba. All varieties of Yoruba display
some kind of RTR harmony; however, there is variation between them with respect to whether
high vowels block harmony, surface as transparent to it, or surface as RTR. Interestingly, these
varieties of Yoruba cover all of the possible patterns of transparency, blocking, and undergoing
of RTR harmony listed in the table in (198).
4.6.2 Yoruba: Transparency and Blocking by High Vowels
The patterns of transparency and blocking in tongue root harmony are demonstrated here
with examples from three varieties of Yoruba. In Standard Yoruba, high vowels block regressive
RTR harmony. In Ife Yoruba, the high vowels are transparent to it. In Ekiti Yoruba, high vowels
surface as RTR in the domain of RTR harmony. The data and its interpretation come from
multiple sources, including Pulleyblank (1988, 1996), Archangeli & Pulleyblank (1989, 1994),
and Ọla Orie (2001, 2003).
Looking first at Standard and Ife Yoruba, the oral vowel inventory of both varieties is
provided in (199).
291
(199) Standard Yoruba and Ife Yoruba oral vowel inventory
34
non-RTR RTR
Front Back Front Back
High i u
Mid e o ɛ ɔ
Low ɑ
While the mid vowels in these dialects have both RTR and non-RTR counterparts, the
high vowels /i/ and /u/ surface only as non-RTR and the low back vowel /ɑ/ surfaces only as
RTR.
The RTR vowels /ɛ/, /ɔ/, and /ɑ/ all trigger regressive RTR harmony in both Standard
Yoruba and Ife Yoruba. The data in (200) illustrate.
(200) RTR harmony in Standard Yoruba and Ife Yoruba
a. [ɛkɛ] ‘forked stick’ *[ekɛ]
b. [ɔbɛ] ‘soup *[obɛ]
c. [ɔkɔ] ‘vehicle’ *[okɔ]
d. [ɛkɔ] ‘pap’ *[ekɔ]
e. [ɛpa] ‘groundnut’ *[epa]
f. [ɔja] ‘market’ *[oja]
In Standard Yoruba, the regressive spread of RTR is blocked when the high vowels /i/
and /u/ precede the triggering word-final RTR vowel, as in the data in (201).
(201) Blocking by high vowels in Standard Yoruba
a. [ewurɛ] ‘goat’
b. [elubɔ] ‘yam flour’
c. [otitɔ] ‘truth’
d. [odidɛ] ‘parrot’
e. [okurɛ] ‘palm kernel’
f. [oriʃa] ‘primordial diety’
g. [oʃupa] ‘moon’
34
Yoruba also has several nasal vowels; however, this is not crucial to the analysis in this section and is therefore set
aside.
292
However, in Ife Yoruba these high vowels are transparent to RTR harmony, as in the data
in (202).
(202) Transparency of high vowels in Ife Yoruba
a. [ɛwurɛ] ‘goat’
b. [ɛlubɔ] ‘yam flour’
c. [ɔtitɔ] ‘truth’
d. [ɔdidɛ] ‘parrot’
e. [ɔrisa] ‘primordial diety’
f. [ɔsupa] ‘moon’
I analyze regressive RTR harmony in Yoruba as the result of the high ranking of the
constraint ANTICIPATE(tongue root retraction), which requires a tongue root retraction gesture in
an output form to be anticipatory (early-activating). However, in the analysis of blocking that
follows I will set this and other constraints on harmony triggering aside and will only compare
candidates in which an anticipatory tongue root retraction gesture surfaces in the final syllable of
a word. The blocking of RTR harmony by high vowels in Standard Yoruba can be analyzed as
the result of the constraint *OVERLAP(high V, RTR) ranked above *INHIBIT. The effect of this
ranking is demonstrated in the tableau in (203) for the Standard Yoruba form [odidɛ] ‘parrot.’
Again, I only include the vowel gestures of candidate output forms, which are displayed in
gestural score form.
293
(203) Tableau for Standard Yoruba [odidɛ] ‘parrot’
Input: / o
1
d i
2
d ɛ
3
/
*OVERLAP(high V, RTR)
*INHIBIT
a. [ɔdɪdɛ]
*!
F b. [odidɛ]
*
In (203), the anticipatory tongue root retraction gesture of [ɛ] in candidate (a) [ɔdɪdɛ] has
extended to overlap all preceding vowels in the word, causing them to surface as RTR. The
overlap of the high vowel gesture by this tongue root retraction gesture fatally violates high-
ranked *OVERLAP(high V, RTR). The winning candidate (b) [odidɛ] avoids this overlap by
including an inhibition relation between the palatal gesture of [i] and the harmonizing tongue
root retraction gesture. While this results in a violation on low-ranked *INHIBIT, it prevents the
tongue root retraction gesture from extending to overlap the high vowel, thus satisfying
*OVERLAP(high V, RTR).
Tongue Body
uvu-phar narrow
1
Tongue Body
palatal narrow
2
Tongue Body
palatal mid
3
Lip
protrusion
1
Tongue Root
retracted
3
294
It should be noted that the representation of the high vowel in the input and outputs of the
tableau in (203) does not include a tongue root advancement gesture. However, there is nothing
in the definition of *OVERLAP(high V, RTR) that requires a blocker to include such a gesture in
its representation; a high vowel gesture will block regardless of whether it possesses a tongue
root advancement gesture. Of course, it may be the case that high vowels in Standard Yoruba are
accompanied by tongue root advancement gestures; there is no evidence here either for or against
this representation.
In contrast, the high vowels of Ife Yoruba do appear to include tongue root advancement
gestures in their representations. According to Ọla Orie (2001), in Ife Yoruba high vowels do not
block RTR harmony. This can be analyzed as the result of a distinct ranking of constraints in this
dialect of the language such that *INHIBIT outranks *OVERLAP(high vowel, RTR). Because of
this ranking, high vowels will not block RTR harmony by inhibiting a harmonizing tongue root
retraction gesture, but will instead be overlapped by it. Because of the presence of tongue root
advancement gestures accompanying these high vowels, however, the result of this overlap will
be transparency to harmony rather than retraction of the tongue root.
This is illustrated by the gestural score in (204) for the Ife Yoruba form [ɔdidɛ] ‘parrot,’
in which word-final [ɛ] triggers RTR harmony by including an anticipatory tongue root retraction
gesture in its representation. This anticipatory gesture overlaps all preceding gestures in the
word. As a result of this overlap, the word-initial mid vowel surfaces as retracted, while the
word-medial [i] includes a relatively high-strength antagonistic tongue root advancement gesture
that results in the transparency of that vowel to RTR harmony. The dashed line indicates the
neutral position of the tongue root.
295
(204) Gestural score for vocalic portion of Ife Yoruba [ɔdidɛ] ‘parrot’
35
Standard and Ife Yoruba can be contrasted with Ekiti Yoruba, in which the high vowels
undergo RTR harmony and surface as [ɪ] and [ʊ], respectively (Ọla Orie 2003). As in Ife
Yoruba, the high vowels do not act as blockers of harmony; *INHIBIT must be ranked high in
these varieties as well. However, unlike in Ife Yoruba, these high vowels do not appear to
include tongue root advancement gestures that are antagonistic to the harmonizing tongue root
retraction gesture. As a result, when a high vowel is overlapped by an anticipatory tongue root
retraction gesture in Ekiti Yoruba, the result is retraction of the tongue root during the production
of these vowels, rather than transparency. This is illustrated by the gestural score in (205) for the
Ekiti Yoruba form [ɔdɪdɛ] ‘parrot,’ a cognate of the Ife Yoruba form in (204) above.
35
Because of the lack of blocking in Ife Yoruba RTR harmony, it is unclear whether this is a case of regressive
harmony from a word-final anticipatory gesture or of progressive harmony from a word-initial persistent gesture.
Because the standard variety of Yoruba does indicate from its patterns of blocking that it is a regressive harmony
process, the tongue root harmony of Ife Yoruba is assumed to also be a regressive harmony process, though this is
not a crucial assumption.
296
(205) Gestural score for vocalic portion of Ekiti Yoruba [ɔdɪdɛ] ‘parrot’
The Standard, Ife, and Ekiti varieties of Yoruba, then, encompass all of the possible ways
in which the high vowels /i/ and /u/ can surface when subject to RTR harmony. Due to their
gestural incompatibility with a retracted position of the tongue root, high vowels may be
compelled by a high-ranked *OVERLAP constraint to inhibit a harmonizing tongue root retraction
gesture, resulting in blocking of RTR harmony. This is exemplified by Standard Yoruba.
Alternatively, this incompatibility may be tolerated by the phonological grammar, in which case
high vowels will undergo harmony. When this occurs, there are two possible outcomes. In Ife
Yoruba, high vowels surface as transparent due to their inclusion of a high-strength tongue root
advancement gesture that is antagonistic to the harmonizing gesture. This will result in
intergestural competition that favors the target articulatory state of the high vowel, resulting in
transparency. In Ekiri Yoruba, on the other hand, high vowels are not accompanied by tongue
root advancement gestures and will surface as RTR when overlapped by a harmonizing tongue
root retraction gesture.
These three outcomes represent all of the possible outcomes predicted by the RTR
typology listed in (198). The same logic that holds here for the possible outcomes of high vowels
in RTR harmony also hold for the converse pattern of the possible outcomes of low vowels in
297
ATR harmony. There are also three ways in which a low vowel is predicted to surface in the
environment of ATR harmony; all are attested, as listed in the table in (197).
4.6.3 Summary
This section has focused on patterns of transparency and blocking in tongue root
harmony, which are considerably simpler than those observed in either nasal harmony (section
4.4) or rounding harmony (section 4.5). Whereas both nasal harmony and rounding harmony
show crosslinguistic asymmetries such that the classes of attested transparent segments are a
subset of the classes of attested blocking segments, in ATR and RTR harmony no such
asymmetry exists. In ATR harmony, low vowels are attested as transparent and blocking
segments. Likewise, in RTR harmony, high vowels are attested as transparent and blocking
segments.
The fact that the Gestural Harmony Model makes no predictions regarding asymmetries
in attested transparent and blocking segments in tongue root harmony is in itself an important
prediction of the model. This model will predict an asymmetry between transparent and blocking
segments precisely where there is an asymmetry in the sets of antagonistic and incompatible
segments. In nasal harmony and rounding harmony, antagonism to a harmonizing gesture arises
from a small set of segment types (obstruents with a velum closure gesture in the case of nasal
harmony, and high front vowels with a lip spreading gesture in the case of rounding harmony).
However, in both of these types of harmony incompatibility with a harmonizing gesture arises
from a larger set of segment types based on both articulatory and perceptual markedness factors.
In contrast, within ATR and RTR harmony the sets of segments that are antagonistic to a
harmonizing tongue root gesture are not proper subsets of the sets of incompatible segments.
298
Therefore, the Gestural Harmony Model is able to accurately predict the existence of
typological asymmetries in attested transparent and blocking segments in the types of harmony
that exhibit such asymmetries (nasal and rounding harmony) while also accurately predicting the
lack of such a typological asymmetry in the types of harmony that do not exhibit these
asymmetries (ATR and RTR harmony). It is in this respect that the Gestural Harmony Model’s
dual mechanisms of transparency via intergestural competition and blending and blocking via
intergestural inhibition show their major strength. Many accounts within featural phonology that
employ a single theoretical mechanism to account for both transparency and blocking do not
match this success in accounting for typological asymmetries between the two. This is discussed
in the following section.
4.7 Comparing Analyses of Transparency and Blocking
The analysis of transparency in harmony as the result of the competition between two
gestures with antagonistic target articulatory states represents a substantial departure from
previous accounts of this phenomenon. This section examines properties of the Gestural
Harmony Model’s representation of transparency and blocking, as well as the typological
predictions made by the model; it also compares these predictions to those made by several
alternative accounts. Earlier work has relied upon derivational opacity or gapped representations
to capture transparency in harmony. More recently, featural co-occurrence constraints have
played a key role in analyses of both transparency and blocking in harmony. Other analyses rely
on a continuum or scale of likelihood of certain segment types to be transparent or to block a
harmony process. This section demonstrates that the Gestural Harmony Model’s representation
of transparency and blocking improves upon many previous analyses in several areas.
299
4.7.1 Locality of Spreading
A long-standing debate within the study of harmony revolves around how (and whether)
to maintain a local representation of a harmonizing element in harmony systems that exhibit
transparency. Some previous proposals rely on derivational opacity in order to capture instances
of transparency while preserving locality of the harmony process. Piggott (1988), for instance,
provides an analysis of Guaraní, which exhibits regressive (leftward) nasal harmony with
transparent voiceless obstruents, in which the feature [+nasal] first spreads onto voiceless
obstruents but is then denasalized by the later default rule [-voice] → [-nasal]. While this
analysis preserves the locality of the harmony process, it does so by relying on a derivationally
opaque rule ordering. As discussed in section 3.6.3, such analyses are incompatible with output-
oriented frameworks such as OT without adopting additional theoretical mechanisms. One such
mechanism is Harmonic Sympathy, a variant of Sympathy Theory (McCarthy 1999) proposed by
Walker (1998/2000, 2003). Walker analyzes transparency in nasal harmony using Harmonic
Sympathy to mimic the effect of derivational opacity within OT. In this analysis, the selected
output form is faithful to another candidate in which nasality has spread fully throughout a word
(including onto any obstruents), but differs from that candidate by not nasalizing those obstruents
in order to satisfy a high-ranked feature co-occurrence constraint.
The Gestural Harmony Model’s account of transparency similarly represents transparent
segments as having undergone harmony; they are overlapped by a persistent and/or anticipatory
gesture. However, such segments’ surfacing as transparent does rely not upon a subsequent step
of repair as Piggott’s (1988) featural account does. Instead, a segment surfaces as transparent to
harmony due to its gestural makeup, i.e. due to its including a gesture that is antagonistic to the
harmonizing gesture that overlaps it. The Gestural Harmony Model also does not require any
elaboration of the constraint evaluation mechanism in OT, as required by Harmonic Sympathy.
300
An alternative method for capturing transparency in harmony involves allowing a feature
to spread beyond a transparent segment without associating with it, resulting in a gapped
autosegmental configuration. In some cases, this skipping is motivated by assuming
subsegmental feature geometric nodes (Clements 1985; Sagey 1986), and identifying the set of
possible targets of a harmony process by which segments possess the proper node to which to
associate a harmonizing feature. This is the case in Piggott’s (1992) analysis of nasal harmony in
Southern Barasano (also known as Barasana-Eduria; Tucanoan; Colombia). Piggott motivates
the skipping of obstruents in nasal harmony by proposing that in languages with transparency to
nasal harmony, the feature [nasal] is a dependent of the Spontaneous Voicing node. When the
feature [nasal] spreads, it will only target segments that bear this node, namely the class of
sonorants, which potentially includes voiced stops. The figure in (206) demonstrates this for the
Southern Barasano form [w
̃ ãtĩ] ‘demon’ provided by Piggott (1992; p. 53). In this figure, the
harmonizing [nasal] feature does not associate with the obstruent [t] as it is lacking a
Spontaneous Voicing node, thereby eliminating it as a possible target of nasal harmony.
(206) Derivation of Southern Barasano [w
̃ ãtĩ] ‘demon’ proposed by Piggott (1992)
→
However, as discussed in section 2.3, work by Gafos (1996/1999), Ní Chiosáin & Padgett
(2001), Walker (1998/2000), and others claims that such gapped autosegmental configurations
are universally ill-formed. According to this view, all spreading must respect strict segmental
301
locality. Skipping of segments in feature spreading is not permitted; therefore, a form such as
that in (206) would be universally ruled out.
The Gestural Harmony Model’s coactivation account of transparency encounters no such
issues concerning gapped representations and locality of spreading. In this model, harmony is the
result of either a persistent (non-self-deactivating) or anticipatory (early-activating) gesture
extending to overlap undergoing segments. Even segments that appear to be skipped over by a
harmony process are overlapped by a harmonizing gesture. However, this continuous activation
of the harmonizing gesture is not perceived during the production of a transparent segment due
to the effects of intergestural antagonism and blending. Because of this, locality of spreading can
be maintained while also maintaining the idea that transparent segments surface without having
taken on a harmonizing property.
4.7.2 Asymmetries in Attested Transparent and Blocking Segments
Many analyses of neutrality in harmony, both rule- and constraint-based, rely on the use
of restrictions on featural co-occurrence to render certain segment types neutral to a harmony
process. The Gestural Harmony Model rejects neutrality as a concept unifying both transparency
and blocking, instead utilizing two distinct mechanisms to account for them. As a result, the
model correctly predicts that the sets of attested transparent and blocking segments in a given
type of harmony may be distinct. However, there are numerous feature-based analyses of
harmony that utilize feature co-occurrence constraints to account for the ‘neutral’ status of both
blockers and transparent segments. Though all of these approaches differ in the details of their
implementation, without invoking additional theoretical assumptions they make the undesirable
prediction that all segments that are attested as blockers for a certain type of harmony should also
be able to be transparent to harmony as well. This runs contrary to the crosslinguistic patterns
302
observed among transparent and blocking segments in nasal harmony and rounding harmony,
reported in section 4.2.
Phonetically grounded featural co-occurrence restrictions have long played a central role
in analyses of transparency and blocking in harmony. In the rule-based framework develop by
Archangeli & Pulleyblank (1994), certain segments may not undergo harmony processes because
the set of targets of a spreading rule may be limited to those that obey a feature co-occurrence
restriction. The only difference between blockers and transparent segments is in the format of the
rule itself: blocking arises when a rule that inserts an association line between a feature and a
segment is prevented from applying to a restricted target. In contrast, transparency arises when a
rule inserts an additional token of a spreading feature on the other side of the transparent
segment, allowing that feature to avoid associating with a potential target that would violate the
rule’s co-occurrence restriction.
Many OT analyses of neutrality in harmony are similar to Archangeli and Pulleyblank’s
approach in their use of feature co-occurrence constraints to motivate both transparency and
blocking, though not necessarily in terms of featural insertion or copying. In several of these
constraint-based models, segments are parsed into the domain of a harmonizing feature that is
optimally aligned with a word edge. In Optimal Domains Theory (Cole & Kisseberth 1994,
1995), Headed Span Theory (McCarthy 2004, O’Keefe 2005), and Smolensky’s (1993) theory of
embedded feature domains, segments are parsed into feature domains that optimally span an
entire word. Transparency arises when a segment in the domain of harmony does not bear the
harmonizing feature due to the incompatibility of that harmonizing feature with one of the
segment’s other features. This incompatibility is captured by a feature co-occurrence constraint.
In all of these theories, in order to motivate transparency the co-occurrence constraint must
303
outrank some constraint that requires all segments in a harmony domain to bear that domain’s
feature. In Optimal Domains Theory, this constraint is called EXPRESSION; in Headed Span
Theory, it is ASSOCIATEHEAD; and in the theory of embedded feature domains, it is *EMBED.
A drawback of these accounts, which use feature co-occurrence constraints to capture
both transparency and blocking effects, is that they incorrectly predict that the sets of attested
blocking and transparent segments will be identical for a given harmony phenomenon. As
discussed in sections 4.2 to 4.4 and by C. Smith (2016a), this is not the case for nasal harmony,
in which all consonants are attested blockers but only obstruents are attested as transparent, and
for rounding harmony, in which all vowels are attested blockers but only high front vowels are
attested as transparent. I illustrate the issue with an example from Optimal Domains Theory, but
it should be noted that this problem extends to all analyses that use feature co-occurrence to
capture both blocking and transparency in harmony, whether rule- or constraint-based.
Cole & Kisseberth (1995) present a case of obstruent blocking of nasal harmony as the
result of the ranking of the co-occurrence constraint *[Nasal, Obstruent] (no nasalized
obstruents) and EXPRESSION (no transparency) over the harmony driving constraint WSA (Wide-
Scope Alignment, aligning a feature domain with a relevant word edge). This is demonstrated in
the tableau in (207) for the hypothetical input form /nata/. Following Cole & Kisseberth, parsing
into a nasal feature domain is indicated by parentheses.
(207) Obstruent blocking due to a featural co-occurrence restriction
Input: /nata/ *[Nasal, Obstruent] EXPRESSION WSA(Right)
a. [ (nãt ̃ã) ] *!
b. [ (nãtã) ] *!
c. [ (nã)ta ] **
304
Candidate (a) [nãt ̃ã] parses all segments in the word into the nasal feature domain, where
they surface as nasalized, satisfying WSA and EXPRESSION but violating *[Nasal, Obstruent],
which prohibits the nasal feature from being realized on the obstruent [t]. Candidate (b) [nãtã],
the transparency candidate, parses all segments into the nasal domain, including the [t], but
violates EXPRESSION by not nasalizing the [t] in order to satisfy *[Nasal, Obstruent]. Candidate
(c) [nãta], the blocking candidate, neither parses the [t] into the nasal feature domain nor
nasalizes it, violating WSA as the feature domain has not reached the right edge of the word.
However, higher-ranked *[Nasal, Obstruent] and EXPRESSION are satisfied, and candidate (c)
[nãta] is chosen as the winner.
Though they do not provide an example of transparency in nasal harmony, Cole &
Kisseberth state that transparency can be achieved by ranking *[Nasal, Obstruent] and WSA over
EXPRESSION. This is illustrated by the tableau in (208), which subjects the candidates in the
tableau in (207) to this new ranking. Now, the winner is the transparent candidate (b), [nãtã],
which violates only low-ranked EXPRESSION.
(208) Obstruent transparency due to a featural co-occurrence restriction
Input: /nata/ *[Nasal, Obstruent] WSA(Right) EXPRESSION
a. [ (nãt ̃ã) ] *!
b. [ (nãtã) ] *
c. [ (nã)ta ] *!*
However, obstruents are not the only attested blockers of nasal harmony, and *[Nasal,
Obstruent] is not the only featural co-occurrence constraint necessary to capture the full range of
nasal harmony blocking patterns (see Walker (1998/2000) for an overview). Liquids also
frequently block nasal harmony, which would presumably be captured by the constraint *[Nasal,
Liquid]. Liquid blocking of nasal harmony would thus be captured by the ranking *[Nasal,
305
Liquid], EXPRESSION >> WSA. However, this constraint set is also able to generate a nasal
harmony system in which liquids are transparent to nasal harmony. This occurs under the
ranking *[Nasal, Liquid], WSA >> EXPRESSION. The tableau in (209) demonstrates.
(209) Liquid blocking and unattested liquid transparency due to a featural co-occurrence
restriction
Input: /nala/ *[Nasal, Liquid] EXPRESSION WSA(Right)
a. [ (nãl ̃ã) ] *!
b. [ (nãlã) ] *!
c. [ (nã)la ] **
Input: /nala/ *[Nasal, Liquid] WSA(Right) EXPRESSION
a. [ (nãl ̃ã) ] *!
b. [ (nãlã) ] *
c. [ (nã)la ] *!*
In (209), the violation profile for the first input mirrors that of the tableau in (207), while
the violation profile for the second input mirrors that of the tableau in (208). Crucially, the
ranking *[Nasal, Liquid], WSA >> EXPRESSION selects a candidate with liquid transparency to
nasal harmony. This is an undesirable outcome, as liquid (and other non-obstruent) transparency
in nasal harmony is unattested according to Piggott (1992) and Walker (1998/2000). A simple
reordering of constraints is able to classify any natural class of segments as either blockers or
transparent segments based on that class’s incompatibility with a harmonizing feature. This leads
to the incorrect prediction that any segmental class that is attested as a blocker of some type of
harmony may also be transparent to that type of harmony.
