2014 Peresectives On Phonological Theory and Development

Perspectives on Phonological Theory and Development
Language Acquisition and Language Disorders

(LALD)
Volumes in this series provide a forum for research contributing to theories of

language acquisition (first and second, child and adult), language learnability,
language attrition and language disorders.
For an overview of all books published in this series, please see
http://benjamins.com/catalog/lald
Series Editors
Roumyana Slabakova Lydia White
University of Iowa McGill University
Editorial Board
Kamil Ud Deen Mabel Rice
University of Hawaii at Manoa University of Kansas
Katherine Demuth Luigi Rizzi
Macquarie University University of Siena
Naama Friedmann Petra Schulz
Tel Aviv University Johann Wolfgang Goethe-Universitt
Heather Goad Bonnie D. Schwartz
McGill University University of Hawaii at Manoa
Barbara Hhle Antonella Sorace
University of Potsdam University of Edinburgh
Nina Hyams Ianthi Maria Tsimpli
University of California at Los Angeles Aristotle University of Thessaloniki
Jrgen M. Meisel
University of Calgary
Volume 56
Perspectives on Phonological Theory and Development.
In honor of Daniel A. Dinnsen
Edited by Ashley W. Farris-Trimble and Jessica A. Barlow
Perspectives on
Phonological Theory
and Development
In honor of Daniel A. Dinnsen
Edited by
Ashley W. Farris-Trimble
Simon Fraser University
Jessica A. Barlow
San Diego State University
John Benjamins Publishing Company

Amsterdam/Philadelphia
TM
The paper used in this publication meets the minimum requirements of
8
theAmerican National Standard for Information Sciences Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.
Library of Congress Cataloging-in-Publication Data
Perspectives on Phonological Theory and Development : In honor of Daniel A. Dinnsen /

Edited by Ashley W. Farris-Trimble and Jessica A. Barlow.
p. cm. (Language Acquisition and Language Disorders, issn 0925-0123 ; v. 56)
Includes bibliographical references and index.
1. Grammar, Comparative and general--Phonology. 2. Language acquisition. I.Farris-
Trimble, Ashley, editor of compilation. II. Barlow, Jessica A., editor of
compilation. III. Dinnsen, Daniel A., honouree.
P215.P47 2014
414--dc23 2013049290
isbn 978 90 272 5318 7 (Hb ; alk. paper)
isbn 978 90 272 7054 2 (Eb)
2014 John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. P.O. Box 36224 1020 me Amsterdam The Netherlands
John Benjamins North America P.O. Box 27519 Philadelphia pa 19118-0519 usa
Table of contents
Foreword and tabula gratulatoria vii
Introduction 1
Steven B. Chin
section 1. Representations and contrast: What does the learner know? 9
Prosodic Licensing and the development of phonological

and morphological representations 11
Katherine Demuth
Covert contrast in the acquisition of second language phonology 25

Fred R. Eckman, Gregory K. Iverson and Jae Yung Song
section 2. Sources of individual differences in phonological acquisition 49
Sibling rivalry: Comparing phonological similarity between

twin and non-twin siblings 53
David Ingram and Virginia L. Dubasik
Abstracting phonological generalizations: Evidence from

children with disorders 71
Judith A. Gierut, Michele L. Morrisette and Caitlin J. Younger
Rapid phonological coding and working memory dynamics in children

with cochlear implants: Cognitive foundations of spoken language processing 91
David B. Pisoni
section 3. Cross-linguistic approaches to phonological acquisition 113
What guides childrens acquisition of #sC clusters? A cross-linguistic account 115

Mehmet Yava
The role of phonological context in childrens overt marking of -s

in two dialects of American English 133
Jessica A. Barlow and Sonja L. Pruitt-Lord
German settlement varieties in Kansas: Some unusual phonological

and morphological developments with the approach of language death 155
William D. Keel
section 4. Theoretical advances in the field:

Constraint-based approaches 173
The role of onsets in primary and secondary stress patterns 175

Laura W. McGarrity
A faithfulness conspiracy: The selection of unfaithful mappings

in Amahls grammar 199
Ashley Farris-Trimble
Superadditivity and limitations on syllable complexity in Bambara words 223

Christopher R. Green and Stuart Davis
Author index 249

Subject index 253
Foreword and tabula gratulatoria
Dan Dinnsens long and illustrious career has spanned many facets of phonology
and language acquisition and has influenced researchers in a number of disci-
plines. The wide variety of topics in this volume (including, among others, both
first and second language acquisition, minority dialects, children with cochlear
implants, genetics, prosody, theoretical phonology, and cross-linguistic typology)
attest to the broad impact of his research.
Dinnsens influence is not simply limited to his research, however. He has men-
tored dozens of students and has taught thousands. He encouraged, supported, and
pushed his students to do their best. As a mentor, he was always available to answer
a question, give advice, or listen to a rant about the futility of a research topic (and
then, of course, to argue otherwise). As a teacher, he was highly respected and had a
reputation for being tough but funny and fair. From his impressions of Kermit the
Frog to his Thanksgiving-time analogies between phonology and turkeys to his sto-
ries about unwittingly frightening parents when he diagnosed their children with
terminal devoicing, he always managed to remind his students and mentees that as
a discipline, phonology can be entertaining. His reminder that nobody dies if a
linguist is wrong encouraged students to take chances and think for themselves.
This volume is a tribute to Dans well-deserved retirement and a testament to
his well-deserved reputation. It is also a thank-you, from the bottom of our hearts,
for all the time and energy he has poured into our development as students and
colleagues over the years. So thanks, Dan, for everything we hope this volume
does you justice.
Finally, wed also like to extend our thanks to a number of people who helped
make this volume possible, including Philip Combiths, Peter Flipsen, Heather
Goad, Matthew Gordon, Jill Hoover, Michael Marlo, Andrew Nevins, Elgustus
Polite, Tom Shannon, Laura Catharine Smith, Carol Stoel-Gammon, Anne-Michelle
Tessier, and Steven Weinberger.
Ashley & Jessica
Tabula gratulatoria
The following students, colleagues, and friends wish to congratulate Dan on his
retirement. All have benefitted greatly from his mentorship and collegiality. Of
course, this list represents only a small fraction of the individuals whose careers
and lives have been influenced by Dan apologies go out to those whom we were
unable to contact.
ra Birna sgeirsdttir Jeannette S. Leonard

Julie Auger Ania ubowicz
Karen Baertsch Paul A. Luce
Eric Bakovi Joshua Lumsden
R. Joe Campbell Michael R. Marlo
Nancy J. Caplow Nikole Miller
Jan Charles-Luce John McCarthy
Cynthia G. Clopper Paul Newman
Andries W. Coetzee Samuel Obeng
Kathleen Currie Hall Tom Perry
Phil J. Connell Thomas W. Powell
Cynthia Core Marwa Ragheb
Kenneth de Jong Heather Rice
Markus Dickinson Rebecca Ronquest
Michael Dow Thomas F. Shannon
Mary Elbert Gerald Sanders
Marilyn Estep B. Devan Steiner
Leah Fabiano-Smith Nola Stephens
Peter Flipsen Jr. Carol Stoel-Gammon
Steven Franks Holly Storkel
Leslie Gabriele Anne-Michelle Tessier
Heather Goad Kathryn Tippetts
Matt Goldrick Mike Trimble
Mike Hammond Melissa Troyer
Nicholas Henriksen Barbara Vance
Larry Humes Mark VanDam
Steliana Ivanova Marc van Oostendorp
Daniel Karvonen Michael S. Vitevitch
Yoshihisa Kitagawa Thomas J. Walsh
Kassim Kone Tania Zamuner
Sandra Kuebler Suzanne Martin Ziemer
Laurence B. Leonard
Introduction
Steven B. Chin
Indiana UniversityPurdue University Indianapolis
Introduction
When Daniel Dinnsen retired from Indiana University in 2012, he was Chancellors
Professor of Linguistics and adjunct professor of speech and hearing sciences and
had served as a faculty member in the Department of Linguistics for 40 years, had
chaired 17 PhD dissertation committees, and had served as a member for 44 oth-
ers. In addition, he had conducted research under two grants from the National
Institutes of Health (providing continuous funding from 1985 to 2014), taught lin-
guistics and phonology to hundreds of undergraduate and graduate students, testi-
fied as an expert witness in federal court, and served for six years as a department
chair. The current volume, a collection of papers by Dinnsens former students and
colleagues, is edited by two of his former graduate students, Ashley Farris-Trimble
(Ph.D., 2008), now a faculty member at Simon Fraser University, and Jessica Barlow
(Ph.D., 1997), now a faculty member at San Diego State University.
Born in Chicago, Dinnsen grew up in Indianapolis and attended Arsenal
Technical High School (Tech), so named because its oldest buildings had been
originally built as a U.S. Civil War arsenal located east of downtown Indianapolis.
There he majored in social studies, Spanish, and mathematics; was a member of
the Key Club; and served as sergeant-at-arms of the Spanish Club. Originally dis-
couraged from pursuing education any further than high school because of his
extremely poor vision, Dinnsen nevertheless found an advocate in one of his
teachers, Anna Parker, who persuaded him and his parents, as well as a number of
college administrators, that going to college was well within his abilities. As an
alumna of Indiana University, located 50 miles to the southwest in Bloomington,
Ms. Parker advocated strongly for Dinnsens enrollment there. Dinnsens high
school interest in Spanish carried over into his college career, and he declared a
Spanish major. R. Joe Campbell, who taught grammar courses in the Department
of Spanish and Portuguese, introduced Dinnsen to concepts of linguistics, and
Dinnsen enrolled in several courses offered by the linguistics department. At the
Steven B. Chin
time, linguistics was not a fully formed undergraduate program, so after taking a
freshman-level introduction to language course, Dinnsen enrolled in several grad-
uate courses, including phonology, with Gerald Sanders, and syntax, with Andreas
Koutsoudas. It was in these courses that Dinnsen was first introduced to contem-
porary theories of language, particularly generative phonology. Still an undergrad-
uate, Dinnsen was befriended by several graduate students in the course who
learned along with him, including Harry Gradman, who would eventually become
Dinnsens department chair in linguistics, just prior to Dinnsens becoming
Gradmans chair in the same department from 1986 to 1992. Dinnsens developing
interest in linguistics coincided with the ascendancy of transformational and gen-
erative grammar during the late 1960s: Noam Chomskys Aspects of the Theory of
Syntax had appeared in 1965, and manuscript versions were being distributed of
parts of Noam Chomsky and Morris Halles The Sound Pattern of English (SPE)
prior to its official publication in 1968.
After graduating Phi Beta Kappa from Indiana University with a double major
in Spanish and linguistics in 1969, Dinnsen enrolled in the Ph.D. program in lin-
guistics at the University of Texas at Austin, where he studied under Robert D.
King. His dissertation, General constraints on phonological rules (1972), argued
against the free application of global rules in phonological derivations. In fact, the
standard theory of generative phonology as presented in SPE had little to say about
constraints on types of rules or on their application, and this gap formed the basis
for further development of phonological theory in the years after 1968.
After completing the Ph.D. in 1972, Dinnsen accepted a position as assistant
professor of linguistics at Indiana University, thus returning to his undergraduate
alma mater after only three years away at Texas. During his early faculty years at
Indiana University, Dinnsen developed Atomic Phonology, an empirically-based
account of universal principles that constrain the types of phonological rules that
could occur in language. Dinnsen postulated that atomic rules are entirely inde-
pendent rules of grammar which are presumed to be the most basic, most specific
rules that can be motivated on empirical grounds. For any given atomic rule, then,
there could be no more specific nor equally specific and independent rule effecting
a same or otherwise related structural change. (Dinnsen, 1979b:31). Conversely,
non-atomic rules are all and only those rules which are complements of atomic
rules or complements of their complements, where two rules are complements if
just those commonly shared (identical) features in the structural description of the
two rules are sufficient to define precisely the same set of input representations
defined by the two rules jointly (Dinnsen, 1979b: 31). Dinnsen provided the fol-
lowing example of an atomic rule (1) and a complementary, non-atomic rule (2)
for terminal devoicing (Dinnsen, 1979b:32):
Introduction
(1) [sonorant, continuant] [voice]/___#