This overgeneration of predicted transparent segments is not unique to Optimal Domain
Theory, however. Analyses of neutrality such as those proposed by Bakovic & Wilson (2000)
and Walker (1998/2000) rely on frameworks that compare candidates exhibiting transparency to
candidates with full harmony. They differ only in the mechanisms by which this comparison
takes place. In Bakovic & Wilson’s theory, inter-candidate comparison takes place because the
306
co-occurrence constraints that motivate segments’ neutrality to harmony are targeted constraints
(Wilson 2000, 2001, 2003). In Walker’s theory, this comparison is situated within Harmonic
Sympathy, as discussed in section 4.7.1. In both cases these mechanisms are triggered by high-
ranking constraints on co-occurrence between a harmonizing feature and the features of a
potential target. Because of this, both theories suffer from the same issue of overgeneration of
possible transparent segment types discussed above for featural domain theories. Walker
addresses this issue with her framework, suggesting a metric, based in perception and ease of
acquisition, by which harmony systems with transparent segments can be evaluated, with
systems with too many predicted transparent segments ruled out. However, this metric represents
a bias against unattested patterns of transparency in harmony rather than an outright ban.
In contrast, the Gestural Harmony Model makes no such predictions as transparency and
blocking are treated as the products of fundamentally different mechanisms with distinct
functional groundings. Transparency in harmony is the result of the concurrent activation and
subsequent blending of two antagonistic gestures, not an operation of the grammar. It is the
underlying representation of obstruents as including a velum closure gesture that leads to their
special status as the only attested transparent consonants in nasal harmony. Likewise, the
representation of high front vowels as including a lip spreading gesture is responsible for their
status as the only attested transparent vowels in rounding harmony. Co-occurrence constraints
crucially play no role in rendering certain segment types transparent to harmony in the Gestural
Harmony Model. Such restrictions do, however, come into play in the analysis of blocking of
nasal harmony. By applying the use of co-occurrence restrictions only to the analysis of blocking
segments, the Gestural Harmony Model is able to make different predictions about the types of
segments that can be transparent versus those that can serve as blockers.
307
Other analyses of neutrality in feature-based frameworks, including those proposed by
Kaun (1995) and Kimper (2011), have attempted to build segments’ tendencies to be transparent
or to block into the phonological grammar. This is achieved by positioning different segment
types along a neutrality scale or continuum, with those segments most likely to block on one end
and those most likely to be transparent on the other. While many of these approaches still utilize
co-occurrence restrictions to determine that some segment is neutral to a given harmony process,
its status as either a transparent or blocking segment is determined by its place on a neutrality
continuum. However, the neutrality continuum approach also makes incorrect predictions about
which segments may surface as transparent to harmony, again overgenerating predicted types of
transparent segments in both rounding and nasal harmony. Furthermore, any attempt to construct
a continuum of transparency and blocking for nasal harmony fails due to the special status of
obstruents as both the most likely blockers and the most likely (in fact, only) transparent
segments. The continuum approach generates several pathological patterns of neutrality in nasal
harmony.
One of the first proposals that neutral segment behavior should be captured by a
neutrality continuum is made by Kaun (1995). According to Kaun, certain segment types are
better suited to being transparent to a harmony process than other segment types. In addressing
this, she states the following:
‘Transparent elements are those elements which may occur during the span of
some feature while still allowing for the interpretation of that span as a cohesive
phonetic event…Elements or strings of segments may exhibit transparency if their
occurrence does not constitute a substantial interruption of the signal associated
with the extended feature’ (p. 211).
Precisely what constitutes a substantial interruption remains undefined, however. One
possibility is that it is the inherent length of a certain type of element that determines how
308
substantial an interruption it poses to an extended feature. Another possibility is that the notion
of a substantial interruption of a featural span is perceptually defined; a related proposal
regarding how speakers perceive harmony spans across transparent segments is made by
Boersma (1998, 2003). In this case, a segment type that is a perceptually strong bearer of some
featural contrast would present a greater interruption of a featural span than a weak bearer of that
featural contrast. This is certainly the case in terms of the different patterning of consonants and
vowels in vowel harmony. Consonants, as weak bearers of vocalic features such as rounding or
tongue root position, are considered to be transparent to vowel harmony in the vast majority of
cases, while vowels are transparent to harmony much less often. However, this does not extend
to the distinction between vowel heights as posited in Kaun’s continuum. In rounding harmony,
for instance, high vowels are strong bearers of a rounding contrast, while nonhigh vowels are
weak bearers of this contrast. This means that rounding is more perceptually salient when it
occurs on high vowels. If this is the case, the perception of a high unround vowel such as /i/ as
unround should be quite salient, while the unround quality of /e/ or /a/ should be less salient. If
perceptual salience is tied to what constitutes a substantial interruption of a featural span, then
unround /i/ would constitute a substantial interruption of a [+round] span relative to nonhigh
unround vowels and would therefore be less likely to behave transparently in rounding harmony.
This is in direct conflict with the typological facts of rounding harmony and the transparency
continuum that Kaun constructs from those facts. This indicates that the likelihood of surfacing
as transparent to vowel harmony is based on some factor other than a lack of perceptual salience,
or that it is based on multiple factors, such as perceptual salience and length.
309
Whatever the basis of segments’ positions along the neutrality continuum, Kaun infers
these positions based on typological patterns of neutrality. Based on these patterns, she provides
the neutrality continuum in (210):
(210) Transparency continuum, modified from Kaun (1995, p. 214)
On one end of this continuum are laryngeal consonants, which are considered the most
likely to be transparent to vowel harmony. Oral consonants are the next most likely segment type
to be transparent to harmony, followed by vowels of various heights. The least likely elements to
be transparent to harmony, and therefore most likely to serve as blockers, are series of two
syllables containing neutral vowels. Kaun proposes that every language with harmony
determines a cutoff point between transparent and blocking segments at some point along this
continuum. This is implemented in the grammar via a family of CONTINUITY constraints, as
defined in (211):
(211) CONTINUITY
X
: No element to the left of X on the transparency continuum may interrupt
an extended feature domain. (Kaun 1995, p. 215)
CONTINUITY is an anti-transparency constraint; it penalizes skipping of a segment in a
harmony process unless that segment is positioned sufficiently close to the right edge of the
continuum in (210). Kaun proposes that there is a CONTINUITY constraint for each position along
the transparency continuum.
When some markedness constraint is ranked above EXTEND(F) (Kaun’s harmony driving
constraint, which like SPREAD(F) (Padgett 1995; Walker 1998/2000) is evaluated by assigning
310
violation marks for non-undergoers), it prevents a segment from taking on a harmonizing feature.
Whether this neutral segment acts as a blocker or is transparent to harmony is determined by the
most stringent CONTINUITY constraint (i.e., the constraint that admits the fewest transparent
segments) that is ranked above EXTEND.
Kaun uses this constraint set to account for the distinct patterning of neutrality in
Mongolic and Tungusic rounding harmonies, both of which hold only among nonhigh vowels.
According to Kaun, these language families differ in what type of neutrality high vowels exhibit
in rounding harmony: in Mongolic languages, high front vowels are transparent, while in
Tungusic languages, high front vowels are blockers. Kaun accounts for high vowel neutrality in
both language families by ranking UNIFORM[round] over harmony-driving
EXTEND(round)IF[-high]; these families differ in which CONTINUITY
X
constraint is ranked above
EXTEND, as in (212).
(212) Constraint rankings for high vowel transparency versus blocking
a. Tungusic (consonants transparent, high vowels block):
CONTINUITY
C0
, GESTUNI >> EXTEND(+R)IF[-high]
b. Mongolic (consonants and high vowels transparent):
CONTINUITY
HighV
, GESTUNI >> EXTEND(+R)IF[-high] >> CONTINUITY
C0
In Mongolic, the most stringent CONTINUITY constraint to outrank EXTEND is
CONTINUITY
HighV
, which sets the cutoff point along the transparency continuum between high
and mid vowels, as in (213).
311
(213) Transparency continuum cutoff for Mongolic languages
In contrast, in Tungusic languages the most stringent CONTINUITY
X
constraint to outrank
EXTEND is CONTINUITY
C0
, resulting in a transparency continuum cutoff in which only consonants
are transparent to rounding harmony, while any neutral vowels will block, as illustrated in (214).
(214) Transparency continuum cutoff for Tungusic languages
While this approach makes more constrained predictions than typical co-occurrence
restriction-based analyses in accounting for certain segment types’ tendencies to be transparent
to or to block harmony, these predictions are still not constrained enough to capture the
typological asymmetries observed in nasal and rounding harmonies (section 4.2). I illustrate the
issue by applying Kaun’s neutrality continuum theory to neutrality in Turkic rounding
harmonies. As illustrated by the case of Tuvan rounding harmony in section 4.5.4, in Turkic
languages neutrality is usually driven by a restriction on nonhigh round vowels and enforced by
the constraint *ROLO (Kirchner 1993, Kaun 1995). With *ROLO above EXTEND(+R), nonhigh
vowels will be rendered neutral segments. In Tuvan and other Turkic languages, these nonhigh
312
vowels block rounding harmony, suggesting that CONTINUITY
C0
is ranked above EXTEND(+R).
The result is a cutoff along the transparency continuum identical to that of the Tungusic
languages; the only difference is in which vowels are determined to be neutral to rounding
harmony.
However, with these constraints it is also possible to generate a rounding harmony system
in which EXTEND(+R) is outranked by the less stringent CONTINUITY
LowV
, while CONTINUITY
C0
is
low-ranked. According to this ranking, nonhigh vowels are neutral to vowel harmony, and all
unround neutral vowels (and consonants) are transparent. This pattern generated by the
constraint ranking in (215b), in which nonhigh vowels such as /e/ and /a/ are transparent to
rounding harmony, is unattested.
(215) Constraint rankings for nonhigh vowel transparency versus blocking
a. Turkic (consonants transparent, nonhigh vowels block):
CONTINUITY
C0
, *ROLO >> EXTEND(+R)
b. Unattested (consonants and nonhigh vowels transparent):
CONTINUITY
LowV
, *ROLO >> EXTEND(+R) >> CONTINUITY
C0
The neutrality continuum approach, then, does not avoid the overgeneration of harmony
patterns with unattested types of transparent segments. While it is certainly true that high vowels
are more likely to be transparent to rounding harmony than mid or low vowels, it is only because
mid and low vowels are entirely unattested as transparent segments in rounding harmony.
However, any analysis relying on a continuum of neutrality is unable to capture this, as it is
designed to make use of the entire continuum and not just one edge of it.
In addition, the neutrality continuum approach is unable to accurately account for
patterns of neutrality in nasal harmony, for which it is unclear how to even establish the
313
neutrality continuum. On one end, obstruents are the most likely segments to be transparent to
harmony, due solely to the fact that they are the only type of segment that can be transparent to
nasal harmony. They are also the most likely type of segments to block harmony, indicating that
they should be simultaneously at either end of the neutrality spectrum. In addition, glides are the
least likely to block nasal harmony but are unattested as transparent segments, and liquids are
moderately likely to block nasal harmony but are also unattested as transparent segments,
rendering their positions on the continuum unclear. Following the line of reasoning that lesser
likelihood to block harmony translates to greater likelihood to be transparent to it, the continuum
in (216) is the closest possible approximation of a neutrality continuum for nasal harmony.
(216) Pathological transparency continuum for nasal harmony
The use of this continuum leads to a number of pathological predictions about possible
patterns of neutrality in nasal harmony. It predicts that not just obstruents, but all consonants
should be able to surface as transparent to nasal harmony in some language. The unique
patterning of obstruent transparency in nasal harmony is precisely what has led Walker
(1998/2000) to build an implicational hierarchy of blocking behavior around the idea that
transparent obstruents are best thought of as ‘permeable’ to nasal harmony, surfacing as
transparent to harmony as a specially conditioned consequence of undergoing harmony. This is
similar to the view of transparent segments in the Gestural Harmony Model, and is responsible
for accurately accounting for the status of obstruents as the only segment type that is able to be
314
transparent to nasal harmony. The use of a neutrality continuum, by contrast, generates a number
of pathological patterns of neutrality in nasal harmony.
Kimper’s (2011) competing triggers analysis of neutrality to harmony bears some
similarity to Kaun’s neutrality continuum approach. In Kimper’s framework, blocking of the
spread of a feature [+F] is modeled as triggering of the spread of the opposite feature value [-F].
When a harmony process encounters a neutral segment, the choice between transparency and
blocking comes down to competition between the segment bearing feature value [+F] and the
neutral segment bearing feature value [-F] to determine which will trigger further feature
spreading. This is illustrated in (217).
(217) Transparency and blocking as the result of competing triggers
a. Transparency b. Blocking
In (217), S
3
is the target of harmony and S
2
is a segment that is neutral to a harmony
process that spreads the feature value [+F]. According to the competing triggers analysis, S
1
and
S
2
compete to determine which will associate its feature specification to segment S
3
. In (217a),
S
1
wins the competition to spread its feature value onto S
3
, and S
2
surfaces as transparent to the
spread of [+F]. In (217b), on the other hand, S
2
wins this competition, and spreads [-F] onto
segment S
3
. This renders it a blocker of the spread of [+F].
Kimper’s competing triggers analysis of harmony is situated within the framework of
Serial Harmonic Grammar (Kimper 2011; Mullin 2011; Pater 2012), a variant of Harmonic
Serialism (McCarthy 2000, 2008a, 2008b; McCarthy & Pater 2016) in which constraints are
weighted rather than ranked. In this framework, harmony is driven by a positively-defined
315
constraint SPREAD(F), which assigns rewards for undergoers of harmony rather than violations
for non-undergoers. When a segment is prevented from undergoing harmony due to the ranking
of a relevant markedness or faithfulness constraint above SPREAD(F), that segment will surface
as neutral to harmony. In such cases, in order to choose between candidates displaying
transparency versus blocking, as in (217), Kimper adopts scaling conditions that alter the amount
of reward incurred by SPREAD(F). One of these scaling conditions increases the reward incurred
by SPREAD(F) according to the degree of perceptual impoverishment of the trigger. When a
reward incurred by spreading feature [-F] from a neutral segment, as in (217b), is greater than the
reward incurred by spreading a feature [+F] across a neutral segment, blocking occurs.
Therefore, a segment’s likelihood to either act as a blocker of harmony or to permit transparency
is determined by its placement on a scale of perceptual impoverishment.
In this way, Kimper’s approach mirrors Kaun’s (1995) appeal to a continuum of
likelihood to be transparent or to block harmony. One set of constraints determines whether a
segment is neutral, and another determines where along a scale or continuum to draw the line
between transparent and blocking segments. As a result, the issues that arise with the neutrality
continuum are mirrored by the competing triggers framework. There is nothing constraining the
values of the scaling factors of SPREAD(F) such that they wholly prevent certain types of
segments from ever surfacing as transparent to harmony. Instead, the competing triggers
framework is only able to represent relative likelihood of a segment to be transparent or to block
harmony. In addition, there remains the ambiguity of obstruent placement on any scale necessary
to analyze transparency and blocking in nasal harmony. The Gestural Harmony Model avoids
these issues by recasting transparent segments as undergoers of harmony, rather than as a type of
neutral segment.
316
4.7.3 Harmony Systems with Transparency and Blocking
Another area in which analyses of harmony can be distinguished is in their ability to
account for harmony systems in which some segments are transparent and some are blockers.
This is the case in many harmony systems, including Halh Mongolian rounding harmony
(Svantesson 1985; Svantesson et al. 2005), discussed in section 4.5.2, Coatzospan Mixtec nasal
harmony (Gerfen 1999, 2001), and Menominee ATR harmony (Archangeli & Pulleyblank 1994;
Archangeli & Suzuki 1995). Because the Gestural Harmony Model splits transparency and
blocking among two distinct theoretical mechanisms, in this model the two mechanisms may
operate concurrently within a single harmony system. As a result, the Gestural Harmony Model
is able to account for such patterns. Feature-based analyses of harmony, however, meet with
mixed success in accounting for such patterns.
An issue with many featural accounts of neutrality in harmony lies in their reliance on the
relative ranking of constraints to determine whether neutral segments in a given harmony system
will be transparent or blocking segments. In Optimal Domains Theory (Cole & Kisseberth 1994,
1995), for instance, the ranking of anti-transparency EXPRESSION over harmony-driving WSA
determines that any neutral segments in a harmony system will be blockers, while the reverse
ranking determines that any neutral segments will be transparent to harmony. This property is
shared by Bakovic & Wilson’s (2000) analysis of transparency and blocking in harmony using
targeted versions of AGREE (Baković 2000). In their theory, segments are neutral to harmony if a
co-occurrence constraint prevents them from taking on a harmonizing feature value; whether that
segment is transparent to or blocks harmony is determined by the relative rankings of different
types of AGREE constraints. If an AGREE constraint requiring a neutral segment to agree with
both adjacent segments for feature F is ranked higher, blocking is generated; if an AGREE
317
constraint requiring a neutral segment to agree with only one adjacent segment is ranked higher,
transparency is generated.
However, this sort of broad determination of what type of neutrality a harmony system
will exhibit is unable to account for harmony systems exhibiting both transparency and blocking.
By splitting transparency and blocking among two distinct theoretical mechanisms, the Gestural
Harmony Model allows these mechanisms to operate independently, and in some cases
concurrently. This is the case in Halh Mongolian rounding harmony (section 4.5.2), in which
high front vowels are transparent to harmony due to their inclusion of a lip spreading gesture that
is antagonistic to harmonizing lip protrusion gestures, and high back vowels blocking harmony
due to a constraint against their concurrent activation with lip protrusion gestures.
This success is shared by approaches that rely on a continuum or scale of segment
neutrality, such as those proposed by Kaun (1995) and Kimper (2011), discussed in greater detail
in section 4.7.2. Both of these frameworks are able to account for concurrent transparency and
blocking. Regarding Halh Mongolian, Kaun claims that while both front and back high vowels
are rendered neutral to rounding harmony by high-ranked UNIFORM[round], preventing cross-
height harmony, they occupy different positions on her proposed transparency continuum and are
on opposite sides of the transparency cutoff point put in place by the constraint CONTINUITY
HighV
,
depicted in the figure in (213) in section 4.7.2. In other frameworks there is one constraint that
determines whether a segment type is neutral to harmony and another constraint that determines
whether the harmony system as a whole exhibits transparency or blocking. In contrast, in Kaun’s
transparency continuum framework, markedness constraints such as UNIFORM[round] and
*ROLO determine which segments are neutral to harmony, and CONTINUITY
X
constraints
determine which segments are transparent and which block.
318
Analyses of transparency and blocking utilizing Agreement by Corespondence are also
successful in accounting for harmony systems exhibiting both transparency and blocking. Such
analyses are proposed by Rhodes (2012) for Halh Mongolian rounding harmony and by Walker
(2009, 2018) for Menominee ATR harmony. In Agreement by Correspondence, harmony is
driven as agreement between segments that are in surface correspondence relations.
Transparency results from a lack of surface correspondence between dissimilar segments, while
blocking results when similar segments are in correspondence but do not agree with one another.
Crucially, both of these configurations may arise in the same language. Like the Gestural
Harmony Model, the Agreement by Correspondence framework is successful in this regard
because is relies on two distinct theoretical mechanism to account for transparency and blocking.
4.7.4 Partial Harmony and Sour Grapes
Different analyses of harmony also make distinct predictions in the generation of ‘sour
grapes’ patterns of harmony. Originally coined by Padgett (1995), the term sour grapes refers to
patterns in which a harmonizing feature either spread throughout a domain, or not at all. That is,
harmony will not be triggered when a blocker is present in a word and harmony is unable to
reach the edge of a domain. As pointed out by Wilson (2003) and McCarthy (2003a, 2004), such
patterns are predicted by analyses that rely on the use of harmony drivers requiring adjacent
segments to agree for some feature value, such as AGREE (Baković 2000; Mahanta 2007, 2009;
Finley 2008, 2010).
This is demonstrated by the tableau in (218). The constraint AGREE(round) is ranked
above IDENT(round)-IO, resulting in rounding harmony. However, ranking of the constraint *y
over AGREE(round) prevents the vowel high front vowel /i/ from undergoing this process. As a
result of the presence of this blocking segment, harmony is prevented altogether.
319
(218) Tableau with sour grapes due to presence of a blocker
Input: /o-a-i/
*y AGREE(round) IDENT(round)-IO
a. [o-o-y] *! **
b. [o-o-i]
* *
c. [o-a-i] *
In candidate (a), all vowels surface as rounded, satisfying the harmony driving constraint
AGREE(round), but fatally violating higher-ranked *y. In candidate (b), rounding spreads onto the
second vowel, which surfaces as [o], but not the third vowel, which surfaces faithfully as [i].
This satisfies *y, but violates AGREE(round) because it includes a round-nonround vowel
sequence. In candidate (c), harmony is not triggered, and all vowels surface faithfully. Of note is
that this candidate also contains a round-nonround vowel sequence, and therefore also violates
AGREE(round). Because candidates (b) and (c) perform equally well with respect to the constraint
AGREE(round), the decision between these candidates falls to the constraint IDENT(round)-IO.
This constraint favors the candidate with no harmony, as it is the most faithful to the input.
The tableau in (218) demonstrates that local agreement constraints such as AGREE(F)
encounter issues with both over- and undergeneration of blocking patterns in harmony. The set
of constraints utilized here generates unattested sour grapes, while at the same time is unable to
generate robustly attested patterns of partial harmony with blocking. The candidate exhibiting
partial harmony is harmonically bounded, indicating that this set of constraints does not generate
partial harmony under any constraint ranking. Various strategies to avoid this undesirable
prediction have been employed. Bakovic & Wilson (2000) propose that blocking can be
generated in an analysis in which the markedness constraints that prevent segments from
undergoing harmony are targeted (on targeted constraints, see Wilson (2000, 2001, 2003)).
However, McCarthy (2002) argues against the effectiveness of targeted constraints in
constraining typological predictions when large sets of candidates are considered. Finley (2008,
320
2010), on the other hand, proposes that partial harmony can be generated, and sour grapes
patterns avoided, by assuming that local agreement constraints are evaluated directionally, i.e.,
by scanning an output candidate either left-to-right or right-to-left. Such directional constraint
evaluation is proposed by Eisner (2000). Under this proposal, a constraint like AGREE(F) could
be defined such that a candidate’s violations of the constraint are less costly the further they
occur from the trigger of harmony. While this approach does successfully generate patterns of
partial harmony, the broader typological consequences of admitting this sort of directional
constraint evaluation into OT remain largely unexplored.