(Stop obstruents are voiceless word-finally.)
(2) [sonorant, +continuant] [voice]/___#
(Fricative obstruents are voiceless word-finally.)
On empirical evidence from such languages as Italian and Turkish, Dinnsen sug-
gested that (1) was the atomic rule for terminal devoicing. Because (2) is comple-
mentary, and thus non-atomic, it is not a possible independently occurring rule of
grammar. Negative empirical evidence bears this out; that is, no language has a
terminal devoicing rule that applies only to fricatives but not stops.
In fact, Dinnsens concern with rules types and constraints on rule application
reflected a growing interest among phonologists in addressing multiple concerns
about phonology that had not been addressed in SPE. By the mid to late 1970s
several modifications of the standard theory had been proposed, including Natu-
ral Phonology (Stampe, 1973) and Autosegmental Phonology (Goldsmith, 1976).
In the fall of 1977, Dinnsen organized a Conference on the Differentiation of Cur-
rent Phonological Theories at Indiana University with the support of the National
Science Foundation and the American Council of Learned Societies, as well as the
Indiana University Linguistics Club, the Indiana University Department of Lin-
guistics, and the Indiana University Office of Research and Graduate Develop-
ment. Nine papers representing different approaches to phonology were presented
by major originators or proponents of those approaches, including both Stampe
and Goldsmith. Two years later, a volume edited by Dinnsen with the nine papers,
along with a commentary article called How different are they? by Fred W.
Householder, Jr. (1979), and several short critiques of the main papers prepared by
conference discussants, was published by Indiana University Press and titled Cur-
rent Approaches to Phonological Theory (Dinnsen, 1979a).
One of the people attending the fall 1977 conference was an assistant professor
from the Department of Speech and Hearing Sciences at Indiana University named
Mary Elbert. Originally trained as a speech-language pathologist, Elbert had re-
ceived her Ph.D. from the University of Kansas and was pursuing research at Indiana
University on speech errors by children with functional misarticulation, that is,
children with speech errors for which no organic basis is known. Shortly after the
conference, Elbert contacted Dinnsen to ask whether some current approach to
phonology might help to explain the observed data from functional misarticula-
tion. This query from Mary Elbert was Dinnsens introduction to clinical linguistics
and the study of first language acquisition. Clinical and acquisition studies would
define the majority of Dinnsens research program from this time to his retirement,
collaborating first with Mary Elbert and then with Judith Gierut, also a faculty
member in the Department of Speech and Hearing Sciences at Indiana University.
Steven B. Chin
Shortly after meeting Mary Elbert, however, Dinnsen commenced pursuit of

another of his interests, that of linguistics education. Dinnsen had originally
planned to teach high school Spanish when he entered Indiana University and had
always considered himself both an educator and a scientist. The 1960s and 1970s
saw a widespread proliferation of undergraduate linguistics major programs in the
United States, much of it attributable to paradigm-shifting developments in lin-
guistic theory and the rise of transformational and generative grammar. Materials
for the instruction of undergraduates and junior graduate students in contem-
porary linguistic theory and analysis were not much available at the time, and to
address this lack, Dinnsen founded and edited a small journal, Innovations in
Linguistics Education, offset-printed and distributed by the Indiana University
Linguistics Club. The journal included articles by linguistics educators from across
the United States on methods for teaching linguistics and exercises designed to
demonstrate linguistic argumentation and state-of-the-art analytical methodology.
This journal was published from 1979 to 1992.
From 1981 to 1982, Dinnsen was a visiting scientist at Bell Laboratories in
Murray Hill, New Jersey. Known for its development of a wide range of revolution-
ary technologies, Bell Labs provided Dinnsen an opportunity to pursue his inde-
pendent research and learn from the many staff and visiting scientists there.
Additionally, Bell Labs gave Dinnsen access to computers, which aside from large
mainframes, were not part of the everyday life of academics at the time. Beginning
in the first grade, Dinnsen had been taught to use a typewriter, the belief at the
time being that typing could provide a means of communication for people with
visual impairment. It is not clear that this logic could possibly have made sense at
the time, but now with the ubiquity of computers with keyboards, it turns out that
somebody did know what he was talking about. During his time in Murray Hill,
and with the aid of new-to-him word processing, Dinnsen completed several
chapters of a planned monograph with Mary Elbert and Gary Weismer from the
speech and hearing sciences department back in Bloomington. Under the tutelage
of Osamu Fujimura, Dinnsen was also able to complete several computer-aided
acoustic analyses of Catalan data for research he was conducting at the time.
After returning to Bloomington, Dinnsen continued to teach and to conduct
research, increasingly now in clinical phonology but also in non-clinical areas,
much of it in collaboration with his graduate students. These included studies of
phonological neutralization in Catalan with Jan Charles-Luce and Janet DeCesaris
and in Polish with Louisa Slowiaczek. Between 1980 and 1992, Dinnsen super-
vised seven Ph.D. dissertations, and only one of these dealt with clinical research,
Edith Maxwells (Ph.D., 1981) A study of misarticulation from a linguistic perspec-
tive. Groundbreaking in its application of generative phonological principles to
the analysis of functional misarticulation, Maxwells dissertation and the research
Introduction
behind it prepared the way for more formal statements of the application of cur-
rent linguistic theory to clinical data, first in an article by Weismer, Dinnsen, and
Elbert (1981) in the Journal of Speech and Hearing Disorders called A study of the
voicing distinction associated with omitted word-final stops, and then in a mono-
graph from the American Speech-Language-Hearing Association (ASHA) called
Phonological Theory and the Misarticulating Child, a collection of articles by Elbert,
Dinnsen, and Weismer (1984). Many of the proposals in the various chapters of
the monograph (which, it may be recalled, Dinnsen was already formulating dur-
ing his time at Bell Laboratories) were relatively shocking to professionals in
speech-language pathology, particularly the contention that childrens underlying
representations could differ from adult ones. The standard approach, based to a
large extent on David Stampes Natural Phonology, held that underlying represen-
tations were uniform across speakers of the language, including children learning
the language, and that surface differences were due to processes that essentially
created unmarked phonetic representations. The notion of incorrect underlying
representations for some children remained a contentious issue in the field of child
phonological acquisition for some time.
Around the same time as the appearance of the ASHA monograph in 1984,
Dinnsen was contacted by representatives of the law firm of Barrett, Barrett and
McNagny of Fort Wayne, Indiana. The law firm was representing Pathfinder Com-
munications, seeking a preliminary injunction against Midwest Communications
(593 F.Supp. 281 (1984)) in United States District Court for the Northern District
of Indiana. Pathfinder claimed that the newly-assigned call letters WMCZ used by
a radio station owned by Midwest were confusable with the established call letters
WMEE used by a station owned by Pathfinder. Such disputes had in the past been
resolved by the Federal Communications Commission (FCC) under a significant
likelihood of confusion standard. The FCC had since relinquished this role, and
the plaintiffs sought the injunction under a likelihood of confusion standard
enunciated by the Seventh Circuit Court of Appeals. To help with its case, the
plaintiff called on Dinnsen to provide expert testimony as to the likelihood of
confusion of the two sets of call letters. To produce a straightforward set of criteria
under which similarity and thus confusability of trademarks could be assessed,
Dinnsen brought to bear two methodological approaches widely used in linguistic
and phonetic analysis: phonetic transcriptions and spectrograms. Dinnsens con-
clusion from these two analyses convinced the court that the two sets of calls letters
were overwhelmingly phonetically similar and could lead to confusion. Evidence
of actual confusion would of course be more difficult to establish, but unfortu-
nately for the defendant, the manager of the defendants radio station confused the
two marks during a discussion with both attorneys; this discussion was tape-
recorded and presented to the court at the evidentiary hearing. Based on Dinnsens
Steven B. Chin
qualifications, methodology, approach, testimony, and general demeanor (593