In contrast, maximal harmony drivers that require a harmonizing feature to be associated
with as many segments in a domain as possible, such as ALIGN(F), EXTEND(F), and SPREAD(F),
generate partial harmony without issue and do not predict sour grapes patterns. In contrast with
local agreement constraints, maximal harmony drivers assign fewer violations to candidates in
which fewer segments fail to undergo a harmony process. As a result, such constraints are able to
account for harmony patterns in which a harmonizing feature spreads as far as it can within a
domain, and stops when it reaches either a blocker or a domain edge.
The Gestural Harmony Model is only partially successful in accounting for partial
harmony without sour grapes. This chapter has provided analyses of several harmony systems in
which a harmonizing gesture extends in activation until it is inhibited by a blocker. See, for
example, the analyses of nasal harmony in Orejón (section 4.4.3) and Capanahua (section 4.4.4),
rounding harmony in Baiyina Oroqen (section 4.5.3) and Tuvan (section 4.5.4), and tongue root
harmony in Standard Yoruba (section 4.6.2). These analyses indicate that the Gestural Harmony
Model is capable of generating patterns of partial harmony with blocking, avoiding the
undergeneration issue encountered by many analyses that rely on local agreement constraints to
321
drive harmony. However, the model does not avoid those constraints’ issue with overgeneration;
it also predicts that sour grapes patterns of harmony are generated under certain constraint
rankings.
As discussed throughout chapter 4, in the Gestural Harmony Model blocking results
when a grammar ranks a constraint from the *OVERLAP family over *INHIBIT. A sour grapes
pattern arises when both of these constraints outrank a constraint requiring a gesture to surface as
persistent or anticipatory. Under such a ranking, the grammar favors candidates in which the
overlap of incompatible gestures is prevented not by intergestural inhibition, but by not
triggering harmony at all. This is illustrated by the tableau in (219).
322
(219) Tableau with sour grapes due to high-ranked *OVERLAP and *INHIBIT
Input: / o
1
a
2
i
3
/
*OVERLAP(high V,
persistent LP)-IO
*INHIBIT
PERSIST(LP)
a.
*!
b.
*!
c.
*
In (219), candidate (a) [o-o-y] contains a persistent lip protrusion gesture that is not
inhibited by the high front vowel, resulting in full harmony and a fatal violation of high-ranked
*OVERLAP(high V, persistent LP). In candidate (b), the high front vowel blocks rounding
harmony via inhibition of the lip protrusion gesture, incurring a violation of *INHIBIT. In the
winning candidate (c), the lip protrusion gesture surfaces as self-deactivating and does not trigger
rounding harmony. This violates low-ranked PERSIST(lip protrusion), but satisfies both higher-
ranked oconstraints *OVERLAP(high V, persistent LP) and *INHIBIT.
Tongue Body
pharyngeal wide
1
Tongue Body
pharyngeal wide
2
Tongue Body
palatal narrow
3
Lip
protrusion
1
323
While the Gestural Harmony Model is able to generate patterns of partial harmony with
blocking, this tableau demonstrates that the model also predicts the possibility of sour grapes
spreading when constraints like *OVERLAP and *INHIBIT are ranked sufficiently high. One
possible solution to this issue is a reconceptualization of intergestural inhibition as the result of
one type of gesture being specified as a deactivator of another type of gesture, rather than as a
means of satisfying constraints from the *OVERLAP family. This possibility is discussed further
in section 6.2.2.
4.7.5 Non-Undergoers of Harmony
The final comparison of model predictions concerns the status of non-undergoers of
harmony in different frameworks, and how different treatments of non-undergoers make distinct
typological predictions. Wilson (2003) and McCarthy (2004) show that maximal harmony
drivers generate several pathological patterns by which the phonological grammar minimizes the
number of segments that do not undergo harmony due to the presence of a blocker in an output
form. This arises due to these constraints’ assigning violations to all non-undergoer segments
within some domain, regardless of their distance from a harmony trigger. Walker argues that this
non-locality between triggers and targets is an asset in generating some patterns of triggering, as
discussed in section 3.4.2. However, this same non-locality also leads to pathologies in blocking.
This is illustrated by the tableau in (220), adapted from an illustration by Wilson (2003,
p. 3). In this hypothetical nasal harmony system, nasal harmony proceeds through sonorants and
is blocked by obstruents. When the constraint that drives this harmony, SPREAD(nasal), is ranked
above another constraint motivating epenthesis, *CC, epenthesis occurs when the resulting
segment undergoes harmony, and does not occur when the resulting segment is inaccessible to a
324
spreading nasal feature due to the presence of a blocker. A high-ranked feature co-occurrence
constraint is assumed to motivate the blocking of nasal harmony by obstruents.
(220) Tableau with deletion of non-undergoers after a blocker
Input: /nawal-t/
SPREAD(nasal) *CC
a. [nãw
̃ ãl̃-t] * *
b. [nãw
̃ ãl̃-ə
̃ t]
*
Input: /nawak-t/
a. [nãw
̃ ãk-t]
** *
b. [nãw
̃ ãk-ət] **!*
For the first input in (220), the winning candidate is candidate (b) [nãw
̃ ãl̃-ə
̃ t], in which
epenthesis breaks up a consonant cluster and nasality spreads onto the epenthetic vowel.
However, the stem of the second input ends in an obstruent. Now the winner is candidate (a), in
which epenthesis does not apply. While candidate (b) satisfies *CC, epenthesis results in an
additional fatal violation of SPREAD(nasal) because the epenthetic vowel is blocked from
undergoing nasal harmony by the stem-final obstruent [k]. Wilson (2003) and McCarthy (2004)
argue that this issue extends beyond blocked epenthesis to include additional pathologies
involving the minimization of the number of non-undergoer segments in a word.
The Gestural Harmony Model, on the other hand, does not generate pathologies related to
the minimization of non-undergoer segments. As discussed in section 3.2, the constraints
PERSIST(Gest
X
) and ANTICIPATE(Gest
X
) are satisfied when a gesture of type X surfaces as
persistent and anticipatory, respectively. Whether a persistent or anticipatory gesture actually
overlaps any other segments in a word is irrelevant. As a result, there is no motivation to
minimize the number of non-undergoers of harmony when harmony is blocked, as is the case for
constraints that drive maximal spreading of a feature. This is an asset to the Gestural Harmony
325
Model, as it allows the model to avoid the generation of the harmony pathologies identified by
Wilson (2003) and McCarthy (2004).
4.8 Summary
One of the primary strengths of the Gestural Harmony Model lies in its distinct
representations of transparent and blocking segments, as evidenced by the discussion of various
types of harmony throughout this chapter. Key to the success of this model lies in its division of
the representations of transparency and blocking between two distinct theoretical mechanisms.
Relying on the Task Dynamic Model’s concepts of gestural competition and blending,
transparency is represented in the Gestural Harmony Model as the result of the concurrent
activation of antagonistic gestures. In this sense, transparent segments are undergoers of
harmony, albeit undergoers that are able to temporarily counteract the effect of a harmonizing
gesture upon the state of the vocal tract. Blocking, on the other hand, is the result of a ban on the
overlap of incompatible gestures, whose concurrent activation is either articulatorily or
perceptually marked in some way. This ban on the overlap of incompatible gestures is
implemented by a newly proposed intergestural relation by which gestures inhibit one another’s
levels of activation. The presence of such inhibition relations is motivated by constraints in the
phonological grammar.
Because transparency and blocking are split among two different theoretical mechanisms
and have different motivations underlying them, the Gestural Harmony Model makes distinct
predictions with respect to what types of segments are able to surface as transparent or blocking
segments. In nasal harmony and rounding harmony, gestural antagonism between a transparent
and a harmonizing gesture arises from a small set of segment type, while gestural incompatibility
with a harmonizing gesture arises from a larger set of segment types. Obstruents are the only
326
segments attested as being transparent to nasal harmony, and the only types of segments that
include a velum closure gesture that renders them antagonistic to a nasal harmony triggering
velum opening gesture. High front vowels are the only segments attested as being transparent to
rounding harmony, and the only types of segments that include a lip spreading gesture that
renders them antagonistic to a rounding harmony triggering lip protrusion gesture. In contrast,
there is no restriction on which types of segments may inhibit one another; any type of gesture
may inhibit any other type of gesture provided that inhibition is motivated by *OVERLAP
constraints in the phonological grammar. Because of this, the types of segments that are
predicted by the Gestural Harmony Model to serve as blockers of harmony is less constrained
relative to the types of segments that are predicted to surface as transparent to harmony. In
capturing the crosslinguistic asymmetries in nasal and rounding harmony discussed in section
4.2, this is a desirable prediction of the Gestural Harmony Model. Featural analyses that assume
the unitary concept of neutrality to harmony and treat transparent and blocking segments simply
as two possible types of neutral segments, on the other hand, do not match these predictions.
The splitting of the mechanisms responsible for transparency and blocking also ensures
that the Gestural Harmony Model is able to generate harmony systems that display both
transparency and blocking. This was exemplified by the examination of Halh Mongolian
rounding harmony in section 4.5.2. While transparency arises directly from segments’ gestural
representations, blocking is a product of the phonological grammar. By occupying distinct
theoretical spaces, the mechanism responsible for transparency and blocking are able to operate
concurrently with one another. Featural analyses that assume no such split between transparency
and blocking, meanwhile, meet with mixed success in accounting for these systems.
327
A major source of the success of the Gestural Harmony Model in accounting for
typological asymmetries between transparent and blocking segments lies in its reliance on the
concepts of intergestural competition and blending in its representation of transparency. Chapter
5 presents a more in-depth examination of these concepts, both in terms of how they are
implemented by the Task Dynamic Model of Speech Production and the role they play in the
phonological grammar.
Appendix B: Constraint Definitions
This appendix contains definitions for all of the constraints used in the analyses presented
in chapter 4.
*INHIBIT: Assign a violation mark to an inhibition relation between gestures in a coupling graph.
*INHIBIT(glide, velum opening): Assign a violation mark for an inhibition relation between a
glide and a velum opening gesture.
*INHIBIT(sonorant C, velum opening): Assign a violation mark for an inhibition relation between
a sonorant consonant and a velum opening gesture.
*INHIBIT(oral C, velum opening): Assign a violation mark for an inhibition relation between an
oral consonant and a velum opening gesture.
*OVERLAP(oral C, velum opening): Assign a violation mark to the gesture(s) of an oral
consonant that is/are active concurrently with a velum opening gesture.
*OVERLAP(nonhigh vowel, lip protrusion): Assign a violation mark to a nonhigh vowel gesture
and a lip protrusion gesture that are concurrently active.
*OVERLAP(front vowel, lip protrusion): Assign a violation mark to a front vowel gesture and a
lip protrusion gesture that are concurrently active.
*OVERLAP(high back vowel, persistent lip protrusion): Assign a violation mark to a high back
vowel gesture that is active concurrently with a persistent lip protrusion gesture.
*OVERLAP(nonhigh vowel, lip protrusion): Assign a violation mark to a nonhigh vowel gesture
and a lip protrusion gesture that are concurrently active.
328
*OVERLAP(lip protrusion; high V, nonhigh V): Assign a violation mark to a lip protrusion
gesture that is active concurrently with both a high vowel gesture and a nonhigh vowel gesture.
*OVERLAP(lip protrusion; high V, nonhigh V): Assign a violation mark to a lip protrusion
gesture that is active concurrently with a high vowel gesture followed by a nonhigh vowel
gesture.
*OVERLAP(high vowel, persistent lip protrusion): Assign a violation mark to a high vowel
gesture and a persistent lip protrusion gesture that are concurrently active.
*OVERLAP(high vowel, tongue root retraction): Assign a violation mark to a high vowel gesture
and a tongue root retraction gesture that are concurrently active.
*OVERLAP(low vowel, tongue root advancement): Assign a violation mark to a low vowel
gesture and a tongue root advancement gesture that are concurrently active.
*COUPLE(nonhigh short V, self-deactivating lip protrusion)
LICENSE(velum closure, glottis open): Assign a violation mark to a velum closure gesture that is
not active concurrently with glottal opening gesture.
LICENSE(lip protrusion, first σ): Assign a violation mark to a lip protrusion gesture that is not in
an initial syllable.
MAX(velum closure)-IO: Assign a violation mark to a segment (set of gestures) that includes a
velum closure gesture in the input if its output correspondent does not include that gesture.
MAX(lip protrusion)-IO: Assign a violation mark to a segment (set of gestures) that includes a lip
protrusion gesture in the input if its output correspondent does not include that gesture.
IDENT(deactivation)-IO: Assign a violation mark to a gesture whose input and output
correspondents do not have identical deactivation specifications.
PERSIST(lip protrusion): Assign a violation mark to a lip protrusion gesture that is self-
deactivating.
SELFDEACTIVATE: Assign a violation mark to a gesture that is not self-deactivating.
329
Chapter 5
A Closer Look at Gestural Strength
5.1 Introduction
The concepts of intergestural competition and blending adopted into the Gestural
Harmony Model from the Task Dynamic Model of speech production (Saltzman & Munhall
1989) are crucial to the representation of transparency to harmony in chapter 4. A segment is
predicted to be able to surface as transparent to a harmony process only if its gestural makeup
includes a gesture that is antagonistic to a harmonizing gesture. The result of this antagonism is
intergestural competition for control over the state of a vocal tract articulator. In this analysis of
transparency, the gesture of the transparent segment is always assumed to be stronger than the
harmonizing gesture that overlaps it, resulting in its ability to temporarily counteract the effect of
the harmonizing gesture on the vocal tract.
In this chapter, I provide a fuller explanation of how the results of intergestural
competition are calculated from gestures’ specified individual strengths and target articulatory
states. By defining intergestural competition, blending, and strength more explicitly, the validity
of the Gestural Harmony Model’s analysis of transparency is further strengthened. Another
aspect of this chapter’s strengthening of the claim that intergestural competition can result in
transparency involves the computational modeling of the articulatory trajectories and acoustic
outputs of concurrently active antagonistic gestures. This modeling is conducted in TADA (Nam,
Goldstein, Saltzman, & Byrd 2004), a MATLAB-based implementation of the Task Dynamic
Model of speech production.
330
I also challenge the assumption that transparency to harmony solely involves
configurations in which a transparent gesture is sufficiently strong to fully counteract the effect
of a harmonizing gesture upon the vocal tract. Gestural strength, as it is defined within the Task
Dynamic Model of speech production, is a gradient property. As such, it is predicted that cases
of partial transparency should also arise from this model. Such a case of partial transparency to
harmony is plausibly attested in the faucal harmony of Coeur d’Alene Salish.
In addition, in this chapter I examine the role that gestural strength may play in the
phonological grammar. Gestural strength is not merely a device that is employed in the
calculations of the Task Dynamic Model; it is a parameter that makes up part of the specification
of the phonological unit of the gesture. As such, there is no reason to assume that this parameter
is invisible to phonology. In particular, I focus on cases in which gestural strength appears to
serve a contrastive function and examine what sorts of phonological patterns this contrastive
gestural strength generates. Analyses based on contrastive gestural strength are proposed for
patterns of vowel retraction in Coeur d’Alene Salish faucal harmony, as well as two vowel-
consonant assimilation processes in Barrow Inupiaq.
The chapter is organized as follows. Section 5.2 provides more formal definitions of
gestural blending and gestural strength within the Task Dynamic Model. It also introduces the
TADA model and demonstrates how it can be used to computationally model the blending of
gestures with antagonistic target articulatory states that results in transparency to harmony. The
following two sections provide phonological analyses that demonstrate the utility of gestural
strength as a phonologically active parameter. Section 5.3 is an examination of faucal harmony
in Coeur d’Alene Salish, which serves as an interesting example of partial transparency to
harmony. It also introduces the possibility of contrastive gestural strength. This ability of
331
gestural strength to be contrastive is further explored in section 5.4 with a case study of two
phonological processes in Barrow Inupiaq. This case study examines the advantages of adopting
contrastive gestural strength as a possible explanation for cases of apparently exceptional
application of phonological processes. Section 5.5 concludes.
5.2 Gestural Strength in the Task Dynamic Model of Speech Production
5.2.1 Formal Definition of Gestural Strength
Discussions of gestural strength thus far have been largely informal in nature, referring to
gestures as ‘strong’ or ‘weak,’ and describing their influence on one another in terms of
‘competition.’ While these terms are descriptively useful, it is difficult to assess their theoretical
status or to examine the predictions they make without more explicit definitions. In this section, I
present these formal definitions.
Articulatory Phonology (Browman & Goldstein 1986, 1989, et seq.) assumes that during
the period of time in which a gesture is active, it gradually and asymptotically achieves its target
articulatory state. The time course of the achievement of this target articulatory state is
determined by a dynamically defined equation of motion, provided in (221).
(221) Dynamically defined equation of motion
ẍ = -k(x – x
0
) – bẋ
In this equation, x represents the current tract variable value, while x
0
represents the value
of this tract variable that is the gesture’s target articulatory state. The ẋ represents the first
derivative (velocity) of the gesture’s attainment of its target state, while ẍ represents its second
derivative (acceleration). The b in this equation is a damping constant, which ensures that the
tract variable value x will asymptotically approach its target articulatory state x
0
, rather than
oscillating around it. As discussed in section 1.2.1, k represents the gesture’s stiffness parameter,
332
which determines how quickly a gesture approaches its target articulatory state. While a gesture
is active, this equation of motion determines the state of the vocal tract with respect to the
specific tract variable referred to by the parameter x.
As discussed throughout chapter 4, concurrently active antagonistic gestures, i.e., those
with conflicting target articulatory states, enter into a state of competition with one another. This
competition is formalized as the blending of these two antagonistic gestures. The blending of two
gestures is calculated as the weighted averaging of the values of the parameters referenced in the
two gestures’ individual dynamically defined equations of motion. The parameters referenced by
that equation are the gestures’ respective target articulatory states x
0
and their respective stiffness
values k. The weighted averaging of these values is determined by each gesture’s strength,
denoted α. After two gestures’ α values are normalized such that they add up to 1, the blending
of those gestures’ individual target articulatory states is calculated as in (222).
(222) Blending of target articulatory states of gestures i and j
x
0ij
= (x
0i
* α
i
) + (x
0j
* α
j
)
Likewise, the blending of two gestures’ individual stiffness parameter values is calculated
as in (223). Again, this equation uses normalized α values that add up to 1.
(223) Blending of stiffness parameter values of gestures i and j
k
ij
= (k
i
* α
i
) + (k
j
* α
j
)
From these blended values for x
0
(target articulatory state) and k (stiffness), a new
equation of motion describing the achievement of the blended target articulatory state of two
gestures can be stated as in (224).
333
(224) Blended equation of motion
ẍ = -k
ij
(x – x
0ij
) – bẋ
If each active gesture is viewed as a control regime that holds over the vocal tract and is
defined by a dynamic equation of motion, then the blending of two gestures can be viewed as the
creation of a new blended control regime for the vocal tract that holds throughout the period of
time during which the two blended gestures are concurrently active. This is depicted in (225).
(225) Gestural score with resulting blended vocal tract control regimes
The proposals made regarding transparency as intergestural competition in chapter 4 can
now be restated more formally. In section 4.4.2, for instance, I propose that an obstruent is
transparent to nasal harmony in Tuyuca because it includes a strong velum closure gesture that
competes with and counteracts the effect of the harmonizing, relatively weak velum opening
gesture that overlaps it. More formally, a strong velum closure gesture is one with a high α value,
while a weak velum opening gesture has a relatively lower α value. The ‘competition’ between
the two gestures is formalized as gestural blending according to the equations in (221)-(224), and
‘winning’ an intergestural competition occurs when one gesture’s target articulatory state is
favored by the weighted averaging of the blending function in (222). Implemented as a weighted
averaging function, gestural blending does not represent a case in which a strong gesture wholly
overpowers a weak gesture. Rather, the equation of motion that results from the blending of the
two gestures more closely resembles that of the gesture with the higher value of α.
334
Demonstrating this result via computational modeling of the Task Dynamic Model of speech
production is the subject of the following section.
5.2.2 Computational Modeling of Transparency to Nasal Harmony
In this section I report the results of computational modeling of transparency via the
concurrent activation of antagonistic gestures. To put the Gestural Harmony Model’s analysis of
transparency to the test, simultaneous activation of velum opening and velum closure gestures of
various strengths were modeled in TADA (Nam et al. 2004), a MATLAB-based implementation
of the Coupled Oscillator Model and the Task Dynamic Model of speech production.
From the activation durations and gestural specifications in a gestural score, including
each gesture’s target articulatory state and strength parameter, TADA calculates the articulatory
trajectories necessary to achieve the target articulatory states specified for each gesture in a
gestural score according to the workings of the Task Dynamic Model of speech production. In
addition, it generates the acoustic output of the vocal tract shapes produced by the movements of
the articulators.
Recall from section 4.4.2 that in Tuyuca voiceless obstruents are transparent to nasal
harmony. This is illustrated by the data in (226), repeated from (130a-d) in section 4.4.2.
(226)
a. [mĩpĩ] ‘badger’
b. [w
̃ ãtĩ] ‘demon’
c. [ȷ
̃ ũkã] ‘yucca soup’
d. [ȷ
̃ õsõ] ‘bird’
TADA was used in order to test whether Tuyuca forms such as [mĩpĩ] ‘badger’ can
indeed be produced with a closed velum for the transparent [p] despite being produced during the
extended activation of a harmonizing velum opening gesture, as proposed in section 4.4.2. To do
this, a gestural score with the gestural specifications and durations in (227) was input to TADA.
335
(227) Gestural score input to TADA
The simulated [mĩpĩ] output by TADA is 55 frames (550 ms) long, with the velum
opening gesture active throughout that entire length of time. During frames 25 to 34, the velum
closure gesture is active as well. The target articulatory states of both gestures are specified
relative to velum aperture; however, they have different target values for the state of the velum.
Therefore, the resulting velum aperture tract variable that governs the state of the velum during
this time period is a weighted average of the velum aperture specifications of the two gestures
according to their α values. According to the gestural parameters built into TADA, the velum
opening gesture is specified for a target aperture of 0.2, while the velum closure gesture is
specified for a target aperture of -0.1.
36
These values are abstract, and their precise values are not
crucial here; what matters is that the target values for each gesture are distinct, with one specified
for an open velum, and the other a closed velum.
Velum closure gestures of various strengths were input to the model in order to examine
how gestural strength affects the resolution of the antagonism between the velum opening and
velum closure gestures. By varying the relative α values for the velum opening and velum
closure gestures of [mĩpĩ], the velum aperture trajectories displayed in (228) were obtained.
36
A target constriction degree with a negative value indicates a tight closure.
336
(228) TADA modeling of velum aperture for antagonistic gestures of different relative
strengths
While the velum opening gesture alone is active, the velum approaches its goal aperture
of 0.2. However, during frames 25 to 34, the velum closure gesture is active and its goal velum
aperture is blended with that of the velum opening gesture. During this period, velum aperture is
reduced to varying degrees according to the gestures’ relative α values. For the first simulation
(in lightest gray), the α value of the velum closure gesture is quite small relative to the α value of
the velum opening gesture, and very little change in velum aperture takes place. This can be
contrasted with the final simulation (in black) in which the α value of the velum closure gesture
is much greater than the α value of the velum opening gesture, resulting in a drastic change in
velum aperture and complete closure of the velopharyngeal port (i.e., velum aperture at or below
0). Intermediate manipulation of relative α values results in intermediate levels of velum aperture
change throughout the concurrent activation interval of the two velum gestures.
It should be noted that these simulations were carried out using TADA’s default gestural
specifications, which specify abstract goal velum apertures for velum closure and velum opening
gestures of -0.1 and 0.2, respectively. The values of α that result in a blended velum aperture of
zero or less for concurrently active gestures are based on these specifications. However, the
specific values used in these simulations are not as crucial as the overall result: it is possible for a
337
closed velum to result during a period of which a velum opening gesture is active. These
simulations therefore lend validity to the Gestural Harmony Model’s analysis of transparency as
the activation of an antagonistic gesture during the extended activation of a harmonizing gesture.
The results of this modeling of transparency to nasal harmony also bring to light
interesting predictions of the Gestural Harmony Model regarding the role of gradience of
gestural strength. Gestures are not simple ‘strong’ or ‘weak,’ but rather may take on any strength
value between one and zero. As a result, many of the velum aperture trajectories output by
TADA and depicted in (228) involve the blending of gestures of roughly equal strengths,
resulting in intermediate target states of the velum, neither fully open nor fully closed. Rather
than being a liability to the Gestural Harmony Model, I claim that the ability of the model to
generate these intermediate blending outcomes due to the gradient nature of the gestural strength
parameter is an asset to the theory.
Recall that in Tuyuca (section 4.4.2), voiced and voiceless obstruents are analyzed as
patterning differently with respect to nasal harmony. Voiced obstruents are overlapped by a
harmonizing velum opening gesture and surface as nasalized, while voiceless obstruents are
overlapped and surface as transparent to nasal harmony. I attribute the distinct patterning of
voiced and voiceless obstruents to the voiced obstruents’ lack of a velum closure gesture in the
surface forms of Tuyuca. However, the results of TADA modeling suggest another possibility:
that voiceless and voiced obstruents are not distinguished from one another based on the
presence versus absence of a velum closure gesture, but instead on the strengths of their velum
closure gestures. It could be the case that while voiceless obstruents possess a high-strength
velum closure gesture whose activation results in full closure of the velum, voiced obstruents
possess a velum closure gesture whose strength is not sufficient to result in full closure of the
338
velum when overlapped by a velum opening gesture. Such an analysis could have implications
for the analysis of across-morpheme nasal harmony in Tuyuca, in which voiced obstruents
pattern with voiceless obstruents in blocking nasal harmony from a root to a suffix (Barnes &
Takagi de Silzer 1976; Barnes 1996; Walker 1998). The matter is certainly worthy of further
study.
Beyond velum gestures, the results of TADA modeling also introduce a more general
prediction of the Gestural Harmony Model that gestures with strengths roughly equal to that of a
harmonizing gesture will surface as only partially transparent to harmony. I propose that such a
case of partial transparency does exist in the faucal harmony system of Coeur d’Alene Salish.
This case is discussed in section 5.3.
5.3 Partial Transparency via Gradient Gestural Strength
5.3.1 Partial Transparency in Coeur d’Alene Salish Faucal Harmony
In Coeur d’Alene Salish (also known as Snchitsu’umshtsn; Salishan; Washington, Idaho),
a process of regressive vowel-consonant harmony causes vowels to surface as retracted and
lowered. The language is described by researchers including Doak (1992) and Bessell (1998) as
having the five surface vowels indicated in (229). (There is also a non-phonemic schwa that
surfaces as the result of vowel reduction.)
(229) Coeur d’Alene Salish surface vowel inventory
37
Front Back
High i u
Nonhigh ɛ ɑ ɔ
37
Cole (1987) transcribes the front vowel [ɛ] as [ä]. Bessell and Doak both transcribe this vowel as [ɛ], though Doak
claims its quality ranges anywhere from [e] to [æ]. While Cole transcribes the vowel [ɔ] as [o] and classifies it as a
mid vowel, I assume that vowels in Coeur d’Alene Salish are only distinguished between high and nonhigh, and
treat [ɔ] as the round version of nonhigh back [ɑ].
339
The surface distributions of these vowels are dependent upon a process referred to as
faucal harmony. In faucal harmony, vowels surface as retracted preceding a ‘faucal’ consonant.
This faucal class includes the primary uvular/pharyngeal consonants /q/, /χ/, and /ʕ/, as well as
the coronal rhotic transcribed as /r/, whose production includes some kind of secondary
pharyngeal constriction.
38
Retraction of vowels preceding these consonants is demonstrated by the data in (230), in
which the alternating vowels are underlined. All data are from Doak (1992) and Bessell (1998).
39
(230) a. [t
s
iʃ-t] ‘it is long’ h. [t
s
ɛʃ-ɑlq
w
] ‘he is tall’
b. [dlim] ‘he galloped hither’ i. [t
ʃ
-dlɑm-ɑlq
w
] ‘train’
c. [q
w
it
s
-t] ‘warm’ j. [q
w
ɑt
s
-qən] ‘hat’
d. [sɛtt
ʃ
-nt
s
] ‘he twisted it’ k. [nɛʔ-sɑtt
ʃ
-ɛʔqs-n] ‘crank (on a car)’
e. [χɛt
s
-p] ‘he become curious’ l. [t-χɑt
s
-χət
s
-us] ‘he has curious eyes’
f. [ʔɛ-niʔ-kus-ɛlstʃn] ‘hair curls back
from forehead’
m. [ʔɑt-kɔs-qn] ‘his hair is curled’
g. [s-t-pum-əlx
w
] ‘hide with fur’ n. [s-pɔm-ɑlqs] ‘fur coat’
Note that the non-retracted vowels in (230c,e) demonstrate that this consonant-triggered
harmony is only regressive; vowels following a triggering faucal consonant are not retracted. In
most cases, uvular and pharyngeal consonants cause preceding vowels to surface as a non-high
back vowel. In the environment of faucal harmony, underlying /i/, /ɛ/, and /ɑ/ surface as [ɑ],
while underlying /u/ surfaces as [ɔ], analyzed here as the round counterpart of [ɑ]. There are,
however, some lexical items (such as (230a,h) above) that indicate that underlying /i/ sometimes
surfaces as [ɛ] rather than [ɑ] in faucal contexts. This vowel shift in faucal harmony contexts is
illustrated in (231).
38
All of these faucal consonants contrast with their labialized and/or glottalized counterparts, which also trigger
faucal harmony.
39
Doak and Bessell use slightly different consonant transcriptions. In particular, the consonant transcribed as [c] by
Doak is transcribed as [t
s
] by Bessell. I follow Bessell’s transcriptions here.
340
(231) Vowel shift in faucal harmony contexts in Coeur d’Alene Salish
This surface distribution of vowel qualities preceding faucal consonants holds not only
for the alternating forms in (230h-n) but also for static, within-morpheme vowel-faucal
sequences. Within a morpheme containing a faucal consonant, that consonant may only be
preceded by the vowels [ɛ], [ɑ], and [ɔ]. This is shown in the data in (232) from Doak (1992), as
well as the forms in (230h,i,k,n) above.
(232) a. [s-lɛq’-m] ‘baking camas in the ground’ *[s-liq’-m]
b. [n-pɑχ
w
-ət] ‘he went outside’ *[n-piχ
w
-ət]
c. [s-tɔpq-s] ‘thread’ *[s-tupq-s]
At first glance, Coeur d’Alene Salish faucal harmony appears to contradict the claim
made in section 2.3 that primary consonantal gestures involving oral closure or critical
constriction degree may not surface as persistent (non-self-deactivating) or anticipatory (early-
activating) and therefore may not trigger harmony. With the exception of /r/, the primary
gestures of the consonants that trigger faucal harmony involve either full or critical closure of the
vocal tract, and are predicted not to be able to extend in duration via persistence or anticipation.
The triggering of harmony by faucal consonants can be resolved by assuming that it is
not their primary consonantal gestures that serve as the triggers of faucal harmony. Instead, the
ability of a faucal consonant to trigger harmony in Coeur d’Alene Salish can be attributed to its
hypothesized gestural makeup including a secondary vocalic gesture for retraction of the
posterior tongue body in the pharyngeal region. I propose that it is this vocalic gesture, rather
than a faucal consonant’s primary oral closure gesture, that triggers faucal harmony. The
inclusion of a vocalic retraction gesture in the representation of rhotic consonants is well
341
supported. Many phonetic studies show that rhotic production often involves pharyngeal
constriction; see, for example, work by Delattre & Freeman (1968), Lindau (1985), Gick,
Iskarous, Whalen, & Goldstein (2003), and Proctor (2009). The inclusion of a retraction gesture
in the representations of the consonants /q/, /χ/, and /ʕ/, on the other hand, is somewhat more
controversial. I propose that the presence of this gesture is motivated as a way of enhancing or
aiding in the production of the uvular/pharyngeal closure gesture of a faucal consonant.
Generating regressive (leftward) faucal harmony in Coeur d’Alene Salish involves the
relative ranking of constraints ANTICIPATE(Gest
X
), SELFACTIVATE, and IDENT(activation)-IO that
should be familiar from the discussion throughout chapter 3. Therefore, I will not provide an
analysis of the constraint interactions necessary to derive the faucal harmony system of Coeur
d’Alene Salish. Instead, I will focus on the results of the gestural blending that arises from
overlap by the faucal harmony triggering tongue body retraction gesture.
Having determined the representations of faucal consonants in Coeur d’Alene Salish,
faucal harmony can be analyzed as being triggered by a tongue body gesture specified for
pharyngeal narrowing. Because it is anticipatory, this pharyngeal gesture extends regressively
(leftward) to overlap other segments. When preceding vowel gestures are overlapped, their target
articulatory states for tongue body position are in conflict with the target state of the anticipatory
tongue body retraction gesture. In order for this overlap to result in retraction of the tongue body,
the pharyngeal constriction gesture of the faucal consonant must be specified for a higher
strength than the gestures of vowels.
The result of overlap of a vowel gesture by an anticipatory tongue body gesture specified
for pharyngeal narrowing is illustrated in (233) for an /iq/ sequence in which the vowel /i/ is
specified for a relatively low strength. The result of overlap of the weak /i/ vowel by a
342
pharyngeal constriction gesture is that the vowel surfaces as [ɑ]. The dashed line indicates a
closed constriction degree.
(233) Full retraction of weak /i/
In (233), despite the gesture of /i/ being specified for a narrow constriction in the palatal
region, the vowel surfaces as [ɑ] due to overlap by the pharyngeal narrowing gesture. The
vowels /ɛ/ and /u/ will similarly surface as retracted when overlapped by this gesture.
However, there appear to be two different versions of the high front vowel /i/ in the
vowel inventory of Coeur d’Alene Salish. While most vowels in the inventory appear to fully
undergo harmony, surfacing as either [ɑ] or its rounded variant [ɔ], some high front vowels only
partially retract, surfacing as [ɛ] rather than [ɑ]. This is exemplified by forms such as [t
s
ɛʃ-ɑlq
w
]
‘he is tall’ (cf. [t
s
iʃ-t] ‘it is long’).
This pattern, in which a gesture has undergone harmony to some extent but appears to
have at least partially resisted it, can be analyzed within the Gestural Harmony Model as a case
of partial transparency to harmony. Rather than being strong enough to surface as fully
transparent to harmony, as in the cases of transparency discussed in chapter 4, in Coeur d’Alene
the /i/ that is found in some morphemes appears to be of an intermediate strength. As a result,
when its target articulatory state is blended with that of the harmonizing pharyngeal constriction
343
gesture, the result will be neither full retraction of the tongue body, nor full transparency. Rather,
the result is a vowel with a target intermediate state between the individual targets of the
overlapped tongue body gestures; this vowel is transcribed as [ɛ].
This overlap of a medium-strength /i/, resulting in partial retraction of the tongue body, is
illustrated in (234).
(234) Partial retraction/partial transparency of medium-strength /i/
The ability of gestural strength to take on intermediate values, rather than simply being
specified as weak or strong, is crucial to the success of this account of Coeur d’Alene Salish
faucal harmony. The partial transparency exhibited by the /i/ → [e] mapping in Coeur d’Alene
Salish faucal harmony is also important in that it fulfills a prediction of the Gestural Harmony
Model’s representation of transparency. If transparency is the result of blending between two
concurrently active antagonistic gestures, and gestural blending strength is a gradient property,
the model predicts that intermediate gestural strength of a transparent gesture will result in partial
transparency.
344
5.3.2 Computational Modeling of Coeur d’Alene Salish Faucal Harmony
The viability of the analysis of vowel retraction in Coeur d’Alene Salish faucal harmony
as a case of partial transparency to harmony is demonstrated in this section via computational
modeling of the harmony process in TADA. In order to test the hypothesis that both the /i
1
/ →
[ɛ] and /i
2
/ → [ɑ] mappings in Coeur d’Alene can result from overlap of the palatal /i/ gesture by
a vocalic gesture specified for pharyngeal constriction, the sequence /adidaq/ was synthesized.
The gestural score in (235) served as the input to TADA.
(235) Gestural score for /adidaq/, the input to TADA
Three versions of this gestural score were input to TADA, each with a different strength
ratio between the palatal gesture of high front vowel /i/ and the pharyngeal constriction gesture.
The table in (236) summarizes.
(236) Blending strength values for inputs to TADA
α for palatal gesture α for pharyngeal gesture
Strong /i/ 0.83 0.17
Medium-strength /i/ 0.5 0.5
Weak /i/ 0.17 0.83
The medium-strength /i/ gesture is intended to correspond to partially transparent /i
1
/ in
the figure in (234), and is equal to the harmonizing tongue body retraction gesture in strength.
The retraction gesture is five times stronger than weak /i/, which is intended to correspond to
fully retracting /i
2
/ in the figure in (233). The strong /i/ gesture is intended to correspond to the
345
fully transparent /i/ vowel that is found in other closely related varieties of Salish according to
Bessell (1998). This strong /i/ gesture is five times stronger than the harmonizing retraction
gesture.
For each gestural score, TADA produced a time series of vocal tract postures, as well as
synthesized audio. The results of TADA modeling of the gestural score in (235) with different
strengths of the palatal constriction gesture for high front /i/ show distinct degrees of tongue
body retraction during the production of /i/, as predicted. The synthesized vocal tract postures
during the production of /i/ are provided in (237).
(237) Synthesized tongue positions during production of underlying /i/ of different strengths
a. Fully retracted tongue
position during production
of weak /i/
b. Partially retracted tongue
position during production
of medium-strength /i/
c. Unretracted tongue
position during production
of strong /i/
For weak /i/ in (237a) the tongue body position achieved during the production of the
medial vowel is strongly retracted, despite this vowel’s underlyingly identity as /i/. For strong /i/
in (237c), the tongue body is quite advanced despite the presence of concurrently active
pharyngeal constriction gesture. Due to its high gestural strength, this /i/ appears to have surfaced
as fully transparent to harmony. Intermediate between these two vocal tract postures is the
medium-strength /i/ in (237b), which has surfaced as partially transparent to faucal harmony.
346
The three degrees to which the overlapped /i/ vowels have retracted as a function of their
different gestural strengths is also apparent in their acoustic signals, provided in the figures in
(238).
(238) Synthesized spectrograms for underlying /adidaq/ with /i/ gestures of different strengths
a. Produced as [ɑdɑdɑq] with weak /i/ gesture
b. Produced as [ɑdɛdɑq] with medium-strength /i/ gesture
c. Produced as [ɑdidɑq] with strong /i/ gesture
In the spectrograms in (238), the formant structure of the medial vowel varies depending
on the specified strength of the palatal gesture of underlying /i/. In (238a), the formant structure
of the medial vowel matches that of the nonhigh back vowels around it. As a result of faucal
347
harmony, the underlying /i/ vowel has surfaced as fully retracted and is indistinguishable from
the nonhigh back vowels around it. This gestural blending strength for /i/ is consistent with the /i/
in Coeur d’Alene Salish forms such as [dlim] ‘he galloped hither’ ~ [t
ʃ
-dlɑm-ɑlq
w
] ‘train.’ In
(238b), the less retracted tongue body results in a raised second formant relative to the
spectrogram of (238a). This intermediate strength for /i/ appears to be consistent with the /i/ in
Coeur d’Alene Salish forms such as [tsiʃ-t] ‘it is long’ ~ [tsɛʃ-ɑlqw] ‘he is tall.’ In (238c), the
spectral profile of the medial vowel closely resembles that of an [i], with a second formant
substantially higher than either of the vowels around it; this is despite having been overlapped by
a tongue body retraction gesture. While this /i/ that exhibits full transparency to faucal harmony
does not appear in Coeur d’Alene Salish, Bessell (1998) notes that there are several closely
related Salish varieties, including the dialect continuum of Spokane-Kalispel-Flathead Salish, in
which /i/ is fully transparent to faucal harmony.
TADA modeling has confirmed that the Task Dynamic Model of speech production can
represent different degrees of /i/ retraction as the result of the blending of tongue body gestures
of different strengths. It also confirms a prediction of the gestural representation of harmony that
there should be attested some harmony system in which a transparent gesture surfaces as only
partially transparent to a harmony process.
5.4 Contrastive Gestural Strength in Barrow Inupiaq
The analysis of Coeur d’Alene Salish presented in section 5.3 introduces the idea that the
parameter of gestural blending strength is able to serve a contrastive function within phonology.
In Coeur d’Alene Salish, lexical items can either contain a weak /i/, which surfaces as fully
retracted as the result of faucal harmony, or a medium-strength /i/, which is partially transparent
to faucal harmony and surfaces as only partially retracted. This introduces the interesting
348
possibility that within gestural phonology it is possible to use the gestural strength parameter to
represent contrasts among segments based on their susceptibility to undergo a phonological
process. This possibility is further explored in this section.
There has been some recent work that suggests that gestural strength plays an active role
in phonological processes. Iskarous, McDonough, & Whalen (2012), for instance, propose that
the velar fricative in Navajo is specified for a low gestural strength, which accounts for its
patterns of allophony based on surrounding vowel context. In the Gestural Harmony Model, high
gestural strength is proposed to be responsible for transparency to harmony processes. These
analyses are based upon the idea that gestural strength is a crucial part of a gesture’s
phonological representation and can determine its participation in certain phonological
processes. However, it is still assumed in these analyses that the specified strengths of gestures,
while phonologically active, are a fixed, language-specific property.
The case of the two high front vowels of Coeur d’Alene Salish suggests that the
phonological role played by gestural strength be expanded to include the possibility that the
setting of a gestural strength parameter can serve a contrastive function. However, instead of a
contrast based on a directly observable property of the production of a segment, such as its place
of articulation, contrastive gestural strength is only observable in terms of whether, or to what
degree, a gesture participates in some phonological process. Because of this, contrastive gestural
strength manifests as patterns that are often analyzed as cases of exceptionality in the application
of phonological processes.
Another case of apparent phonological exceptionality that can be analyzed instead as the
result of contrastive gestural strength comes from Barrow Inupiaq (Inuit; Alaska; L. Kaplan
(1981), Archangeli & Pulleyblank (1994), C. Smith & Blaylock (2017)).
I focus in particular on
349
two phonological processes in this language. First, a process of coronal palatalization causes
underlyingly coronal consonants to become palatalized following a high front vowel. This
process can occur across an intervening consonant and targets the coronals /t/, /n/, and /l/, which
surface as [s]~[c], [ɲ], and [ʎ], respectively. What is interesting about this process of coronal
palatalization is that it is triggered by some /i/ vowels and not others. This is illustrated by the
data in (239). In (239a-b), the high front vowel triggers palatalization of the following lateral,
while in (239c-d), palatalization does not take place. All data are from L Kaplan (1981).
(239) a. [iki-ʎu] ‘and wound’ c. [ini-lu] ‘and place’
b. [sɑvig-ʎu] ‘and knife’ d. [kamig-lu] ‘and boot’
Another process, dorsal assimilation, causes the high front vowel /i/ to retract and surface
as [ɑ] before a velar or a uvular consonant. Again, the process does not apply uniformly across
all lexical items. Interestingly, it is precisely those lexical items that trigger coronal palatalization
that fail to undergo dorsal assimilation. This is illustrated by the data in (240), which include the
same roots as (239). In (240a-b), the vowel of the second syllable surfaces as [i], while in (240c-
d) the vowel retracts, surfacing as [ɑ].
(240) a. [iki-k] ‘wound (dual)’ c. [innɑ-k] ‘place (dual)’
b. [sɑvvi-k] ‘knife (dual)’ d. [kɑmmɑ-k] ‘boot (dual)’
It is apparent from the data in (239) and (240) that these processes do not apply uniformly
among all instances of the vowel /i/. Instead there are a set of /i/ vowels that triggers
palatalization while failing to undergo dorsal assimilation, and a set of /i/ vowels that fail to
triggers coronal palatalization while undergoing dorsal assimilation.
L. Kaplan (1981) and Archangeli & Pulleyblank (1994) analyze the inconsistent
application of coronal palatalization and dorsal assimilation in Barrow Inupiaq by proposing that
there are two vowels in the language that surface as [i]. L. Kaplan claims that underlyingly
350
Barrow Inupiaq has the vowels /i/ and /ī/, the latter of which he describes as a schwa-like vowel.
Under his analysis, while both /i/ and /ī/ surface as [i] outside of the environment for dorsal
assimilation, their underlying featural specifications cause them to participate in these
phonological processes differently. He claims that at the step in the derivation at which coronal
palatalization applies, /ī/ lacks the featural specification to trigger this process alongside high
front /i/. It is this same featural specification that renders it a target of dorsal assimilation, while
underlying /i/ is not targeted. Similarly, Archangeli & Pulleyblank analyze the [i] that triggers
coronal palatalization and resists dorsal assimilation as underlying /i/, while an underlyingly
featureless vowel fails to trigger palatalization and undergoes dorsal assimilation.
I also adopt the line of analysis in which there are two /i/ vowels in the phonological
inventory of Barrow Inupiaq. However, the basis of the contrast between these vowels is
considerably different. Following C. Smith & Blaylock (2017), I propose that these two /i/
vowels have identical palatal constriction targets, but contrast for gestural blending strength. The
processes of coronal palatalization and dorsal assimilation in Barrow Inupiaq are the results of
the overlap and subsequent blending of the conflicting goal states of two antagonistic gestures.
While strong /i/ triggers coronal palatalization and resists dorsal assimilation, weak /i/ does not
trigger coronal palatalization and undergoes dorsal assimilation. Like Coeur d’Alene Salish,
then, Barrow Inupiaq appears to have a surface phonological inventory that includes two /i/
vowels that contrast for gestural blending strength.