F.Supp. 283), the court credited Dinnsen wherever there was a conflict between
Dinnsen and the defendants witness, concluded that there was a likelihood of con-
fusion, and granted the injunction. The district courts acceptance of the method-
ology used by Dinnsen resulted in his being retained as an expert witness in an-
other dozen similar trademark cases through 2006.
As interesting an intellectual diversion as trademark law was, the majority of
Dinnsens scientific attention now needed to be directed to clinical linguistics, be-
cause in 1985, Mary Elbert and he were awarded their first grant from the U.S.
National Institutes of Health to conduct a five-year research project called Phono-
logical knowledge and learning patterns. Phonological knowledge referred to a
childs knowledge of the surrounding languages phonological system within a gen-
erative phonological framework, thus knowledge of underlying representations,
knowledge of phonological rules, and so forth. The grant-supported project was at
heart a treatment study to determine whether targeting areas of more knowledge
vs. areas of less knowledge would be most efficacious in treating functional misar-
ticulation. Various side studies, of course, departed from the treatment aspects of
the project as the investigators and their graduate students examined various as-
pects of the phonological systems of the children who participated in the study.
Eventually, as well, the theoretical paradigm in which the analyses were conducted
changed, and changed multiple times. Throughout the 1980s and 1990s new ap-
proaches to phonological analysis came and went, including Feature Geometry and
Underspecification Theory. While Mary Elbert kept watch over the clinical treat-
ment phases of the project, Dinnsen and his research assistants kept one eye on the
data and the other on current theoretical literature, adapting their analyses to
changing theoretical conceptions of representations, rules, and constraints. At clin-
ical conferences, Dinnsen and his colleagues were often the only ones to present in
a particular theoretical framework, and at theoretical conferences, they were often
the only ones to present clinical data from children. Because of this, Dinnsen,
Elbert, and their Indiana University colleagues achieved a certain notoriety, par-
ticularly among their more clinical colleagues. Still, though, with each new analysis
within a different framework, former mysteries were solved, and the use of clinical
data provided good tests of the robustness of new approaches to phonological anal-
ysis. It was during this period that I worked with Dinnsen, both as a Ph.D. student
and as a research associate for the NIH-supported project. My 1993 dissertation
examined how distinctive features are organized and specified in the phonological
systems of children with functional misarticulation (Chin, 1993). Later, as a faculty
member at the Indiana University School of Medicine, I extended the analysis tech-
niques that I had learned and developed while working with Dinnsen to the pho-
nological systems of deaf children who used cochlear implants.
Introduction
By far the biggest game changer since SPE for Dinnsen, his colleagues, and the
way they viewed their data was Optimality Theory (OT), a constraint-based, non-
derivational approach to phonology and phonological analysis. Starting with his
graduate student Jessica Barlow (Ph.D., 1997), Dinnsens Clinical Phonology Lab-
oratory quickly adapted to the new framework, analyzing and reanalyzing clinical
data and phonological phenomena such as conspiracies, opacity, and sympathy in
terms of Optimality Theory. Ph.D. dissertations supervised by Dinnsen that incor-
porated OT analyses included those by Jessica Barlow (1997), Laura McGarrity
(2003), Kimberly Swanson (2006), and Ashley Farris-Trimble (2008). Dinnsen had
already been working for some time with Judith Gierut, another faculty member
in the Department of Speech and Hearing Sciences. Her 1985 Indiana University
dissertation, On the relationship between phonological knowledge and generaliza-
tion learning in children (Gierut, 1985) had defined for her a career-long research
program pursued with support by the National Institutes of Health. After Mary
Elberts retirement, Dinnsen signed on as a co-principal investigator with Gierut
on an NIH-funded research project called Development of phonological catego-
ries. Dinnsen and Gieruts edited 2008 volume, Optimality Theory, Phonological
Acquisition and Disorders, with articles by colleagues and current and former stu-
dents, demonstrated the extent to which Optimality Theory had provided a suffi-
ciently rich and robust framework for the analysis of clinical and acquisition data.
With the exception of his place of birth and a few years in graduate school and
at Bell Laboratories, Daniel Dinnsen has spent his entire life in Indiana. From his
modest offices in Lindley Hall and then Memorial Hall on the Indiana University
Bloomington campus, Dinnsen trained successive generations of linguistic scien-
tists and changed the way people view the role of acquisition and clinical data in
the development of linguistic theory. As of this writing it is not clear whether In-
diana can be considered a destination, but to the scores of colleagues who have
worked with him, the hundreds of students who have studied with him, and the
dozens of friends who have laughed with him, it has been well worth the trip.
References
Barlow, J.A. 1997. A Constraint-Based Account of Syllable Onsets: Evidence from Developing
Systems. PhD dissertation, Indiana University.
Chin, S.B. 1993. The Organization and Specification of Features in Functionally Disordered
Phonologies. PhD dissertation, Indiana University.
Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press.
Chomsky, N. & Halle, M. 1968. The Sound Pattern of English. New York NY: Harper & Row.
Dinnsen, D.A. 1972. General Constraints on Phonological Rules. PhD dissertation, University
of Texas at Austin.
Steven B. Chin
Dinnsen, D.A. (ed.). 1979a. Current Approaches to Phonological Theory. Bloomington IN:
Indiana University Press.
Dinnsen, D.A. 1979b. Atomic phonology. In Current Approaches to Phonological Theory, D.A.
Dinnsen (ed.), 3149. Bloomington IN: Indiana University Press.
Dinnsen, D.A. & Gierut, J.A. 2008. Optimality Theory, Phonological Acquisition and Disorders.
London: Equinox.
Elbert, M., Dinnsen, D.A. & Weismer G. 1984. Phonological Theory and the Misarticulating
Child [ASHA Monographs 22]. Rockville MD: American Speech-Language-Hearing
Association.
Farris-Trimble, A.W. 2008. Cumulative Faithfulness Effects in Phonology. PhD dissertation,
Indiana University.
Gierut, J.A. 1985. On the Relationship Between Phonological Knowledge and Generalization
Learning in Children. PhD dissertation, Indiana University.
Goldsmith, J.A. 1976. Autosegmental Phonology. Bloomington IN: Indiana University Linguis-
tics Club.
Householder, F.W. 1979. How different are they? In Current Approaches to Phonological Theory,
D.A. Dinnsen (ed.), 251264. Bloomington, IN: Indiana University Press.
Maxwell, E.M. 1981. A Study of Misarticulation from a Linguistic Perspective. PhD dissertation,
Indiana University.
McGarrity, L.W. 2003. Constraints on Patterns of Primary and Secondary Stress. PhD disserta-
tion, Indiana University.
Stampe, D. 1973. A Dissertation on Natural Phonology. PhD dissertation, University of Chicago.
Swanson, K.A. 2006. Acquisition Versus Suppression of Phonological Processes in the Second
Language Acquisition of French and English. PhD dissertation, Indiana University.
Weismer, G., Dinnsen, D. A. & Elbert, M. 1981. A study of the voicing distinction associated
with omitted word-final stops. Journal of Speech and Hearing Disorders 46: 320328.
section 1
Representations and contrast

What does the learner know?
Much of Dinnsens early research into phonological acquisition was concerned

with understanding what children actually know about the language they are ac-
quiring. That is, what constitutes childrens phonological and lexical representa-
tions? Are they highly specified and adult-like, only generally specified, or simply
incorrect? Toward this end, Dinnsen examined covert contrasts, a phenomenon
that occurs when a child makes a contrast between two sounds that is not easily
perceptible to the listener. Such contrasts can be identified with careful acoustic
measurements (Forrest, Weismer, Hodge, Dinnsen & Elbert, 1990; Gierut &
Dinnsen, 1986). For instance, Weismer, Dinnsen and Elbert (1981) illustrated
that some children who delete word-final stops nevertheless make a duration
contrast in the previous vowel that indicates knowledge of the voicing feature
of the deleted segment. This parallels some adult languages in which the neutral-
ization of a contrast is made incomplete by some remaining phonetic contrast
(e.g., Dinnsen, 1985; Dinnsen & Charles-Luce, 1984). Dinnsen also examined the
development of contrastive features and underlying representations in children
(Dinnsen, 1996; Dinnsen & Barlow, 1998), arguing that phonological theories
constructed to describe adult languages can be extended to describe developing
systems as well. He revisited the issue in a plenary address to the Boston Univer-
sity Conference on Language Development in 2002, noting that more recent pho-
nological theories have allowed researchers to revise their initial assumptions
about childrens representations.
As acoustic measurements have become more fine-tuned and new experimen-
tal paradigms have made mental representations more accessible for study, the is-
sue of what learners know, especially as reflected in phonological contrast, has
become central once again. In the two chapters in this section, the issues of repre-
sentation and contrast are re-visited. Demuth (Prosodic Licensing and the devel-
opment of phonological and morphological representations) and Eckman et al.
(Covert contrast in the acquisition of second language phonology) examine the
acquisition of contrast in children learning a first language and adults learning a
second language, respectively.
Demuth reviews a number of findings that suggest that much of childrens

early acquisition of phonological and morphological representations can be ex-
plained under the Prosodic Licensing Hypothesis. This helps account for much of
the variability seen in early development by predicting in which phonological po-
sitions children are expected to first represent contrasts.
Eckman et al. challenge the view that second language learners are unaware of
some non-native contrasts by presenting evidence of covert contrasts in the pro-
ductions of these learners, contrasts which are not perceived by native speakers of
the language but nevertheless reveal the second language learners sophisticated
linguistic awareness.
The analyses in Chapters 2 and 3 (of contrast relative to prosodic position and
covert contrast in L2 learners, respectively) are both novel. Moreover, they ap-
proach a similar set of questions in two types of acquisition L1 and L2. Bridging
the two lines of research and finding questions (and answers!) that link the two
contributes greatly to a unified understanding of language acquisition. This set of
chapters thus brings a variety of new perspectives to a question that has long been
fundamental to the field.
References
Dinnsen, D.A. 1985. A re-examination of phonological neutralization. Journal of Linguistics 21:

265279.
Dinnsen, D.A. 1996. Context-sensitive underspecification and the acquisition of phonemic con-
trasts. Journal of Child Language 23: 5779.
Dinnsen, D.A. 2002. A reconsideration of childrens phonological representations. In B.
Skarabela, S. Fish & A. H.-J. Do (Eds.), Proceedings of the 26th Annual Boston University
Conference on Language Development (pp. 123). Somerville, MA: Cascadilla Press.
Dinnsen, D.A. & Barlow, J.A. 1998. On the characterization of a chain shift in normal and de-
layed phonological acquisition. Journal of Child Language 25: 6194.
Dinnsen, D.A. & Charles-Luce, J. 1984. Phonological neutralization, phonetic implementation
and individual differences. Journal of Phonetics 12: 4960.
Forrest, K., Weismer, G., Hodge, M., Dinnsen, D.A., & Elbert, M. 1990. Statistical analysis of
word-initial /k/ and /t/ produced by normal and phonologically disordered children. Clini-
cal Linguistics & Phonetics 4: 327340.
Gierut, J.A., & Dinnsen, D.A. 1986. On word-initial voicing: Converging sources of evidence in
phonologically disordered speech. Language and Speech 29: 97114.
Weismer, G., Dinnsen, D.A. & Elbert, M. 1981. A study of the voicing distinction associated with
omitted, word-final stops. Journal of Speech and Hearing Disorders 46: 320328.
Prosodic Licensing and the development
of phonological and morphological
representations
Katherine Demuth
Macquarie University
One of the challenges for understanding the processes underlying the acquisition
of phonology has been the variability found in early speech productions.
Our recent research suggests that much of this is due to the phonological
(or prosodic) context in which words (and their segments) appear. This paper
explores some of the recent findings on childrens acquisition of phonological/
prosodic units as a function of syllable and word structure, showing how
acoustic analysis provides evidence of childrens developing phonological
representations from their first words. It then shows that similar processes can
account for the variable emergence of early grammatical morphemes, suggesting
that these are also Prosodically Licensed. These findings are discussed in terms of
a developmental model of language planning and production.
Introduction
There has been much research on the development of phonological representa-

tion, with gradual learning curves being the norm, even for an individual child.
Thus, a given child at a given stage of development may produce a particular seg-
ment, syllable structure, prosodic word structure, or grammatical morpheme only
25%, 50%, or 75% of the time, before finally exhibiting adult-like use of target
forms in obligatory contexts. Such variability in the development of phonological
and morphological structures is characterized differently across studies. At the
segmental level, some report first appearance of a sound, as well as the age at which
most children achieve more systematic use of a particular segment (Dodd, Holm,
Hua & Crosbie 2003; Ingram 1981; Smit 1993). However, a childs use of a particular
segment may depend, in part, on where in the syllable or word it occurs (e.g., onset
consonants vs. coda consonants). Thus, we might expect a particular segment to
Katherine Demuth
be more reliably produced in some phonological/prosodic contexts than others.

However, until recently, little was known about the possible phonological/prosodic
context effects on childrens early acquisition of segments, how this interacts with
the use of particular syllable/word structures, and the implications this has for the
development of phonological representations.
The same situation exists for studies reporting on childrens use of grammati-
cal morphemes. For example, in the classic study of the emergence of grammatical
morphology in the speech of Adam, Eve and Sarah, Brown (1973) documented
when each child had acquired each grammatical morpheme. This was measured
in terms of 95% percent use in obligatory contexts across three consecutive record-
ing sessions. This is a fairly rigorous test of morpheme use, and much higher than
that used by others in the field (e.g., above 80% use overall is typically considered
quite good e.g. Demuth & McCullough 2009). This raises the question, then, of
what kind of morphological representations a child may have when a grammatical
morpheme is produced only 50% of the time. Should this be considered as not
having any representation at all (i.e. random use)? Many researchers have sug-
gested that such behavior implies a lack of adult-like syntactic structure (Radford
1990) or a lack of semantic understanding about the use of such morphemes
(Hyams 2007). However, research by Gerken and McIntosh (1993), Gerken (1996)
and Demuth (1994; 2001) and colleagues suggests that the phonological/prosodic
context in which a grammatical morpheme appears can influence the likelihood
that a child will produce it. This could therefore account for much of the variabil-
ity found in childrens early productions. This in turn raises the question of what
children actually know about the structure of their language, and when. In par-
ticular, it suggests that, just as discourse context is essential for assessing childrens
knowledge of syntax, so too prosodic context plays a critical role in assessing the
nature of childrens developing phonological and morphological representations.
The implications for understanding the mechanisms underlying phonological and
morphological development, as well as for assessing the language abilities of those
with language delay (phonological delay, SLI, hearing loss, etc.), is enormous.
This paper reviews what is known about prosodic effects on childrens devel-
opment of phonological and morphological representations, focusing on interac-
tions at the levels of the mora, syllable, foot, prosodic word, and phonological
phrase. In particular, it shows that the acquisition of phonology goes far beyond
the acquisition of segments alone, interacting with many other levels of prosodic
structure. This is perhaps most clearly illustrated with reference to the Prosodic
Hierarchy, as developed by Nespor and Vogel (1986) and Selkirk (1984; 1996).
Viewed from this perspective, it is perhaps not surprising that the acquisition of
phonology takes years to master, playing an important role in understanding chil-
drens planning and production of words, morphemes, and utterances.
Prosodic Licensing and development
Interactions at the segmental/prosodic interface
A childs use of a particular segment (phoneme) may be influenced by the prosodic