The gestural analysis of coronal palatalization and dorsal assimilation in Barrow Inupiaq
is centered upon the gestural blending that takes place between these /i/ vowels and gestures with
which it is concurrently active. Under this analysis, palatalization occurs due to the overlap of an
alveolar constriction gesture by a preceding palatal gesture for an /i/. The result of this overlap is
351
antagonism with respect to the position of the tongue blade and anterior tongue body, as
illustrated by the vocal tract diagrams in (241).
(241) Antagonism between high front vowel and apical coronal gestures
a. Tongue position for /i/ b. Tongue position for coronal consonant
The /i/ gesture requires the anterior tongue body to be high and domed, bringing the
tongue tip to a lower position, while the alveolar gesture requires a lower position of the tongue
blade and anterior tongue body in order to achieve an apical constriction at the alveolar ridge.
Due to conflicting target articulatory states, when these gestures are concurrently active they will
undergo gestural blending to determine what position the tongue will take. The result of this
gestural blending is determined by the gestures’ relative blending strength values.
In the case of strong /i/, gestural blending favors the target articulatory state of the strong
/i/ gesture at the expense of the alveolar gesture. The result is a palatal constriction during the
time in which the two gestures are concurrently active. Palatalization is a direct result of the
overlap of the alveolar gesture by the palatal gesture, and is output by the calculations of the
Task Dynamic Model of speech production. The process is illustrated by the gestural score and
resulting constriction location time course in (242).
352
(242) Gestural score for sequence of strong /i/ and coronal consonant
In the case of weak /i/, gestural blending favors the target articulatory state of the alveolar
gesture over that of the concurrently active palatal gesture. Therefore, palatalization does not
result during the overlap of the two gestures, and alveolar consonants surface as alveolar. The
failure of this /i/ to trigger palatalization can be attributed to its low gestural blending strength.
This is illustrated in (243).
(243) Gestural score for sequence of weak /i/ and coronal consonant
The same distinction between strong and weak /i/ comes into play in the analysis of
dorsal assimilation. However, in this case the distinction affects whether or not /i/ is targeted by
the process, rather than triggering it. Again, the process is analyzed as the result of the overlap of
two antagonistic gestures. I claim that dorsal assimilation in Barrow Inupiaq is the result of a
uvular or velar consonantal gesture not strictly following a vowel gesture, but instead
overlapping a preceding vowel gesture to some degree, as in the gestural scores in (244) and
353
(245). I assume that these consonantal gestures are of a sufficiently low stiffness that they will
not reach their consonantal target closures until after the deactivation of the vowel gesture.
Overlap between the vowel gesture for the /i/ and the gesture of a velar or uvular consonant
results in antagonism; /i/ requires a high front position of the tongue body, while a velar or
uvular consonantal gesture requires the tongue body to be retracted in the uvular region. Again,
the outcome of the overlap of these antagonistic gestures is determined by their relative
strengths.
The blending of a uvular closure gesture and a weak palatal gesture for /i/ will favor
retraction of the tongue body. The result of this overlap will be the back vowel /ɑ/, as in (244).
(244) Gestural score for weak /i/ overlapped by following uvular consonant
The goal articulatory state of a strong /i/, on the other hand, will be favored over that of a
uvular consonantal gesture. As a result, vowel retraction does not result during the overlap of the
two gestures, and /i/ is able to surface as palatal. The /i/ gesture has resisted dorsal assimilation
due to its high specified strength. This is illustrated in (245).
354
(245) Gestural score for strong /i/ overlapped by following uvular consonant
There are a number of advantages to the contrastive gestural strength analysis of these
processes in Barrow Inupiaq. First, it recruits an independently necessary element of gestural
representations, gestural strength, to account for this phonological behavior. In addition, it
provides a unified explanation for the apparently exceptional applications of palatalization and
dorsal assimilation, rather than treating them as unrelated processes. Finally, the gestural strength
analysis also eliminates the need to rely on various theoretical mechanisms that are typically
employed to account for cases of apparent exceptionality.
Because they include no direct analog to gestural strength, more traditional feature-based
analyses of apparently exceptional or inconsistent application of a phonological process must
distinguish between seemingly identical segments or lexical items in other ways. In derivational
analyses, that often involves positing a fully abstract underlying phoneme that never appears in
surface forms. This is the case for analyses of Barrow Inupiaq by L. Kaplan (1981) and
Archangeli & Pulleyblank (1994), who posit an abstract underlying vowel that later merges with
/i/. Under these analyses, a coronal palatalization rule will apply before the abstract vowel is
merged with /i/, preventing it from triggering coronal palatalization. This is a case of
counterfeeding opacity. Likewise, the dorsal assimilation rule will apply before the rule that
355
derives [i] from the underlying abstract vowel, accounting for its susceptibility to dorsal
assimilation.
As discussed in section 3.6.3, this sort of analysis based on absolute neutralization with
counterfeeding opacity is incompatible with parallel phonological frameworks such an
Optimality Theory or Harmonic Grammar, unless additional theoretical architecture is adopted.
The gestural account of Barrow Inupiaq avoids this issue as it does not rely on abstract
underlying phonemes, absolute neutralization, or rule ordering. Instead, under this account the
contrastive element, gestural strength, is not neutralized; it persists from the underlying to the
surface form and results in the different consequences of overlap observed in (242)-(245).
One way of dealing with these sorts of apparently exceptional patterns in a way that is
compatible with parallel phonological frameworks is indexation between morphemes and
constraints (Pater 2000, 2009a; Flack 2008; Becker 2009). Exceptional triggering of
palatalization can be generated by indexing a high ranked markedness constraint to triggering
morphemes; for the sake of simplicity this constraint will be referred to as PALATALIZE. Ranking
a relevant faithfulness constraint between the indexed and unindexed versions of this constraint,
as in (246), ensures that only morphemes bearing the index X will trigger palatalization.
(246) Constraint ranking for exceptional triggering of palatalization by X-indexed morphemes
PALATALIZE
X
>> IDENT(high)-IO >> PALATALIZE
Similarly, exceptional failure to undergo dorsal assimilation can be generated via
indexing of a high-ranked faithfulness constraint. Ranking the indexed version of this constraint
above the markedness constraint that drives dorsal assimilation and the unindexed version of this
constraint below that markedness constraint, as in (247), ensures that the grammar will only
preserve the underlying backness value of vowels in morphemes bearing the index Y.
356
(247) Constraint ranking for exceptional resistance to dorsal assimilation by Y-indexed
morphemes
IDENT(back)-IO
Y
>> ASSIMILATE >> IDENT(back)-IO
While these rankings will generate the patterns of palatalization and dorsal assimilation
exhibited by Barrow Inupiaq, several issues arise by adopting this constraint indexation analysis.
The first issue is that if a morpheme contains multiple high front vowels, the indexation of that
morpheme to PALATALIZE and IDENT(back)-IO predicts uniform behavior of all /i/ vowels in the
morpheme with respect to palatalization and dorsal assimilation. In other words, a morpheme
will contain either all palatalizing /i/ vowels or all non-palatalizing /i/ vowels. However, there are
a number of morphemes in Barrow Inupiaq that contain both palatalizing and non-palatalizing /i/
vowels. One example is the word [ilvi-ʎʎi] ‘and you, in your turn,’ in which the first [i] does not
trigger palatalization of the immediately following [l], but the second [i] does. Another example
is the stem /siɲik/ ‘sleep.’ The first [i] triggers palatalization of following [ɲ], as in [siɲik-pa] ‘is
he sleeping?,’ while the second [i] does not, as in [siɲik-tuq] ‘sleeps’ (*[siɲik-cuq]). This
indicates that the exceptional patterning of vowels in this language must be assessed for an
element smaller than the morpheme.
A possible solution is to consider that constraints are indexed to individual segments
rather than entire morphemes, as proposed by Temkin Martínez (2010). With the indexed
constraint rankings necessary for an account of Barrow Inupiaq, an inventory with two /i/ vowels
can be posited, with one /i/ indexed to both high-ranked constraints, PALATALIZE and
IDENT(back)-IO, and one that is indexed to neither. This vowel inventory is provided in (248).
357
(248) Barrow Inupiaq vowel inventory with two indexed /i/ phonemes
/ɑ/, /u/, /i
XY
/, /i/
However, under the constraint indexation approach, the indexation of the gestural
analysis’ strong /i/ to both PALATALIZE and IDENT(back)-IO is entirely accidental. It is also
possible to generate a system in which PALATALIZE and IDENT(back)-IO are not indexed to the
same sets of /i/ vowels. That is, an /i/ vowel may be indexed only to PALATALIZE, or only to
IDENT(back)-IO, and a phonological inventory may include any combination of these indexed /i/
vowels, or even all of these vowels. This is illustrated by the possible vowel inventory in (249).
(249) Predicted possible high vowel inventory with two indexed constraints
/i
XY
/, /i
X
/, /i
Y
/, /i/
When it is only necessary to propose the indexation of two constraints, the predicted
increase in the size of a possible phonological inventory is not ideal, but also not a serious
problem. The indexation of two constraints results in four possible vowel-index combinations.
However, there are additional phonological processes in Barrow Inupiaq that rely on the
distinction between what the gestural analysis calls strong and weak /i/. According to L. Kaplan
(1981), there is also a process in which weak /i/ retracts to surface as [u] in certain environments,
as well as a process in which weak /i/ is deleted. The analysis of each of these processes will
need to make use of both an indexed and a general version of a constraint in order to differentiate
between weak /i/ and strong /i/. With four indexed constraints, there are sixteen possible vowel-
index pairs, and therefore a predicted phonological inventory that contains sixteen /i/ vowels,
each with a different set of constraint indices. This points out an important point about the
constraint indexation approach to dealing with apparent exceptionality: as the number of indexed
constraints increases, the size of a language’s possible segment inventory will increase at a rate
358
of 2
n
, where n is the number of indexed constraints in the grammar. By allowing the constraints
that motivate different processes to index to segments independently from one another, this
approach misses a generalization about the relatedness of certain processes.
The account of apparent phonological exceptionality in Barrow Inupiaq based on
contrastive gestural strength encounters no such issues with exploding inventory size. In contrast,
the gestural strength account of coronal palatalization and dorsal assimilation in Barrow Inupiaq
relies on a single gestural parameter, gestural blending strength, to generate the patterning of the
high front vowel with respect to multiple processes. In doing so, this analysis captures the
generalization that a vowel’s ability to trigger palatalization and its failure to undergo dorsal
assimilation are related. By capturing this relatedness, the contrastive gestural strength analysis
avoids the kinds of inventory size explosions predicted by an account based on segment
indexation to constraints.
There is an open question as to how many gestural strength parameter settings may be
utilized contrastively by a language. The cases of Coeur d’Alene Salish and Barrow Inupiaq
discussed here are each analyzed as involving a contrast between vowels of two different
gestural strengths. However, because gestural strength is specified on a gradient scale,
potentially much more fine-grained contrasts are also predicted to be possible. For instance, a
language could in principle contrast five /i/ vowels: very weak (α=0.20), weak (α=0.40), medium
(α=0.60), strong (α=0.80), and very strong (α=1.0). There is no theoretical upper limit on how
fine-grained the contrasts between gestures of different strengths may be, and therefore no
theoretical limit on the number of segments that may contrast for gestural strength in some
language.
359
However, I claim that contrastive gestural strength still does not necessarily predict the
types of inventory size explosions as constraint indexation, because the contrasts between
gestures based on their strength parameter settings must be sufficiently perceptually distinct.
While it is true that a language could, in principle, contrast two gestures with specified strengths
of 0.82 and 0.83, such a contrast is unlikely to ever arise. The differences in how these gestures
would blend with and affect the production of the gestures around them would be so slight as to
render those differences imperceptible. As a result, any contrast between them would likely be
highly unstable. Contrasts between gestures based on their strengths are only predicted to arise if
those distinct gestural strengths result in perceptually distinct results of blending. Therefore, the
number of gestural strength parameter settings that can be utilized contrastively by a language is
constrained by perceptual factors.
This section has demonstrated that the gestural analysis of multiple apparently
exceptional processes in Barrow Inupiaq avoids issues of both undergeneration and
overgeneration of phonological patterns. Beyond that, it has also demonstrated the utility of
gestural blending strength in accounting for phonological processes other than harmony. This
suggests that many of the innovations introduced by the Gestural Harmony Model can be
generalized beyond the study of harmony. In the case of granting contrastive status to gestural
blending strength, these innovations extend into the realm of phenomena that are often treated as
cases of phonological exceptionality.
5.5 Summary
This chapter has provided an in-depth examination of gestural strength and blending,
concepts that play a key role within gestural phonology. Originally conceived within the Task
Dynamic Model of speech production, gestural strength and blending are recruited by the
360
Gestural Harmony Model to account for transparency as the result of overlap between
antagonistic gestures. A major consequence of this approach is in the ability of the Gestural
Harmony Model to account for asymmetries between attested transparent and blocking segments,
as discussed in chapter 4. In addition, the gradient nature of gestural strength allows the Gestural
Harmony Model to account for cases of partial transparency, exemplified by Coeur d’Alene
Salish faucal harmony in section 5.3.
Looking beyond the study of harmony, the case of Barrow Inupiaq fulfills a prediction
that gestural strength, as a phonological parameter of a gesture, should play an active role in
phonology, even taking on a contrastive function in some languages. In addition to the case of
Barrow Inupiaq, there are a number of other possible cases of apparent phonological
exceptionality that could benefit from an analysis based on contrastive gestural strength as well.
These include processes that distinguish between the standard and ‘superclose’ high vowels in
various Bantu languages, including Bemba (Hyman 1994; Zoll 1995); the two /i/ vowels of
Kashaya (Buckley 1994); and the two /w/ glides of Fula (S. Anderson 1976a, 1976b). In Kashaya
(Pomoan; northern California), Buckley analyzes the idiosyncratic patterning of several
processes that target the vowel /i/ as the result of the language’s inventory containing an abstract
vowel. This is remarkably similar to the analyses of Barrow Inupiaq by L. Kaplan (1981) and
Archangeli & Pulleyblank (1994). In Bemba (also known as Cibemba; Bantu; Zambia), Zoll
(1995) analyzes the idiosyncratic ability of high surface vowels to trigger various consonant
mutations as the result of the distinction between underlying vowels that are historically derived
from the Bantu ‘superclose’ vowels and underlying vowels that are derived from standard high
vowels.
361
Fula represents an especially interesting case of apparent exceptionality. S. Anderson
(1976, 1976) suggests that there are two types of /w/ in the language, one in which the labial
component is consonantal, and one in which the dorsal component is consonantal.
40
For each of
these versions of /w/, its consonantal portion determines how it patterns with respect to a process
of consonant mutation involving alternation between stop and continuant. Such a pattern could
be captured straightforwardly within gestural phonology. In an inventory with two versions of
/w/, both would have a lip protrusion gesture and a tongue body gesture specified for uvular
constriction. However, they would differ as to which of the gestures is the stronger of the two. In
addition to providing another potential case of contrastive gestural strength, the two /w/ glides of
Fula also speak to the assumption within gestural phonology that consonantal gestures are
generally strong and vocalic gestures are generally weak. Fula would provide a valuable case
through which to examine how gestures are classified as consonantal and vocalic, and the role
that gestural strength plays in that classification.
40
Thanks to Nick Danis for bringing this case to my attention.
362
Chapter 6
Conclusion and Further Issues
6.1 Summary of the Dissertation
This dissertation has introduced the Gestural Harmony Model and argued for the
representation of harmony as the result of a gesture’s extended duration such that it overlaps
other gestures in a word. The ability of a gesture to surface with this extended duration is the
result of two newly proposed gestural parameters. One parameter determines whether a gesture
self-deactivates once its target articulatory state is achieved, or whether it is a persistent gesture
that does not self-deactivate and therefore acts as a trigger of progressive harmony. The other
parameter determines whether a gesture activates at the 0º phase of its planning oscillator, or
whether it is an anticipatory gesture that activates before this phase and acts as a trigger of
regressive harmony.
By casting the triggering of harmony as the result of a gestural parameter setting, the
grammar of triggering in the Gestural Harmony Model is one based on shaping surface
phonological inventories such that they contain harmony-triggering (persistent and/or
anticipatory) gestures, as well as placing distributional restrictions on members of those
inventories. In reconceptualizing the driving of harmony in this way, the Gestural Harmony
Model provides novel solutions to some of the issues that arise in the analysis of harmony
triggering patterns. The model successfully accounts for harmony systems in which harmony
triggers are restricted to specific positions (such as Kyrgyz rounding harmony, sections 2.2.1 and
3.2.1), as well as those that place conditions on the identities of triggers of harmony (such as
Baiyina Oroqen rounding harmony, section 3.4.2). The Gestural Harmony Model finds particular
363
success in accounting for patterns of contrastive triggering in harmony, including those in which
the contrast between triggers and non-triggers is restricted to privileged positions. This is
exemplified by analyses of nasal harmony in Acehnese and Rejang (section 3.3), as well as
tongue root harmony in Classical Manchu (section 3.5).
The Gestural Harmony Model also makes major inroads in the study of transparency and
blocking, and the observed typological asymmetries between them. In this model, a transparent
segment is analyzed as a special type of undergoer, one that includes a gesture that is
antagonistic to the harmonizing gesture. This antagonism results in the temporary counteracting
of the effect of the harmonizing gesture on the vocal tract due to the blending of the antagonistic
gestures’ individual target articulatory states. Blocking, on the other hand, is the result of a
grammatical restriction on the overlap of incompatible gestures, whose concurrent activation is
either articulatorily or perceptually marked in some way. This restriction is enforced by a newly
proposed type of intergestural relation: inhibition. When two gestures are in an inhibition
relation, an inhibiting gesture will leech the activation from the inhibited gesture, either
preventing it from activating (in the case of regressive harmony) or deactivating it (in the case of
progressive harmony).
By analyzing transparent segments as undergoers of harmony, the Gestural Harmony
Model makes important predictions about the types of segments that should be able to surface as
transparent to harmony. Only those segments whose representations include a gesture that is
antagonistic to a harmonizing gesture are predicted to be transparent to harmony. Because of
this, the Gestural Harmony Model correctly predicts observed typological asymmetries between
attested transparent and blocking segments in nasal harmony and rounding harmony. Obstruents
are the only segments attested as being transparent to nasal harmony, and the only types of
364
segments that include a velum closure gesture that is antagonistic to a velum opening gesture that
is responsible for nasal harmony. High front vowels are the only segments attested as being
transparent to rounding harmony, and the only types of segments that include a lip spreading
gesture that is responsible for rounding harmony. In contrast, there is no restriction on which
types of segments may inhibit one another. Because of this, the types of segments that are
predicted by the Gestural Harmony Model to serve as blockers of harmony are unconstrained
relative to the types of segments that are predicted to surface as transparent to harmony.
There are a number of additional advantages to the Gestural Harmony Model’s
representations of transparency and blocking. By representing transparency as the result of
blending between two gestures according to their specified strength parameters, the Gestural
Harmony Model is able to account for cases of partial transparency, exemplified by Coeur
d’Alene Salish faucal harmony (section 5.3). In addition, the splitting of transparency and
blocking among two distinct theoretical mechanisms ensures that the Gestural Harmony Model is
able to generate harmony systems that display both transparency and blocking, such as Halh
Mongolian rounding harmony (4.5.2).
In addition, the gestural analysis of coronal palatalization and dorsal assimilation in
Barrow Inupiaq (section 5.4) suggests that the innovations of the Gestural Harmony Model are
not limited to the study of harmony. By expanding the phonological role of gestural blending
strength to include cases in which gestural strength serves a contrastive function, gestural
representations are able to account for patterns that have previously been treated as cases of
phonological exceptionality.
365
6.2 Further Issues
The development of the Gestural Harmony Model advances our understanding of what
drives many of the crosslinguistic patterns observed across harmony systems. However, this
dissertation also leaves some questions open to further study. In this section I discuss some of
these open questions, as well as avenues for further research. I focus on issues involving (1) the
representation of vowel place within gestural phonology, (2) an elaboration of the theory of
intergestural inhibition, and (3) directionality of harmony.
6.2.1 Gestural Representation of Vowels
Throughout this dissertation, the focus has been on harmony systems that are based on
the extended duration of a secondary gesture of a segment, e.g. nasal harmony (velum opening),
rounding harmony (lip protrusion), and tongue root harmony (tongue root
advancement/retraction). It has not addressed two common types of vowel harmony: those based
on vowel height and backness. This is because as the gestural phonology framework currently
stands, these types of harmony cannot be represented as the result of extended gestural duration
due to the absence of gestural analogs to height and backness.
Recall from section 1.2.1 that Articulatory Phonology assumes that lingual gestures
(those for either the tongue tip or tongue body) for both consonants and vowels are specified in
terms of constriction location and constriction degree. Front vowels are specified for palatal
constriction, while high back vowels are specified for uvular constriction and nonhigh back
vowels are specified for pharyngeal constriction. This gestural coordinate system, based on
constriction location and constriction degree, is illustrated in (250).
366
(250) Gestural representation of constriction location for vowels
This polar coordinate system can be contrasted with a typical featural mapping of vowel
place along a Cartesian coordinate system with two dimensions: height and backness. Assuming
a featural system for vowels based upon these dimensions, the representation of height harmony
and backness harmony as feature spreading is straightforward. The same cannot be said for a
gestural framework that assumes vowel representations based on palatal, uvular, and pharyngeal
constriction location. The question, then, is how to reconcile attested harmony systems based on
the spread of vowels’ properties of height and backness with a gestural representational system
that does not rely on these dimensions in the representation of vowels.
One possible solution is to redesign the representation of vowels within gestural
phonology such that they do rely on gestures that specify the height and backness of the tongue
body, rather than the location and degree of tongue body constriction. Instead of relying on
gestures specified for the dimensions of constriction location and degree, this new gestural
system for vowels could be constructed using a set of one-dimensional tongue body raising
and/or lowering gestures, as well as a set of one-dimensional tongue body retraction and/or
advancement gestures. This new coordinate system for vowel gestures is illustrated in (251).
367
(251) Proposed new gestural representation of vowel place
A vowel, then, would be specified for height and backness according to its unique
combination of raising/lowering and advancement/retraction gestures of the tongue body. These
gestures would then only need to be specified for a single dimension, constriction degree (how
raised, how retracted, etc.). For instance, the vowel /u/ could be represented by the combination
of a tongue body raising gesture and a tongue body retraction gesture (as well as a lip protrusion
gesture responsible for rounding). The result of the concurrent activation of these lingual
gestures is a high back tongue body position. This is somewhat similar to proposals made within
particle theory (Schane 1984, 1990) and element theory (Kaye, Lowenstamm, & Vergnaud
1985), in which vowels are represented by combinations of subsegmental particles or elements.