structure in which it occurs. Given that English exhibits word-minimality effects
(a word must contain two moras (i.e. a foot) of structure (Prince & Smolensky 2004)),
we wondered if perhaps children would also be more likely to preserve a coda con-
sonant in the context of a preceding short vowel (sit) compared to a preceding long
vowel (seat). If so, this would provide support for the observation that early codas are
also prosodically licensed, being more likely to occur when they are required to pre-
serve word-minimality. Data from 2-year-olds using an elicited imitation task (with
picture prompt) suggests that this is the case, with children more likely to preserve
the coda when the preceding vowel is short (monomoraic) than long (bimoraic)
(Miles, Cox, Yuen & Demuth in submission).
Segments are also variably produced depending on whether they occur at
the beginning or end of a word or syllable. Children typically acquire onset con-
sonants before coda consonants, such that a high frequency segment such as /t/
might be realized in a word like top, but not in a word like cat, even though /t/
tends to be the first coda consonant acquired in English (cf. Kehoe & Stoel-
Gammon 2001; Stites, Demuth & Kirk 2004; Zamuner, Gerken & Hammond
2005). Similarly, a significant number of children show acquisition of // in
Brazilian Portuguese in syllable-initial-within-word position long before they
acquired the same sound in syllable-final-within-word position, and stress does
not seem to be a factor (Yava 1988).
Positional effects can be found in complex clusters. For example, /s/ tends to
appear earlier in childrens coda clusters in a word like box or wasp than it does in
onset clusters such as sky or spot (Kirk & Demuth 2005). Thus, the position in
which a target segment appears in a syllable or word can have a major effect on the
likelihood that it will be produced at a certain stage of development.
This is all the more interesting since we tend to think of these segments being
the same phoneme, regardless of the context in which they appear. Although there
are obvious acoustic differences in the realization of an onset vs. coda stop (e.g.,
VOT vs. closure, etc.) and an onset vs. coda fricative (e.g., differences in frication
duration), the assumption is typically that this should not impact on the realiza-
tion of these segments. This might, however, be a factor for the acquisition of /l/,
where some consider light (onset) /l/ and dark (coda) /l/ to be two different seg-
ments (cf. leap vs. peel). Indeed, Smit (1993) suggests that /l/s are first acquired in
onset position around the age of 4, and only later acquired in coda position, around
the age of 6. This may be due to the challenges of learning to coordinate the two
articulatory gestures needed to produce this segment in these two different pro-
sodic contexts (Lin & Demuth 2013). However, for most other consonants, this is
Katherine Demuth
less of an issue. Thus, all else being equal, we might expect the acquisition of a seg-
ment to occur simultaneously across all positions in a word. That this is not the
case suggests that the use of a particular segment at a specific point in develop-
ment is heavily influenced by the prosodic context in which it occurs.
Further support for this claim comes from the observation that word-internal
coda consonants tend to appear more often in stressed compared to unstressed
syllables in English (Kirk & Demuth 2006), and similar findings are reported for
Spanish (Lle 2003). Thus, the coda consonant /k/ in a nonce word is more likely
to be produced when it occurs in a syllable that is stressed (e.g., BAKnal) com-
pared to the same sequence when it is unstressed (e.g., bakNAL). A possible expla-
nation for this finding is that the stressed syllable is longer in duration than the
unstressed syllable, providing the young child with additional time to produce the
coda consonant. This is apparently confirmed by the observation that English-
speaking children are also good at producing coda consonants in word-final
position, even when these occur in an unstressed syllable (e.g., NALbak). In a non-
word imitation task, the word-final consonant is also phrase-final, and therefore
subject to phrase-final lengthening (Lehiste 1972), providing more time for the
child to fully articulate the coda. Thus, both stressed and final syllables facilitate
coda production, whereas unstressed word-medial coda consonants are more like-
ly to be omitted in childrens early speech. Once again, the production of a segment
interacts with the prosodic environment, being influenced not only by syllable
structure, but also prosodic word structure, stress, and phrase-final lengthening.
Interactions at the morphology/syllable structure interface
Many inflectional morphemes in English are encoded with a consonant, with /s, z/
and /t, d/ being the most frequent (e.g., plural, 3rd person singular, past tense). The
plural is typically acquired early, perhaps due to its high frequency in the input
children hear (around 75% of all -s morpheme tokens and types are plurals vs. 20%
for 3rd person singular). However, the acquisition of tense morphemes has been
notorious for the variable and protracted acquisition patterns found, leading to
proposals that these are syntactically difficult (Radford 1990). However, Marshall
and van der Lely (2007) found that SLI childrens use of the past tense morpheme
was worse with increasing phonotactic complexity in the coda. Thus, the produc-
tion of the past tense morpheme in a word like sewed was good, where the mor-
pheme was the only coda consonant. However, when the coda was more complex
(e.g., kicked), performance diminished, and was even lower in a three coda cluster
(e.g., danced). These findings suggest that the more complex the syllable structure,
the lower the use of the grammatical morpheme.
Similar findings have now been reported for the third person singular mor-
pheme, where typically developing 2-year-olds are more likely to produce the
morpheme in a simple coda (sees) compared to a complex coda (hits), in both
spontaneous speech and in elicited imitation tasks (Song, Sundara & Demuth
2009). Using a similar task, Theodore, Demuth and Shattuck-Hufnagel (2011)
found that the plural morpheme was generally preserved in the context of a com-
plex coda, but that cluster simplification also occurred (pigs > piss). This was also
occasionally found in the case of possessives (Mealings & Demuth, in press-a).
Thus, though the effects of the complex coda are seen, they may differentially af-
fect the various segments of the coda cluster depending on the robustness of the
morpheme being acquired. Since the plural is earlier acquired, and the morpho-
logical representation therefore more robust, cluster simplification leaves the
morpheme intact at the expense of reducing the consonant of the lexical form.
Interestingly, this tends to happen much more when the target word occurs utter-
ance medially compared to utterance finally.
These findings are interesting in light of a recent study examining the produc-
tion of morphemic vs. non-morphemic coda clusters (e.g., rocks vs. box) (Song,
Demuth, Shattuck-Hufnagel & Menrd 2013). It was found that 2-year-olds used
different articulatory gestures in producing the two different types of /ks/ coda
clusters, with /k/ appearing to be the articulatory target in the lexical item box, but
the /s/ appearing to be the articulatory target in the morphologically complex
rocks. This would be consistent with the Theodore et al. findings above, where
morphemic -s is retained at the cost of omitting the coda of the lexical base. This
suggests that the representation of morphemic inflectional morphemes differs
from that of tautomorphemic clusters, even at the early age of 2. Perhaps children