A major advantage to this approach is in its ability to represent harmonies based on the
spreading height and backness. For instance, if a tongue body retraction gesture surfaces as either
persistent (non-self-deactivating) or anticipatory (early-activating), it will trigger backness
harmony. If a tongue body raising gesture surfaces as either persistent or anticipatory, it will
triggering height harmony.
Another advantage to this approach to vowel representation is its unification of gestural
types. In the current model, non-lingual gestures are based on the one-dimensional position of an
articulator. For instance, the velum is specified only for height/aperture. Even with this one-
368
dimensional representation of articulator position for a given gesture, it is still possible to
represent the possibility of two-dimensional articulator movement. Lip gestures, for instance,
may specify either the vertical distance between the lips, as for lip closure gestures, or for
horizontal position, as for lip protrusion and spreading. However, these distinct vertical and
horizontal lip positions are the results of different gestures; there is no single lip gesture with two
tract variables, one for the vertical dimension and one for the horizontal dimension. Similarly,
using this newly proposed coordinate system, both the horizontal and vertical position of the
tongue body can be specified; however, this position is the result of different gestures. This
model of vowel place would bring vowel gestures in line with many other non-lingual gestures
based on the one-dimensional position of an articulator.
Under this new model of vowel place, several questions remain and should be addressed
in further research. For one, it is not clear how to represent an unmarked vowel, i.e., a vowel
whose place corresponds to the neutral position of the tongue body. In a language with active
tongue body raising, a high vowel would be represented by a tongue body raising gesture, while
a nonhigh vowel would be represented by the absence of such a gesture. However, there must
still be some gestural element present that acts as the nucleus of a syllable; otherwise, onset and
coda consonants would have nothing to which to couple. Further work should explore what
exactly that nuclear gestural element is, whether it is a gesture specifying the neutral position of
the tongue body, a gesture that represents a syllable timing node, or something else entirely.
Another question that arises from the adoption of this new coordinate plane for vocalic
gestures concerns what, if anything, should be done with respect to the gestural representation of
lingual consonants. If vowels are represented by one or more gestures for tongue body height and
backness, the question arises as to whether consonants should be represented in the same way.
369
There has been a great deal of work conducted within featural phonology that suggests that
consonants and vowels should be represented by the same sets of features in order to account for
phenomena in which consonants and vowels interact with one another (see Clements & Hume
(1995) for a summary). It makes intuitive sense that the same logic could transfer to gestural
phonology, and that all types of gestures should be specified within a single coordinate system.
Otherwise, it is unclear how the result of blending between a consonantal and a vocalic gesture
should be calculated if the two gestures are specified along two completely different sets of
dimensions.
One possibility is to assume that while vowels are phonologically specified for height and
backness, the instructions that are sent to the articulators (following the calculation of
articulatory trajectories by the Task Dynamic Model) are formatted in the polar coordinate
system that is assumed within Articulatory Phonology. The primary drawback to such an
approach is the addition of what is essentially a phonetic implementation component that
performs the translation from phonological to phonetic specifications for vocalic gestures. While
such a phonetic implementation component is generally assumed within feature-based
phonological frameworks, one of the central tenets of Articulatory Phonology is that no such
component is necessary. In this framework, the phonological and phonetic units of representation
are identical. This insight would be lost if a phonetic implementation component were added to
translate the constriction parameters of vowels.
Alternatively, it could be the case that both consonantal and vocalic tongue body gestures
specify the target position of the tongue in terms of height and backness. Within this system,
consonants would be represented by tongue body raising/lowering and advancement/retraction
gestures whose constriction degrees result in either critical (fricative noise producing)
370
constriction degree or full closure of the vocal tract. Further research is necessary to determine
the viability of this proposal, as well as the full set of predictions that is made by such a system
of consonantal representation.
6.2.2 Intergestural Inhibition
Section 4.3 introduced the concept of intergestural inhibition as a mechanism for
accounting for the blocking of harmony. Its current implementation within the Gestural Harmony
Model is fairly uncomplicated. In the case of progressive harmony, an inhibiting gesture
completely and immediately deactivates an inhibited gesture in order to avoid the overlap of
incompatible gestures. In the case of regressive harmony, an inhibiting gesture prevents an
inhibited gesture from activating until immediately after the deactivation of the inhibiting
gesture. This is sufficient to capture basic cases of blocking by segments that are deemed by the
phonological grammar to be incompatible with a harmonizing property. However, there are
many harmony systems whose patterns of blocking do not fit this basic schema. Further research
should examine whether the mechanism of intergestural inhibition can be elaborated in order to
account for these more complex patterns of blocking. In particular, this research should focus on
questions of whether deactivation due to intergestural inhibition need always be complete and
immediate, or whether it can be partial and/or gradual.
As discussed in section 4.3, the Gestural Harmony Model’s representation of blocking as
intergestural inhibition is consistent both with a representation of gestural activation as involving
instantaneous activation and deactivation as well as ramped activation and deactivation. In a
traditional model of gestural activation, a gesture’s activation level goes from zero to one
instantaneously, and during deactivation, that activation level goes from one to zero
371
instantaneously. In a model of ramped gestural activation, the transition is not instantaneous, but
is still rapid. This is illustrated by the figure in (252), repeated from (133) in section 4.3.
(252) Persistent gesture (dashed) is fully inhibited and immediately deactivated by a following
blocking gesture (solid)
It is assumed that during one of these transition periods, a gesture’s activation level may
only take on a value between zero and one; a gesture cannot settle into an activation level other
than zero or one. However, doing away with this assumption may have interesting consequences
for the types of intergestural inhibition it is possible to represent within the Gestural Harmony
Model.
One issue worth examining is whether intergestural inhibition should always involve the
complete and immediate deactivation of an inhibited gesture, or whether gradience in the
strength of an inhibition relation, and in the resulting level of activation of an inhibited gesture,
should be admitted to the theory. Such gradient inhibition has several possible advantages. One
is the ability to account for harmony systems in which blockers of harmony partially take on the
harmonizing property. For instance, in Terena (Arawakan; Brazil), progressive (rightward) nasal
harmony targets vowels and glides and is blocked by obstruents (Bendor-Samuel 1960, 1966).
Interestingly, while these obstruents block nasal harmony, they also partially undergo it,
surfacing as prenasalized. A similar pattern is found in Epena Pedee (Harms 1985, 1994; Walker
1998/2000); its progressive nasal harmony targets vowels, glides, and liquids, while obstruents
372
block harmony and surface as prenasalized. Both of these languages present cases of nasal
harmony in which a blocker seemingly does not immediately deactivate a harmonizing gesture.
Within the Gestural Harmony Model, this could be accounted for if inhibition were not assumed
to always involve complete and immediate deactivation of a harmonizing gesture. While
deactivation of a persistent velum opening gesture begins when the activation of an obstruent
begins, this deactivation of the velum opening gesture would not be completed until well into the
period of activation of the obstruent. The figure in (253) demonstrates.
(253) Persistent gesture (dashed) is gradiently inhibited and gradually deactivated by a
following blocking/partially undergoing gesture (solid)
If gestural activation levels and the strengths of the inhibition relations between them are
more gradient than previously assumed, such cases could potentially be provided with a
straightforward analysis. It may also be possible to leverage the idea of gradual deactivation to
capture what Jurgec (2011) refers to as ‘icy targets,’ those segments that fully undergo a
harmony process but halt the further spread of a harmonizing property. Within the Gestural
Harmony Model, icy targets could represent a special case of gradual inhibition in which the
deactivation of a harmonizing gesture occurs at such a slow rate that it overlaps a blocking
gesture entirely. This is illustrated in the figure in (254).
373
(254) Persistent gesture (dashed) is gradiently inhibited and gradually deactivated by a
following ‘icy target’ gesture (solid)
The ability for gestural activation levels to take on values other than zero or one would
also have implications for harmony patterns in which it appears that blocking occurs only if two
potential blocking conditions are present. For instance, in Tuyuca (section 4.4.2; Barnes &
Takagi de Silzer (1976), Barnes (1996)), obstruents do not block nasal harmony within a
morpheme, but do block nasal harmony when in the initial position of a suffix that follows a
harmony-triggering root. This suggests that while obstruents cannot block nasal harmony on
their own, the combination of an obstruent and a morpheme boundary is enough to block nasal
harmony. Similarly, tongue root harmony in Lango (Western Nilotic; Uganda; Noonan (1992),
Archangeli & Pulleyblank (1994), Smolensky (2006)) displays complex patterns of blocking
based on multiple sources of markedness. As part of this complex pattern, ATR harmony is
blocked when it would proceed across a geminate consonant and also target a nonhigh vowel.
However, geminate consonants and nonhigh vowels do not block harmony on their own.
If intergestural inhibition relations are allowed to take on intermediate degrees of
strength, these more complex patterns of blocking can be accounted for within the Gestural
Harmony Model. Taking the case of Tuyuca nasal harmony as an example, the account would be
roughly as follows. In Tuyuca, a persistent (non-self-deactivating) velum opening gesture is
inhibited by obstruents; however, this inhibition is too weak to actually deactivate the velum
374
opening gesture. As a result, obstruents do not act as blockers of within-morpheme nasal
harmony. A persistent (non-self-deactivating) velum opening gesture is also inhibited by a
following suffix. There are a number of ways that this could be implemented. The inhibition
could come from the gestures in the segment at the edge of the morpheme. Alternatively, this
inhibition could come from a gesture corresponding to a higher organizational node for the
morpheme as a whole. For some suffixes, this inhibition is strong enough to deactivate the velum
opening gesture, and the suffix surfaces as a fixed oral suffix. For other suffixes, this inhibition is
not strong enough, and the suffix undergoes nasal harmony rather than blocking it. However,
when a suffix begins with an obstruent, it will always block nasal harmony due to the combined
effect of two inhibition relations (one from the morpheme boundary, one from the obstruent)
working to deactivate the harmonizing velum opening gesture. Under this analysis, gradient
intergestural inhibition would account for the somewhat complex patterning of fixed and varying
suffixes in Tuyuca nasal harmony.
While incorporating gradience of gestural activation and intergestural inhibition relations
into the Gestural Harmony Model would make it possible to account for more complicated
patterns of blocking than those addressed in chapter 4, this gradience also introduces some
significant theoretical questions. Chief among these is what it means for a gesture to be partially
active, both to the phonological grammar and to the Task Dynamic Model of speech production.
It is currently unclear whether a partially active gesture should count as present in a phonological
form, and whether it would be fully visible to constraints in the grammar. From the perspective
of speech production, it is also unclear whether a partially active gesture would only command
the vocal tract to an intermediate degree, thus affecting the achievement of its target articulatory
state.
375
One way of addressing these issues is to incorporate some kind of thresholding
mechanism for gestural activation, as in the model of gestural activation proposed by Tilsen
(2013, 2016). In Tilsen’s Selection-Coordination Model, gestures are gradiently active at some
abstract level of representation, but are categorically present within a gestural score during the
period in time in which their activation levels are greater than some threshold value. This ability
to represent gestural activation as both gradient and categorical could be a useful addition to the
Gestural Harmony Model’s inhibition mechanism if it were elaborated to include gradient
activation. It is also possible that Smolensky & Goldrick’s (2016) Gradient Symbolic
Computation framework, in which input elements may have gradient activation levels but output
elements must be categorically present or absent, could provide some useful insights in this area.
Their framework could prove especially useful in addressing questions concerning how the
phonological grammar references gradiently active gestures, and how it manipulates this
gradience between underlying and surface forms.
The discussion of blocking of nasal harmony by certain suffixes in Tuyuca brings up
another issue regarding patterns of blocking and the representation of intergestural inhibition in
the Gestural Harmony Model. In many harmony systems, morphemes can be classified as being
either fixed or alternating for a harmonizing property. While alternating morphemes undergo
harmony when attached to another morpheme containing a trigger of harmony, fixed or invariant
morphemes do not undergo harmony. This is the case in several nasal harmony systems,
including Tuyuca nasal harmony (section 4.4.2; Barnes & Takagi de Silzer (1976), Barnes
(1996)). Fixed morphemes are also commonly found among tongue root harmony systems,
including Nandi ATR harmony (sections 2.2.3 and 3.2.2; Creider & Creider (1989)). In these
376
tongue root harmony systems, morphemes may have a fixed ATR value, a fixed RTR value, or
an alternating value for tongue root position.
If the presence of intergestural inhibition relations in surface forms is motivated solely by
markedness constraints such as *OVERLAP, there is no clear way to account for the existence of
morphemes that appear to idiosyncratically block harmony. To address this, it may be
advantageous to redefine the mechanism of intergestural inhibition even further by recasting a
gesture’s ability to block harmony as a property of that gesture rather than as the result of an
intergestural relation. Under this approach, deactivators would be represented as phonological
objects that are associated with certain gestures. A deactivator could be conceived of as an object
that is somehow affiliated with a gesture; for instance, the velum closure gesture of an obstruent
that blocks nasal harmony could include a deactivator for a velum opening gesture.
Alternatively, it is possible that the gestures themselves could serve as deactivators of certain
other gestural types. In either case, casting a gestural deactivator as a type of phonological entity
would allow it to appear not only in surface forms as dictated by the phonological grammar, but
also in underlying forms as part of the phonological specification of a lexical item. As a result,
these deactivators would be able to serve a contrastive function and provide an account for
idiosyncratic blocking of harmony. In addition, casting gestures as being specified to deactivate
certain other types of gestures, rather than relying on *OVERLAP constraints to motivate
intergestural inhibition, could possibly address the Gestural Harmony Model’s issue with
generating sour grapes patterns of harmony, discussed in section 4.7.4.
Under the gestural deactivator approach, the question remains as to whether contrastive
deactivation should be permitted at the level of the segment/gesture or at the level of the
morpheme. If the ability of a gesture to serve as a deactivator of another gesture is encoded
377
within its representation, then this predicts that this contrastive blocking may occur at the
segment level. However, it is unclear whether this is supported by crosslinguistic typology.
While morpheme-level idiosyncratic blocking of harmony is attested, examples of idiosyncratic
blocking at the level of the segment are elusive. Such analyses have been suggested for some
harmony patterns; perhaps the most well-known example is Hungarian backness harmony (see
Törkenczy (2011) for an overview). However, it is possible that these harmony systems instead
represent cases of contrastive triggering, similar to those discussed in sections 3.3 and 3.5, rather
than contrastive blocking. For instance, rather than analyzing Hungarian as exhibiting backness
harmony with front vowels that idiosyncratically block harmony, this pattern could instead be
analyzed as a case of frontness harmony in which front vowels idiosyncratically trigger harmony.
With the development of a new set of vowel representations to include the representations of
vowel height and backness, as discussed in the previous section, such cases should be examined
thoroughly in order to determine the optimal way in which to analyze them.
In any case, the idea of inhibition as the product of gestural activation rather than
intergestural relations is an intriguing one that should be examined further. It provides an
appealing parallel to the account of idiosyncratic harmony triggering via gestural parameter
presented in sections 3.3 and 3.5, which has already been demonstrated to provide significant
advantages in accounting for sometimes complex patterns exhibited by harmony patterns.
6.2.3 Directionality of Harmony
The current implementation of gestural duration extension in the Gestural Harmony
Model includes no directional asymmetries. Progressive harmony is the result of gestural
persistence, by which a gesture does not self-deactivate when it reaches its target articulatory
state. Regressive harmony is the result of gestural anticipation, by which a gesture activates
378
before the starting 0º phase of its clock. Further work within the Gestural Harmony Model
should focus on whether and how the model should incorporate any asymmetries relating to the
representation of progressive and regressive harmony.
Claims about directional asymmetries in harmony are numerous and sometimes
inconsistent. Kaun (1995) claims that the directionality of a harmony system is predictable based
on which positions in a word a harmony-triggering segment may surface. Baković (2000) makes
a similar claim that directionality is directly predictable from the morphological structure of a
language. While these generalizations seem to hold true of the rounding harmonies of Altaic
languages that make up Kaun’s survey and the Niger-Congo and Nilo-Saharan languages that
make up Bakovic’s survey, these claims do not seem to hold more generally. Hyman (2002)
picks up on this and suggests that where morphology does not determine the direction of
harmony, there is a regressive (leftward) bias. Again, however, this claim does not appear to
generalize to different types of harmony.
When examining crosslinguistic tendencies in the directionality of harmony, the
following observations can be made. The typological patterns of directionality in harmony vary
by the harmonizing element, by language family, and by area. While some types of harmony do
appear to show a regressive (leftward) bias (e.g., post-velar vowel-consonant harmonies), many
others (e.g., nasal harmony, rounding harmony, backness harmony) show a bias toward
progressive (rightward) harmony. A successful theory of harmony should be able to account for
these biases, and their possibly articulator-specific nature.
If there is a directional bias, whether across all types of harmony or within a given type of
harmony, further research should work to determine its source. If harmony is driven by a desire
to maximize the period of activation of a gesture such that its perceptibility is maximized, then a
379
difference in the attestation of progressive versus regressive harmony could be based in an
asymmetry in the perceptual payoff of extending in one direction or the other. For instance, if
regressive harmony is less prevalent, it could be the case that anticipatory gestures do not involve
simple early activation, but should instead be modeled via some kind of gestural stretching in
which a gesture activates earlier than its 0º phase, but reaches its target articulatory state at the
same time it would have as a typical gesture. This could be achieved by assuming that the early
activation of a gesture is automatically accompanied by a lowering of gestural stiffness such that
the time point at which it achieves its target articulatory state is stable, regardless of the time
point at which the gesture activates. This is shown in the figure in (255), using velum opening
gestures as an example.
(255) Possible alternative representation of gestural anticipation as stretching
Alternatively, if progressive harmony is less prevalent, it could be the case that persistent
gestures do not involve simply staying active as long as they are able, but instead represent a
gradual deactivation. This could be modeled by some kind of decay parameter that determines
the rate at which a gesture’s activation reaches zero. While a typical gesture would have a
relatively high decay rate, and would deactivate simultaneously with the achievement of its
380
target articulatory state, a persistent gesture could be modeled as a gesture with a low decay rate
that causes it to remain active for a longer period of time. This is illustrated in the figure in (256).
(256) Possible alternative representation of gestural persistence as gradual decay
It could also be the case that the gestural representations of harmony in the progressive
and regressive directions vary across different types of harmony. This is a matter for further
study. This study will need to explore the full range of asymmetries that might arise between
progressive and regressive harmony. These could include asymmetries in blocking, in which
harmony is blocked in one direction and not the other. These could also include asymmetries in
triggering, potentially affecting conditional triggering in one direction and not the other.
381
References
Ajíbóyè, Ọládiípọ
̀ . (2001). Nasalization in Mọ
̀ bà. In S. Oh, N. Sawai, K. Shiobara, & R. Wojdak
(Eds.), Proceedings of the Northwest Linguistics Conference 2001 (pp. 1–18). Vancouver:
University of British Columbia Working Papers in Linguistics Volume 8.
Ajíbóyè, Ọládiípọ
̀ , & Pulleyblank, Douglas. (2008). Mọ
̀ bà Nasal Harmony. unpublished
manuscript, University of Lagos and University of British Columbia.
Akinlabi, Akinbiyi. (1994). Alignment Constraints in ATR Harmony. Studies in the Linguistic
Sciences, 24(1/2), 1–18.
Anderson, Gregory D. S. (1998). Historical Aspects of Yakut (Saxa) Phonology. Turkic
Languages, 2, 1–32.
Anderson, Stephen R. (1976a). On the Description of Consonant Gradation in Fula. Studies in
African Linguistics, 7(1), 93–136.
Anderson, Stephen R. (1976b). On the description of multiply-articulated consonants. Journal of
Phonetics, 4, 17–27.
Archangeli, Diana, & Pulleyblank, Douglas. (1989). Yoruba Vowel Harmony. Linguistic inquiry,
20(2), 173–217.
Archangeli, Diana, & Pulleyblank, Douglas. (1994). Grounded Phonology. MIT Press.
Archangeli, Diana, & Pulleyblank, Douglas. (2007). Harmony. In P. de Lacy (Ed.), The Cambridge
Handbook of Phonology (pp. 353–378). Cambridge University Press.
Archangeli, Diana, & Suzuki, Keiichiro. (1995). Menomini Vowel Harmony: O(pacity) &
T(ransparency) in OT. In K. Suzuki & D. Elzinga (Eds.), South Western Optimality Theory
Workshop 1995 (pp. 1–17). University of Arizona.
Avery, Peter, & Rice, Keren. (1989). Segment structure and coronal underspecification.
Phonology, 6, 179–200.
Baković, Eric. (2000). Harmony, Dominance and Control. Ph.D. Dissertation, Rutgers, The State
University of New Jersey.
Bakovic, Eric, & Wilson, Colin. (2000). Transparency, Strict Locality, and Targeted Constraints.
In Billerey & Lillehaugen (Eds.), Proceedings of the Nineteenth West Coast Conference on
Formal Linguistics (WCCFL 19) (pp. 43–56). Somerville, MA: Cascadilla Press.
Barnes, Janet. (1996). Autosegments with Three-Way Lexical Contrasts in Tuyuca. International
Journal of American Linguistics, 62(1), 31–58.
Barnes, Janet, & Takagi de Silzer, Sheryl. (1976). Fonología del tuyuca. In Sistemas fonológicos
de idiomas colombianos, Tomo III (pp. 123–137). Lomalinda: Editorial Townsend.
Becker, Michael. (2009). Phonological Trends in the Lexicon: the Role of Constraints. Ph.D.
Dissertation, University of Massachusetts Amherst.
Beckman, Jill N. (1997). Positional faithfulness, positional neutralisation and Shona vowel
harmony. Phonology, 14, 1–46.
Beckman, Jill N. (1998). Positional Faithfulness. Ph.D. Dissertation, University of Massachusetts
Amherst.
Bell-Berti, Fredericka. (1976). An Electromyographic Study of Velopharyngeal Function in
Speech. Journal of Speech and Hearing Research, 19(2), 225–240.
Bell-Berti, Fredericka, & Hirose, Hajime. (1975). Palatal activity in voicing distinctions: a
simultaneous fiberoptic and electromyographic study. Journal of Phonetics, 3, 69–74.
Bendor-Samuel, John T. (1960). Some Problems of Segmentation in the Phonological Analysis of
Tereno. Word, 16, 348–355.
382
Bendor-Samuel, John T. (1966). Some Prosodic Features in Terena. In C. E. Bazell, J. C. Catford,
M. A. K. Halliday, & R. H. Robins (Eds.), In Memory of J.R. Firth (pp. 30–39). London:
Longmans.
Benus, Stefan, & Gafos, Adamantios I. (2007). Articulatory characteristics of Hungarian
“transparent” vowels. Journal of Phonetics, 35, 271–300.
Bessell, Nicola J. (1998). Local and non-local consonant-vowel interaction in Interior Salish.
Phonology, 15, 1–40.
Blumenfeld, Lev, & Toivonen, Ida. (2016). A featural paradox in Votic harmony. Natural
Language and Linguistic Theory, 34(4), 1167–1180.
Blust, Robert A. (1972). Sketches of the Morphology and Phonology of Bornean Languages 1:
Uma Juman (Kayan) (No. 33). Pacific Linguistics.
Boersma, Paul. (1998). Functional Phonology: Formalizing the interactions between articulatory
and perceptual drives. Ph.D. Dissertation, University of Amsterdam.
Boersma, Paul. (2003). Nasal harmony in functional phonology. In J. van de Weijer, V. J. van
Heuven, & H. van der Hulst (Eds.), The Phonological Spectrum Volume I: Segmental Structure
(pp. 3–35). Amsterdam: John Benjamins Publishing Company.
Botma, Bert. (2004). Phonological Aspects of Nasality: An Element-Based Dependency Approach.
Ph.D. Dissertation, University of Amsterdam.
Botma, Bert, & Smith, Norval. (2007). A dependency-based typology of nasalisation and voicing
phenomena. In Linguistics in the Netherlands 2007 (pp. 36–48).
Boyce, Suzanne E. (1990). Coarticulatory organization for lip rounding in Turkish and English.
Journal of the Acoustical Society of America, 88(6), 2584–2595.
Bradley, Travis G. (2005). Systemic Markedness and Phonetic Detail in Phonology. In R. S. Gess
& E. J. Rubin (Eds.), Theoretical and Experimental Approaches to Romance Linguistics (pp.
41–62). Amsterdam: John Benjamins Publishing Company.
Browman, Catherine P., & Goldstein, Louis. (1986). Towards an Articulatory Phonology.
Phonology Yearbook, 3, 219–252.
Browman, Catherine P., & Goldstein, Louis. (1989). Articulatory gestures as phonological units.
Phonology, 6(2), 201–251.
Browman, Catherine P., & Goldstein, Louis. (1992). Articulatory Phonology: An Overview.
Phonetica, 49, 155–180.
Browman, Catherine P., & Goldstein, Louis. (2000). Competing constraints on intergestural
coordination and self-organization of phonological structures. Bulletin de la Communication
Parlée, 5, 25–34.
Buckley, Eugene. (1994). Default feature prespecification: the two [i]’s of Kashaya. In
Proceedings of the 24th Annual Meeting of the North East Linguistic Society (pp. 17–30).
Amherst, MA: Graduate Linguistic Student Association.
Byrd, Dani. (1996). A phase window framework for articulatory timing. Phonology, 13, 139–169.
Byrd, Dani, & Saltzman, Elliot. (2003). The elastic phrase: modeling the dynamics of boundary-
adjacent lengthening. Journal of Phonetics, 31, 149–180.
Casali, Roderic F. (2008). ATR Harmony in African Languages. Language and Linguistics
Compass, 2(3), 496–549.
Chamora, Berhanu, & Hetzron, Robert. (2000). Inor. Lincom Europa.
Chomsky, Noam, & Halle, Morris. (1968). The Sound Pattern of English. New York: Harper &
Row.
383
Clements, George N. (1976a). Vowel Harmony in Nonlinear Generative Phonology: An
Autosegmental Model. Unpublished manuscript, Harvard University.
Clements, G. N. (1976b). Neutral Vowels in Hungarian Vowel Harmony: an Autosegmental
Interpretation. In Proceedings of the 7th Annual Meeting of the North East Linguistic Society
(pp. 49–64).
Clements, George N. (1981). Akan Vowel Harmony: a Nonlinear Analysis. Harvard Studies in
Phonology, 2, 108–177.
Clements, G. N. (1985). The geometry of phonological features. Phonology Yearbook, 2, 225–
252.
Clements, G. N., & Hume, Elizabeth V. (1995). The Internal Organization of Speech Sounds. In
J. Goldsmith (Ed.), Handbook of Phonology (pp. 245–306). Blackwell Publishing.
Clements, G. N., & Sezer, Engin. (1982). Vowel and consonant disharmony in Turkish. In H. van
der Hulst & N. Smith (Eds.), The Structure of Phonological Representations (Part II) (pp. 213–
255). Dordrecht: Foris Publications.
Coady, James, & McGinn, Richard. (1982). On the So-Called Implosive Nasals of Rejang. In R.
Carle, M. Heinschke, P. W. Pink, C. Rost, & K. Stadtlander (Eds.), Gava’: Studies in
Austronesian Languages and Cultures (pp. 437–449). Berlin: Dietrich Reimer Verlag.
Cohn, Abigail C. (1990). Phonetic and Phonological Rules of Nasalization. Ph.D. Dissertation,
University of California Los Angeles.
Cohn, Abigail C. (1993a). A survey of the phonology of the feature [±nasal]. Working Papers of
the Cornell Phonetics Laboratory (Vol. 8).
Cohn, Abigail C. (1993b). The Status of Nasalized Continuants. In M. K. Huffman & R. A.
Krakow (Eds.), Phonetics and Phonology 5: Nasals, Nasalization, and the Velum (pp. 329–
367). Academic Press.
Cole, Jennifer. (1987). Planar Phonology and Morphology. Ph.D. Dissertation, Massachusetts
Institute of Technology.
Cole, Jennifer, & Kisseberth, Charles. (1994). An Optimal Domains Theory of Harmony. Studies
in the Linguistic Sciences, 24(2), 101–114.
Cole, Jennifer, & Kisseberth, Charles. (1995). Nasal Harmony in Optimal Domains Theory. In
Proceedings of the Western Conference on Linguistics (pp. 44–58).
Cole, Jennifer, & Trigo, R. Lorenza. (1988). Parasitic Harmony. In Features, Segmental Structure
and Harmony Processes (Part II).
Comrie, Bernard. (1981). The Languages of the Soviet Union. Cambridge: Cambridge University
Press.
Court, Christopher. (1970). Nasal Harmony and Some Indonesian Sound Laws. In S. A. Wurm &
D. C. Laycock (Eds.), Pacific Linguistic Studies in Honour of Arthur Capell (pp. 203–217).
Canberra: Australian National University.
Cowan, H. K. J. (1981). An Outline of Achehnese Phonology and Morphology. Bulletin of the
School of Oriental and African Studies, 44(3), 522–549.
Cowell, Andrew, & Moss, Alonzo. (2008). The Arapaho Language. Boulder, CO: University Press
of Colorado.
Creider, Chet A., & Creider, Jane Tapsubei. (1989). A Grammar of Nandi. Hamburg: Buske.
Danis, Nicholas. (2017). Complex Place and Place Identity. Ph.D. Dissertation, Rutgers, the State
University of New Jersey.
384
Davidson, Lisa. (2003). The Atoms of Phonological Representation: Gestures, Coordination and
Perceptual Features in Consonant Cluster Phonotactics. Ph.D. Dissertation, The Johns
Hopkins University.
Davis, Stuart. (1995). Emphasis Spread in Arabic and Grounded Phonology. Linguistic Inquiry,
26(3), 465–498.
de Lacy, Paul. (2002). The Formal Expression of Markedness. Ph.D. Dissertation, University of
Massachusetts Amherst.
Delattre, Pierre, & Freeman, Donald C. (1968). A Dialect Study of American R’s by X-Ray Motion
Picture. Linguistics, 44, 29–68.
Doak, Ivy G. (1992). Another look at Coeur d’Alene harmony. International Journal of American
Linguistics, 58(1), 1–35.
Dresher, B. Elan, & Nevins, Andrew. (2017). Conditions on Iterative Rounding Harmony in
Oroqen. Transactions of the Philological Society, 115(3), 365–394.
Durie, Mark. (1985). A Grammar of Acehnese on the Basis of a Dialect of North Aceh. Dordrecht:
Foris Publications.
Eisner, Jason. (2000). Directional Constraint Evaluation in Optimality Theory. In Proceedings of
the 18th International Conference on Computational Linguistics.
Essegbey, James, & McCollum, Adam. (2017). Initial prominence and progressive harmony in
Tutrugbu. Unpublished manuscript, University of Florida and University of California San
Diego.
Finley, Sara. (2008). Formal and Cognitive Restrictions on Vowel Harmony. Ph.D. Dissertation,
The Johns Hopkins University.
Finley, Sara. (2010). Exceptions in vowel harmony are local. Lingua, 120(6), 1549–1566.
Flack, Kathryn. (2008). Templatic Morphology and Indexed Markedness Constraints. Linguistic
Inquiry, 38, 749–758.
Flemming, Edward. (1995). Auditory Representations in Phonology. Ph.D. Dissertation,
University of California Los Angeles.
Flemming, Edward. (2004). Contrast and perceptual distinctiveness. In B. Hayes, R. Kirchner, &
D. Steriade (Eds.), Phonetically Based Phonology (pp. 232–276). Cambridge University Press.
Gafos, Adamantios. (1996). The Articulatory Basis of Locality in Phonology. Ph.D. Dissertation,
The Johns Hopkins University. Published 1999.
Gafos, Adamantios I. (2002). A Grammar of Gestural Coordination. Natural Language and
Linguistic Theory, 20(2), 269–337.
Gafos, Adamantios I., & Beňuš, Štefan. (2006). Dynamics of Phonological Cognition. Cognitive
Science, 30, 905–943.
Gallagher, Gillian. (2010). The Perceptual Basis of Long-Distance Laryngeal Restrictions. Ph.D.
Dissertation, Massachusetts Institute of Technology.
Gerfen, Chip. (1999). Phonology and Phonetics in Coatzospan Mixtec. Dordrecht: Kluwer
Academic Publishers.
Gerfen, Chip. (2001). Nasalized Fricatives in Coatzospan Mixtec. International Journal of
American Linguistics, 67(4), 449–466.
Gick, Bryan, Iskarous, Khalil, Whalen, D. H., & Goldstein, Louis. (2003). Constraints on Variation
in the Production of English /r/. In S. Palethorpe & M. Tabain (Eds.), Proceedings of the 6th
International Seminar on Speech Production (pp. 73–78).
Gick, Bryan, Pulleyblank, Douglas, Campbell, Fiona, & Mutaka, Ngessimo. (2006). Low vowels
and transparency in Kinande vowel harmony. Phonology, 23, 1–20.
385
Goldsmith, John A. (1976). Autosegmental Phonology. Ph.D. Dissertation, Massachusetts Institute
of Technology.
Goldstein, Louis. (1991). Lip Rounding as Side Contact. In Proceedings of the 12th International
Congress of Phonetic Sciences (pp. 97–101). Aix-en-Provence: Université de Provence.
Gregores, Emma, & Suárez, Jorge A. (1967). A Description of Colloquial Guaraní. The Hague:
Mouton.
Hadding, Kerstin, Hirose, Hajime, & Harris, Katherine S. (1976). Facial muscle activity in the
production of Swedish vowels: an electromyographic study. Journal of Phonetics, 4, 233–245.
Hall, Nancy. (2003). Gestures and Segments: Vowel Intrusion as Overlap. Ph.D. Dissertation,
University of Massachusetts Amherst.
Hall, Nancy. (2006). Cross-linguistic patterns of vowel intrusion. Phonology, 23, 387–429.
Hall, T. A. (2007). Comparative Markedness vs. Standard Faithfulness Theory: A typological
comparison. Unpublished manuscript, Indiana University.
Hansson, Gunnar. (2001). Theoretical and Typological Issues in Consonant Harmony. Ph.D.
Dissertation, University of California Berkeley. Published in 2010 by University of California
Press.
Harms, Philip L. (1985). Epena Pedee (Saija): Nasalization. In R. M. Brend (Ed.), From Phonology
to Discourse: Studies in Six Colombian Languages.
Harms, Phillip Lee. (1994). Epena Pedee Syntax. Studies in the Languages of Colombia 4. Summer
Institute of Linguistics and the University of Texas at Arlington.
Harrison, Kevin David. (2000). Topics in the Phonology and Morphology of Tuvan. Ph.D.
Dissertation, Yale University.
Hayes, Bruce, & Londe, Zsuzsa Cziráky. (2006). Stochastic phonological knowledge: the case of
Hungarian vowel harmony. Phonology, 23, 59–104.
Herzallah, Rukayyah S. (1990). Aspects of Palestinian Arabic phonology: A nonlinear approach.
Ph.D. Dissertation, Cornell University.
Hong, Sung-Hoon. (1994). Issues in round harmony: Grounding, identity and their interaction.
Ph.D. Dissertation, The University of Arizona.
Hualde, Jose Ignacio. (1989). Autosegmental and metrical spreading in the vowel-harmony
systems of northwestern Spain. Linguistics, 27(5), 773–805.
Hume, Elizabeth, & Odden, David. (1996). Reconsidering [consonantal]. Phonology, 13, 345–376.
Hyman, Larry M. (1994). Cyclic Phonology and Morphology in Cibemba. In J. Cole & C.
Kisseberth (Eds.), Perspectives in Phonology (pp. 81–112). CSLI Publications.
Hyman, Larry M. (1998). Positional prominence and the “prosodic trough” in Yaka. Phonology,
15, 41–75.
Hyman, Larry M. (2002). Is There a Right-to-Left Bias in Vowel Harmony? In 9th International
Phonology Meeting.
Inkelas, Sharon. (1994). The Consequences of Optimization for Underspecification. In J. Beckman
(Ed.), Proceedings of the 25th Annual Meeting of the North East Linguistic Society (pp. 287–
302). Amherst, MA: Graduate Linguistic Student Association.
Inkelas, Sharon, Orgun, C. Orhan, & Zoll, Cheryl. (1997). The Implications of Lexical Exceptions
for the Nature of Grammar. In I. Roca (Ed.), Constraints and Derivations in Phonology (pp.
393–418). New York: Oxford University Press.
Iskarous, Khalil, McDonough, Joyce, & Whalen, D. H. (2012). A gestural account of the velar
fricative in Navajo. Laboratory Phonology, 3, 195–210.
386
Itô, Junko, & Mester, Armin. (1998). Markedness and word structure: OCP effects in Japanese.
Unpublished manuscript, University of California Santa Cruz.
Itô, Junko, Mester, Armin, & Padgett, Jaye. (1995). Licensing and Underspecification in
Optimality Theory. Linguistic Inquiry, 26(4), 571–613.
Jakobson, Roman, Fant, C. Gunnar M., & Halle, Morris. (1952). Preliminaries to Speech Analysis:
The Distinctive Features and Their Correlates. Cambridge, MA: MIT Press.
Jurgec, Peter. (2011). Feature Spreading 2.0: A Unified Theory of Assimilation. Ph.D.
Dissertation, University of Tromsø.
Kaplan, Aaron F. (2008). Noniterativity is an Emergent Property of Grammar. Ph.D. Dissertation,
University of California Santa Cruz.
Kaplan, Lawrence D. (1981). Phonological Issues in North Alaskan Inupiaq. Alaska Native
Langage Center Research Papers No. 6Alaska Native Language Center Research Papers (Vol.
6). Fairbanks, Alaska: Alaska Native Langauge Center.
Kaun, Abigail. (1995). The Typology of Rounding Harmony: An Optimality Theoretic Approach.
Ph.D. Dissertation, University of California Los Angeles.
Kaun, Abigail R. (2004). The typology of rounding harmony. In B. Hayes, R. Kirchner, & D.
Steriade (Eds.), Phonetically Based Phonology (pp. 87–116). Cambridge University Press.
Kaye, Jonathan, Lowenstamm, Jean, & Vergnaud, Jean-Roger. (1985). The internal structure of
phonological elements: a theory of charm and government. Phonology Yearbook, 2(1985),
305–328.
Keating, Patricia A. (1988). Underspecification in phonetics. Phonology, 5, 275–292.
Keating, Patricia A. (1990). The window model of coarticulation: articulatory evidence. In J.
Kingston & M. E. Beckman (Eds.), Papers in Laboratory Phonology I: Between the Grammar
and the Physics of Speech (pp. 451–470). Cambridge University Press.
Kimper, Wendell A. (2011). Competing Triggers: Transparency and Opacity in Vowel Harmony.
Ph.D. Dissertation, University of Massachusetts Amherst.
Kirchner, Robert. (1993). Turkish Vowel Harmony and Disharmony: An Optimality Theoretic
Account. In Rutgers Optimality Workshop I (ROW-I).
Krämer, Martin. (2003). Vowel Harmony and Correspondence Theory. Studies in Generative
Grammar 66. Berlin: Mouton de Gruyter.
Krueger, John R. (1962). Yakut Manual. Uralic and Altaic Series. Indiana University Publications.
Kutsch Lojenga, Constance. (1994). Kibudu a bantu language with nine vowels. Africana
Linguistica, 11, 127–133.
Kutsch Lojenga, Constance, & Waag, Christine. (2004). The Sounds and Tones of Fur. Occasional
Papers in the Study of Sudanese Languages, 9, 1–25.
Ladefoged, Peter. (1990). On dividing phonetics and phonology: comments on the papers by
Clements and by Browman and Goldstein. In J. Kingston & M. Beckman (Eds.), Papers in
Laboratory Phonology I: Between the Grammar and the Physics of Speech (pp. 398–405).
Cambridge University Press.
Legendre, Géraldine, Miyata, Yoshiro, & Smolensky, Paul. (1990). Can Connectionism
Contribute to Syntax? Harmonic Grammar, with an Application. In Proceedings of the 30th
Meeting of the Chicago Linguistics Society.
Li, Bing. (1996). Tungusic Vowel Harmony: Description and Analysis. Ph.D. Dissertation,
Universiteit van Amsterdam.
Lindau, Mona. (1985). The Story of /r/. In V. A. Fromkin (Ed.), Phonetic Linguistics: Essays in
Honor of Peter Ladefoged (pp. 157–168). Orlando: Academic Press.
387
Lindblom, Björn E. F., & Sundberg, Johan E. F. (1971). Acoustical Consequences of Lip, Tongue,
Jaw, and Larynx Movement. Journal of the Acoustical Society of America, 50(4), 1166–1179.
Linker, Wendy. (1982). Articulatory and Acoustic Correlates of Labial Activity in Vowels: A
Cross-linguistic Study. Ph.D. Dissertation, Working Papers in Phonetics.
Local, John, & Lodge, Ken. (2004). Some auditory and acoustic observations on the phonetics of
[ATR] harmony in a speaker of a dialect of Kalenjin. Journal of the International Phonetic
Association, 34(1), 1–16.
Lodge, Ken. (1995). Kalenjin phonology and morphology: A further exemplification of
underspecification and non-destructive phonology. Lingua, 96, 29–43.
Loos, Eugene Emil. (1967). The Phonology of Capanahua and its Grammatical Basis. Ph.D.
Dissertation, University of Texas Austin. Published 1969.
Lubker, James F. (1968). An Electromyographic-Cinefluorographic Investigation of Velar
Function During Normal Speech Production. Cleft Palate Journal, 5, 1–18.
Łubowicz, Ania. (2003). Local conjunction and comparative markedness. Theoretical Linguistics,
29, 101–112.
Mahanta, Shakuntala. (2007). Directionality and Locality in Vowel Harmony. Ph.D. Dissertation,
Universiteit Utrecht.
Mahanta, Shakuntala. (2009). Sequential Markedness Constraints and Iterativity. In A. Schardl,
M. Walkow, & M. Abdurrahman (Eds.), Proceedings of the 38th Annual Meeting of the North
East Linguistic Society (pp. 111–122). Graduate Linguistic Student Association.
McCarthy, John J. (1999). Sympathy and phonological opacity. Phonology, 16, 331–399.
McCarthy, John J. (2000). Harmonic Serialism and Parallelism. In M. Hirotani (Ed.), Proceedings
of the Thirtieth Annual Meeting of the North East Linguistic Society (pp. 501–524).
McCarthy, John J. (2002). On targeted constraints and cluster simplification. Phonology, 19, 273–
292.
McCarthy, John J. (2003a). OT constraints are categorical. Phonology, 20, 75–138.
McCarthy, John J. (2003b). Comparative markedness. Theoretical Linguistics, 29, 1–51.
McCarthy, John J. (2004). Headed spans and autosegmental spreading. unpublished manuscript,
University of Massachusetts Amherst.
McCarthy, John J. (2008a). The gradual path to cluster simplification. Phonology, 25, 271–319.
McCarthy, John J. (2008b). The serial interaction of stress and syncope. Natural Language and
Linguistic Theory, 26, 499–546.
McCarthy, John J., & Pater, Joe (Eds.). (2016). Harmonic Grammar and Harmonic Serialism.
Equinox.
McCarthy, John J., & Prince, Alan S. (1995). Faithfulness and Reduplicative Identity (No. 18).
McGinn, Richard. (1982). Outline of Rejang Syntax. Jakarta: Badan Penyelenggara Seri NUSA.
Mullin, Kevin. (2011). Strength in Harmony Systems: Trigger and Directional Asymmetries.
Unpublished manuscript, University of Massachusetts Amherst.
Nam, Hosung. (2007). Syllable-level intergestural timing model: Split-gesture dynamics focusing
on positional asymmetry and moraic structure. In J. Cole & J. I. Hualde (Eds.), Papers in
Laboratory Phonology 9 (pp. 483–506). Berlin: Walter de Gruyter.
Nam, Hosung, Goldstein, Louis, Saltzman, Elliot, & Byrd, Dani. (2004). TADA: An enhanced,
portable Task Dynamics model in MATLAB. Journal of the Acoustical Society of America,
115, 2430.
388
Nam, Hosung, & Saltzman, Elliot. (2003). A Competitive, Coupled Oscillator Model of Syllable
Structure. In Proceedings of the 15th International Congress of Phonetic Sciences. Barcelona,
Spain.
Ní Chiosáin, Máire, & Padgett, Jaye. (1997). Markedness, Segment Realisation, and Locality in
Spreading. Report No. LRC-97-01.
Ní Chiosáin, Máire, & Padgett, Jaye. (2001). Markedness, Segment Realization, and Locality in
Spreading. In L. Lombardi (Ed.), Segmental Phonology in Optimality Theory: Constraints and
Representations (pp. 118–156). Cambridge University Press.
Noonan, Michael. (1992). A Grammar of Lango. Berlin: Mouton de Gruyter.
Noske, Manuela. (1995). The Ternary Use of Distinctive Features: In Defense of the Number
Three. Ph.D. Dissertation, University of Chicago.
O’Keefe, Michael. (2005). Transparency in Span Theory. University of Massachusetts Occasional
Papers in Linguistics 33: Papers in Optimality Theory 3. Amherst, MA.
Odden, David. (1978). Abstract Vowel Harmony in Manchu. Linguistic Analysis, 4(2), 149–165.
Odden, David. (1991). Vowel geometry. Phonology, 8, 261–289.
Ohala, John J., & Ohala, Manjari. (1993). The Phonetics of Nasal Phonology: Theorems and Data.
In M. K. Huffman & R. A. Krakow (Eds.), Phonetics and Phonology 5: Nasals, Nasalization,
and the Velum (pp. 225–249). Academic Press.
Öhman, S. E. G. (1966). Coarticulation in VCV Utterances: Spectrographic Measurements.
Journal of the Acoustical Society of America, 39(1), 19–36.
Ọla Orie, Ọlanikẹ. (2001). An alignment-based account of vowel harmony in Ifẹ Yoruba. Journal
of African Languages and Linguistics, 22, 117–143.
Ọla Orie, Ọlanikẹ. (2003). Two harmony theories and high vowel patterns in Ebira and Yoruba.
The Linguistic Review, 20, 1–35.
Padgett, Jaye. (1995). Partial Class Behavior and Nasal Place Assimilation. In K. Suzuki & D.
Elzinga (Eds.), Proceedings of the Arizona Phonology Conference: Workshop on Features in
Optimality Theory (pp. 145–183). Tucson: University of Arizona.
Paradis, Carole. (1986). Phonologie et morphologie lexicales: Les classes nominales en Peul
(Fula). Ph.D. Dissertation, Université de Montréal.
Parkinson, Frederick. (1996). The Representation of Vowel Height in Phonology. Ph.D.
Dissertation, The Ohio State University.
Pater, Joe. (1999). Austronesian nasal substitution and other NC ̥ effects. In R. Kager, H. van der
Hulst, & W. Zonneveld (Eds.), The Prosody Morphology Interface (pp. 310–343). Cambridge
University Press.
Pater, Joe. (2000). Non-uniformity in English secondary stress: the role of ranked and lexically
specific constraints. Phonology, 17, 237–274.
Pater, Joe. (2009a). Morpheme-Specific Phonology: Constraint Indexation and Inconsistency
Resolution. In S. Parker (Ed.), Phonological Augmentation: Essays on Evidence and
Motivation (pp. 123–154). London: Equinox.
Pater, Joe. (2009b). Review: Paul Smolensky and Géraldine Legendre, The harmonic mind: from
neural computation to optimality-theoretic grammar. Phonology, 26, 217–226.
Pater, Joe. (2012). Serial Harmonic Grammar and Berber Syllabification. In T. Borowsky, S.
Kawahara, M. Sugahara, & S. Takahito (Eds.), Prosody Matters: Essays in Honor of Elisabeth
Selkirk (pp. 43–72). Equinox.
389
Piggott, Glyne. (1988). A parametric approach to nasal harmony. In H. van der Hulst & N. Smith
(Eds.), Features, Segmental Structure and Harmony Processes (Part I) (pp. 131–167).
Dordrecht: Foris Publications.
Piggott, G. L. (1992). Variability in Feature Dependency: The Case of Nasality. Natural Language
and Linguistic Theory, 10, 33–77.
Potts, Christopher, Pater, Joe, Jesney, Karen, Bhatt, Rajesh, & Becker, Michael. (2010). Harmonic
Grammar with linear programming: from linear systems to linguistic typology. Phonology,
27(1), 77.
Prince, Alan, & Smolensky, Paul. (1993). Optimality Theory: Constraint Interaction in Generative
Grammar. Manuscript, Rutgers University and University of Colorado Boulder. Published
2004 by Blackwell Publishing.
Proctor, Michael. (2009). Gestural Characterization of a Phonological Class: The Liquids. Ph.D.
Dissertation, Yale University.
Pulleyblank, Douglas. (1988). Vocalic Underspecification in Yoruba. Linguistic Inquiry, 19(2),
233–270.
Pulleyblank, Douglas. (1989). Patterns of Feature Cooccurrence: The Case of Nasality. In S. L.
Fulmer, M. Ishihara, & W. Wiswall (Eds.), Proceedings of the Arizona Phonology Conference
2/Coyote Papers 9 (pp. 98–115).
Pulleyblank, Douglas. (1996). Neutral Vowels in Optimality Theory: A Comparison of Yoruba
and Wolof. Canadian Journal of Linguistics, 41(4), 295–347.
Rhodes, Russell. (2012). Vowel Harmony as Agreement by Correspondence. UC Berkeley
Phonology Lab Annual Report.
Ribeiro, Eduardo Rivail. (2002). Directionality in Vowel Harmony: The Case of Karajá (Macro-
Jê). In J. Larson & M. Paster (Eds.), Proceedings of the Twenty-Eighth Annual Meeting of the
Berkeley Linguistics Society (pp. 475–485). Linguistic Society of America.
Rice, Keren D. (1993). A Reexamination of the Feature [sonorant]: The Status of “Sonorant
Obstruents.” Language, 69(2), 308–344.
Rice, Keren. (1996). Default Variability: The Coronal-Velar Relationship. Natural Language and
Linguistic Theory, 14, 493–543.
Rice, Keren, & Avery, Peter. (1989). On the Interaction between Sonorancy and Voicing (No. 10).
Toronto Working Papers in Linguistics.
Ringen, Catherine O. (1988). Underspecification Theory and Binary Features. In H. van der Hulst
& N. Smith (Eds.), Features, Segmental Structure and Harmony Processes (Part II) (pp. 145–
160). Dordrecht: Foris Publications.
Ringen, Catherine O., & Vago, Robert M. (1998). Hungarian vowel harmony in Optimality
Theory. Phonology, 15, 393–416.
Ritchart, Amanda, & Rose, Sharon. (2017). Moro vowel harmony: implications for transparency
and representations. Phonology, 34, 163–200.
Robins, R. H. (1957). Vowel Nasality in Sundanese: A Phonological and Grammatical Study. In
Studies in Linguistic Analysis (pp. 87–103). London: Basil Blackwell.
Rose, Sharon, & Walker, Rachel. (2004). A Typology of Consonant Agreement as
Correspondence. Language, 80(3), 475–531.
Rose, Sharon, & Walker, Rachel. (2011). Harmony Systems. In J. A. Goldsmith, J. Riggle, & A.
C. L. Yu (Eds.), The Handbook of Phonological Theory (pp. 240–290). Blackwell Publishing.
Safir, Ken. (1982). Nasal Spreading in Capanahua. Linguistic Inquiry, 13(4), 689–694.
390
Sagey, Elizabeth. (1986). The Representation of Features and Relations in Non-Linear Phonology.
Ph.D. Dissertation, Massachusetts Institute of Technology.
Sagey, Elizabeth. (1988). On the III-Formedness of Crossing Association Lines. Linguistic
Inquiry, 19(1), 109–118.
Saltzman, Elliot, & Byrd, Dani. (2000). Task-dynamics of gestural timing: Phase windows and
multifrequency rhythms. Human Movement Science, 19(4), 499–526.
Saltzman, Elliot, & Munhall, Kevin G. (1989). A Dynamical Approach to Gestural Patterning in
Speech Production. Ecological Psychology, 1(4), 333–382.
Sasa, Tomomasa. (2001). Yakut vowel harmony: An Optimality Theory account. Turkic
Languages, 5, 270–287.
Sasa, Tomomasa. (2009). Treatments of Vowel Harmony in Optimality Theory. Ph.D. Dissertation,
The University of Iowa.
Schane, Sanford A. (1984). The fundamentals of particle phonology. Phonology Yearbook,
1(1984), 129–155.
Schane, Sanford A. (1990). Lowered height, laxness, and retracted tongue root: Different
manifestations of phonological aperture. Word, 41(1), 1–16.
Schourup, Lawrence. (1972). Characteristics of Vowel Nasalization. Papers in Linguistics, 5, 530–
548.
Scott, N. C. (1957). Notes on the Pronunciation of Sea Dayak. Bulletin of the School of Oriental
and African Studies, 20(1), 509–512.
Scott, N. C. (1964). Nasal Consonants in Land Dayak (Bukar-Sadong). In D. Abercrombie, D. B.
Fry, P. A. D. MacCarthy, N. C. Scott, & J. L. M. Trim (Eds.), In Honour of Daniel Jones (pp.
432–436). Longmans.
Sebeok, Thomas A, & Ingemann, Frances J. (1961). An Eastern Cheremis Manual. Uralic and
Altaic Series. Bloomington, Indiana: Indiana University.
Shahin, Kimary. (1997). Postvelar Harmony: An Examination of its Bases and Crosslinguistic
Variation. Ph.D. Dissertation, University of British Columbia.
Siptár, Péter, & Törkenczy, Miklós. (2000). The Phonology of Hungarian. Oxford: Oxford
University Press.
Smith, Caitlin. (2016a). A gestural account of neutral segment asymmetries in harmony. In G. Ó.
Hansson, A. Farris-Trimble, K. McMullin, & D. Pulleyblank (Eds.), Proceedings of the 2015
Annual Meeting on Phonology. Linguistic Society of America.
Smith, Caitlin. (2016b). Morphological Consonant Mutation as Gestural Affixation. In
Proceedings of the Fiftieth Annual Meeting of the Chicago Linguistic Society (pp. 411–426).
Smith, Caitlin. (2017a). Harmony Triggering as a Contrastive Property of Segments. In K. Jesney,
C. O’Hara, C. Smith, & R. Walker (Eds.), Proceedings of the 2016 Annual Meeting on
Phonology. Linguistic Society of America.
Smith, Caitlin. (2017b). Harmony Triggering as a Segmental Property. In Proceedings of the 47th
Annual Meeting of the North East Linguistic Society.
Smith, Caitlin, & Blaylock, Reed. (2017). Deriving Exceptional Phonological Patterns from
Contrastive Gestural Strength. Presentation at the Dynamic Modeling in Phonetics and
Phonology workshop at the 53rd Annual Meeting of the Chicago Linguistic Society.
Smith, Jennifer L. (2002). Phonological Augmentation in Prominent Positions. Ph.D. Dissertation,
University of Massachusetts Amherst.
Smolensky, Paul. (1993). Harmony, Markedness, and Phonological Activity. Presentation at
Rutgers Optimality Workshop I (ROW-I).
391
Smolensky, Paul. (2006). Optimality in Phonology II: Harmonic Completeness, Local Constraint
Conjunction, and Feature Domain Markedness. In P. Smolensky & G. Legendre (Eds.), The
Harmonic Mind: From Neural Computation to Optimality-Theoretic Grammar, Volume 2.
Linguistic and Philosophical Implications. Cambridge, MA: MIT Press.
Smolensky, Paul, & Goldrick, Matthew. (2016). Gradient Symbolic Representations in Grammar:
The case of French Liaison. Unpublished Manuscript, The Johns Hopkins University,
Northwestern University.
Smolensky, Paul, & Legendre, Géraldine (Eds.). (2006). The Harmonic Mind: From Neural
Computation to Optimality-Theoretic Grammar, Volume 1. Cognitive Architecture.
Cambridge, MA: The MIT Press.
Sproat, Richard, & Fujimura, Osamu. (1993). Allophonic variation in English /l/ and its
implications for phonetic implementation. Journal of Phonetics, 21, 291–311.
Steriade, Donca. (1987). Locality Conditions and Feature Geometry. In J. McDonough & B.
Plunkett (Eds.), Proceedings of the 17th Annual Meeting of the North East Linguistic Society
(pp. 595–617). Amherst, MA: Graduate Linguistic Student Association.
Steriade, Donca. (1995). Positional Neutralization. unpublished manuscript, University of
California Los Angeles.
Steriade, Donca. (1997). Phonetics in Phonology: The Case of Laryngeal Neutralization.
unpublished manuscript, University of California Los Angeles.
Steriade, Donca. (2001). Directional Asymmetries in Place Assimilation: A Perceptual Account.
In E. Hume & K. Johnson (Eds.), The Role of Speech Perception in Phonology (pp. 219–250).
San Diego: Academic Press.
Steriade, Donca. (2009). The Phonology of Perceptibility Effects: The P-Map and its consequences
for constraint organization. In K. Hanson & S. Inkelas (Eds.), The Nature of the Word: Studies
in Honor of Paul Kiparsky (pp. 151–179). Cambridge, MA: MIT Press.
Sussman, Harvey M, & Westbury, John R. (1981). The Effects of Antagonistic Gestures on
Temporal and Amplitude Parameters of Anticipatory Labial Coarticulation. Journal of Speech
and Hearing Research, 46, 16–24.
Svantesson, Jan-Olof. (1985). Vowel Harmony Shift in Mongolian. Lingua, 67, 283–327.
Svantesson, Jan-Olof, Tsendina, Anna, Karlsson, Anastasia, & Franzén, Vivan. (2005). The
Phonology of Mongolian. Oxford University Press.
Tejada, Laura. (2012). Tone Gestures and Constraint Interaction in Sierra Juarez Zapotec. Ph.D.
Dissertation, University of Southern California.
Temkin Martínez, Michal. (2010). Sources of Non-Conformity in Phonology: Variation and
Exceptionality in Modern Hebrew Spirantization. Ph.D. Dissertation, University of Southern
California.
Ternes, Elmar. (1973). The phonemic analysis of Scottish Gaelic: Based on the dialect of
Applecross, Ross-shire. Hamburg: Buske.
Tilsen, Sam. (2013). A Dynamical Model of Hierarchical Selection and Coordination in Speech
Planning. PLoS ONE, 8(4).
Tilsen, Sam. (2016). Selection and coordination: The articulatory basis for the emergence of
phonological structure. Journal of Phonetics, 55, 53–77.
Törkenczy, Miklós. (2011). Hungarian Vowel Harmony. In M. van Oostendorp, C. J. Ewen, E.
Hume, & K. Rice (Eds.), The Blackwell Companion to Phonology. Blackwell Publishing.
Trigo, Rosario Lorenza. (1988). The Phonological Derivation and Behavior of Nasal Glides. Ph.D.
Dissertation, Massachusetts Institute of Technology.
392
Vago, Robert M. (1973). Abstract Vowel Harmony Systems in Uralic and Altaic Languages.
Language, 49(3), 579–605.
Vago, Robert M. (1976). Theoretical Implications of Hungarian Vowel Harmony. Linguistic
Inquiry, 7(2), 243–263.
van der Hulst, Harry, & Smith, Norval. (1982a). An Overview of Autosegmental and Metrical
Phonology. In H. van der Hulst & N. Smith (Eds.), The Structure of Phonological
Representations (Part I) (pp. 1–46). Dordrecht: Foris Publications.
van der Hulst, Harry, & Smith, Norval. (1982b). Prosodic Domains and Opaque Segments in
Autosegmental Theory. In H. van der Hulst & N. Smith (Eds.), The Structure of Phonological
Representations (Part II) (pp. 311–336). Dordrecht: Foris Publications.
van der Hulst, Harry, & Smith, Norval. (1987). Vowel harmony in Khalkha and Buriat (East
Mongolian). In F. Beukema & P. Coopmans (Eds.), Linguistics in the Netherlands 1987 (pp.
81–90). Dordrecht: Foris Publications.
van der Hulst, Harry, & van de Weijer, Jeroen. (1995). Vowel Harmony. In J. A. Goldsmith (Ed.),
The Handbook of Phonological Theory (pp. 495–534). Oxford: Blackwell Publishers.
Velie Gable, Daniel. (1975). Bosquejo de la fonología y gramatica del idioma Orejón (Coto).
Lima, Peru: Instituto Lingüístico de Verano.
Walker, Rachel. (1998). Nasalization, Neutral Segments, and Opacity Effects. Ph.D. Dissertation,
University of California Santa Cruz. Published 2000.
Walker, Rachel. (2001). Round Licensing, Harmony, and Bisyllabic Triggers in Altaic. Natural
Language and Linguistic Theory, 19, 827–878.
Walker, Rachel. (2003). Reinterpreting transparency in nasal harmony. In J. van de Weijer, V. J.
van Heuven, & H. van der Hulst (Eds.), The Phonological Spectrum Volume I: Segmental
Structure (pp. 37–72). John Benjamins Publishing Company.
Walker, Rachel. (2005). Weak Triggers in Vowel Harmony. Natural Language and Linguistic
Theory, 23(4), 917–989.
Walker, Rachel. (2009). Similarity-sensitive Blocking and Transparency in Menominee. Talk
presented at the 83rd Meeting of the Linguistic Society of America.
Walker, Rachel. (2011). Vowel Patterns in Language. Cambridge University Press.
Walker, Rachel. (2014). Nonlocal Trigger-Target Relations. Linguistic Inquiry, 45(3), 501–523.
Walker, Rachel. (2017a). Temporal Structure in Phonology. Handout from colloquium presented
at Massachusetts Institute of Technology, February 2017.
Walker, Rachel. (2017b). Cumulative Constraint Interactions: Violations and Domains
Encompassing Segmental Spans. In A. Kaplan, A. Kaplan, M. K. McCarvel, & E. J. Rubin
(Eds.), Proceedings of the 34th West Coast Conference on Formal Linguistics (pp. 1–15).
Cascadilla Proceedings Project.
Walker, Rachel. (2018). Feature Identity and Icy Targets in Menominee Vowel Harmony.
Unpublished manuscript, University of Southern California.
Warner, Natasha, Brenner, Daniel, Schertz, Jessamyn, Fisher, Muriel, & Hammond, Michael.
(2015). The aerodynamic puzzle of nasalized fricatives: Aerodynamic and perceptual evidence
from Scottish Gaelic. Laboratory Phonology, 6(2), 197–241.
Watson, Janet C. E. (2002). The Phonology and Morphology of Arabic. New York: Oxford
University Press.
Wilson, Colin. (2000). Targeted Constraints: An Approach to Contextual Neutralization in
Optimality Theory. Ph.D. Dissertation, The Johns Hopkins University.
393
Wilson, Colin. (2001). Consonant cluster neutralisation and targeted constraints. Phonology, 18,
147–197.
Wilson, Colin. (2003). Analyzing unbounded spreading with constraints: marks, targets, and
derivations. Unpublished manuscript, University of California Los Angeles.
Younes, Munther. (1982). Problems in the Segmental Phonology of Palestinian Arabic. Ph.D.
Dissertation, University of Texas Austin.
Younes, Munther A. (1993). Emphasis Spread in Two Arabic Dialects. In M. Eid & C. Holes
(Eds.), Perspectives on Arabic Linguistics V: Papers from the Fifth Annual Symposium on
Arabic Linguistics (pp. 119–145). Amsterdam: John Benjamins Publishing Company.
Zhang, Xi. (1996). Vowel Systems of the Manchu-Tungus Languages of China. Ph.D. Dissertation,
University of Toronto.
Zoll, Cheryl. (1995). Consonant Mutation in Bantu. Linguistic Inquiry, 26, 536–545.
Zoll, Cheryl. (1996). Parsing Below the Segment in a Constraint-Based Framework. Ph.D.
Dissertation, University of California Berkeley.
Zsiga, Elizabeth. (1993). Features, gestures, and the temporal aspects of phonological
organization. Ph.D. Dissertation, Yale University.
Zsiga, Elizabeth C. (1997). Features, Gestures, and Igbo Vowels: An Approach to the Phonology-
Phonetics Interface. Language, 73(2), 227–274.
Zsiga, Elizabeth C. (2000). Phonetic alignment constraints: consonant overlap and palatalization
in English and Russian. Journal of Phonetics, 28, 69–102.
Abstract (if available)
Abstract
In this dissertation, I develop the Gestural Harmony Model, a model of harmony situated within a phonological framework that assumes gestures as the units of subsegmental representation. Originally developed within Articulatory Phonology (Browman & Goldstein 1986, 1989, et seq.), gestures are dynamically defined units of phonological representation that are specified for a target articulatory state of the vocal tract. In the Gestural Harmony Model, harmony is triggered when a gesture extends its period of activation and overlaps other segments in a word. To model this ability of a gesture to extend its activation, I propose the addition of two new parameters to the representation of gestures: persistence and anticipation. With the addition of these parameters, gestures can be specified as either self-deactivating or persistent (non-self- deactivating), and as either self-activating or anticipatory (early-activating). A persistent gesture is one that does not self-deactivate when its goal articulatory state is achieved, thus overlapping following segments and triggering progressive (rightward) harmony. An anticipatory gesture is one that is activated early, thus overlapping preceding segments and triggering regressive (leftward) harmony. ❧ In addition to these representational innovations, I develop a phonological grammar, situated within the framework of Optimality Theory (Prince & Smolensky 1993/2004), that operates over gestural representations. The presence of harmony in a language is attributed to whether the segments in a language’s surface phonological inventory contain persistent and/or anticipatory gestures. As a result, in the Gestural Harmony Model patterns of harmony triggering result from the interaction of markedness and faithfulness constraints that shape the surface inventory and determine the distributions of the segments in that inventory. One of the major advantages of the approach to harmony triggering in the Gestural Harmony Model is that harmony systems in which bearers of a harmonizing property idiosyncratically trigger or fail to trigger harmony can be attributed to preservation of a contrast between persistent and self-deactivating gestures in the case of progressive harmony, and anticipatory and self-activating gestures in the case of regressive harmony. This approach to harmony triggering avoids the pathological predictions made by some other analyses of phonological idiosyncrasy and exceptionality. ❧ The Gestural Harmony Model’s representation of harmony also proves advantageous in the analysis of transparency and blocking. In this model, transparency and blocking are the results of two distinct theoretical mechanisms, successfully accounting for the distinct crosslinguistic patterns in the attestation of transparent and blocking segments in some types of harmony. I analyze transparent segments as undergoers of harmony that include in their representations a gesture that is antagonistic to a harmonizing gesture. Antagonistic gestures are those that are specified for directly conflicting target articulatory states of the vocal tract, and as a result enter into competition with one another. Transparency arises when intergestural competition is resolved in favor of the gesture of the transparent segment due to its greater specified gestural strength. Blocking of harmony, on the other hand, results from a different theoretical mechanism: intergestural inhibition, by which one gesture deactivates another. The Gestural Harmony Model’s splitting of transparency and blocking among two distinct theoretical mechanisms makes several advantageous typological predictions. Chief among these is that in some types of harmony, the set of attested transparent segments is a subset of the set of attested blocking segments. This is attributed to the idea that only certain types of segments possess the gestural makeup necessary to surface as transparent to harmony when overlapped by a harmonizing gesture.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Articulatory dynamics and stability in multi-gesture complexes
PDF
Soft biases in phonology: learnability meets grammar
PDF
Articulatory knowledge in phonological computation
PDF
Beatboxing phonology
PDF
Tone gestures and constraint interaction in Sierra Juarez Zapotec
PDF
Individual differences in phonetic variability and phonological representation
PDF
The prosodic substrate of consonant and tone dynamics
PDF
Syntax-prosody interactions in the clausal domain: head movement and coalescence
Asset Metadata
Creator
Smith, Caitlin Michele
(author)
Core Title
Harmony in gestural phonology
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Linguistics
Publication Date
07/30/2018
Defense Date
04/30/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
articulatory phonology,gestural phonology,gestures,Harmony,OAI-PMH Harvest,vowel harmony,vowel-consonant harmony
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Walker, Rachel (
committee chair
), Goldstein, Louis (
committee member
), Iskarous, Khalil (
committee member
), Jesney, Karen (
committee member
), Nayak, Krishna (
committee member
)
Creator Email
caitlin.smith14@gmail.com,smithcm@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-42567
Unique identifier
UC11670352
Identifier
etd-SmithCaitl-6574.pdf (filename),usctheses-c89-42567 (legacy record id)
Legacy Identifier
etd-SmithCaitl-6574.pdf
Dmrecord
42567
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Smith, Caitlin Michele
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
articulatory phonology
gestural phonology
gestures
vowel harmony
vowel-consonant harmony