of this age are already prosodifying this morpheme at the level of the prosodic
word, above the level of the lexical item itself (cf. Goad, White & Steele 2003;
Selkirk 1996). Or perhaps these findings have more to do with lexical access and
online processing of morphological composition. This finding points to the need
for a developmental model of speech planning and production in order to better
understand the nature of childrens developing phonological and morphological
representations, and the implications this has for understanding how and why
children produce the forms they do.
Interactions at the morphology/phrasal interface
Many of the above effects (coda/morpheme omission, cluster reduction) are found
especially in utterance medial position, and less so utterance finally (cf. Mealings
& Demuth in press-b). Recall that coda consonants are more likely to be produced
Katherine Demuth
in the context where syllables have a longer duration, i.e. in either a stressed syl-
lable or word and phrase finally. In the study of childrens spontaneous use of the
3rd person singular -s, it was therefore interesting to find that 23-year-olds were
more likely to produce the morpheme utterance finally than utterance medially
(Song et al. 2009). We suspect that this is also due to the fact that phrase final posi-
tion, where the final syllable is longer than others, affords more time to produce
the entire syllable. Thus, the coda/morpheme that occurs phrase finally will be
more likely to be produced.
We might then also expect that morphemes that tend to occur phrase finally
tend to be more perceptible. If so, it may be easier for children to learn such mor-
phemes, since they can be better perceived. This in turn should enhance encoding
of these morphemes in the lexicon, thereby facilitating subsequent production as
well. Since English is an SVO language, verbs tend to occur phrase medially,
whereas nouns tend to occur phrase finally. In fact, corpus counts suggest that, on
average, about 75% of 3rd person inflected verbs occur in phrase medial position,
in both child and child-directed speech (Song et al. 2009). This means that only
25% of inflected verbs occur in the privileged phrase final position. In contrast,
nouns occur in this phrase final position at least 50% of the time. Since nouns are
also more frequent than verbs, the child hears many more plurals in phrase final
position, facilitating encoding of the morphemes in this position. Perhaps, then, it
is not surprising the plural morphology is acquired before verbal tense/agreement
morphology, given the different prosodic contexts in which plurals tend to appear
(Hsieh, Leonard & Swanson 1999; Song et al. 2009).
In order to test this perceptual hypothesis, Sundara, Demuth and Kuhl (2011)
conducted an infant speech perception/looking study to determine if children around
the age of 2 years notice the difference between grammatical and ungrammatical
forms (where the 3rd person singular morpheme is missing). Indeed, children no-
ticed the difference, showing a difference in looking time between the grammatical
and ungrammatical sentences in phrase final position (e.g., Now she cries vs. *Now she
cry). However, when the verb was embedded phrase medially, children did not show
a looking time difference between the two forms (e.g., She cries now vs. *She cry now).
This indicates that it is more challenging for children to perceive the 3rd person sin-
gular morpheme in utterance medial position the context where it typically appears.
These results suggest that learning verbal inflections may be delayed due to prosodic
context effects. Thus, although learning about tense and agreement may be semanti-
cally more challenging than learning about number and plurality, the fact that the
plural morpheme occurs both more often, as well as in a perceptually more salient
context, may help explain why it tends to be earlier produced.
We have now replicated this utterance medial effect in several follow-up stud-
ies with a variety of morphemes. This effect has been found with plurals when they
were part of a consonant cluster (Theodore et al. 2011; 2012) (as mentioned above),
and in -es /z/ forms of the 3rd person singular as well. Thus, controlling for word
length by using CVCz matched words (possible in a dialect like Australian
English: ladders vs. buses), 2-year-olds are less likely to produce the full /z/ in
buses (but not farmers) when the word is embedded in utterance medial compared
to utterance final position (e.g., The buses came vs. See the buses) (Mealings, Cox &
Demuth 2013). Though one might think that this could be a particular problem
with producing a fricative+schwa+fricative sequence, this appears to be a more
general problem of producing a C1+schwa+C1 sequence, since it appears to gener-
alize to the past tense morpheme as well. For example, in a study with 45-year-old
children diagnosed with SLI, all syllabic morphemes are particularly challenging,
with very low use across morphemes (e.g., 3rd person singular: catches, possessive:
horses, and past tense: added). Interestingly, this problem appears in both verbal
and nominal morphemes, and is not restricted to fricative contexts. It therefore
seems more like an OCP effect in the context of a reduced vowel.
Phrase medial effects are also seen for the production of 3rd person singular -s
when sentence length is manipulated. When 3-year-olds were asked to repeat
3-word and 5-word sentences in the context of a visual prompt, there was no effect
on morpheme production utterance finally, with near ceiling performance for both.
However, there was a large drop in performance utterance medially, with the mor-
pheme produced in the longer 5-word utterances only 48% of the time (e.g., He sits
back vs. He sits back and swings) (Mealings & Demuth in press-b). Thus, with in-
creased grammatical complexity, these children were much more likely to omit the
grammatical morpheme, but only in utterance medial context. Follow-up acoustic
analysis of both the prompt children heard, and childrens own productions, found
no difference in the duration of these morphemes within the medial context for the
3-word and 5-word conditions. This then suggests that the effect found utterance
medially was truly the combination of shorter duration plus the increased gram-
matical complexity and processing load needed to plan the rest of the (longer)
sentence. (cf. Valian 1991). Recall that all these elicited production tasks are carried
out in the context of a supportive picture, thereby reducing the need to remember
what was said. However, planning for the upcoming words nonetheless appears to
tax these childrens processing abilities, resulting in fragile grammatical morphemes
being omitted (see Valian (1991) for further discussion of such effects).
Interactions at the morphology/prosodic word interface
Research in the early 1990s began to report that childrens variable use of gram-
matical morphemes such as articles could be conditioned by prosodic context
Katherine Demuth
(e.g. Gerken & McIntosh 1993; Demuth 1994: Gerken 1996). In a series of elicited
imitation experiments, Gerken (1996) showed that 2;3-year-olds were more likely
to produce an article when it followed a monosyllabic verb than a disyllabic verb.
Thus, use of the object article was significantly higher in sentences like [Tommy]
[kicks the] [rabbit] than in a sentence like [Tommy] [catches] the [rabbit]. Note that
in the first sentence, the article can be prosodified with the previous word to form
a stressed-unstressed (Sw) trochaic foot. But this is not possible in the second sen-
tence, since the disyllabic verb catches is already a Sw foot. This results in the article
being left unfooted, where it is the subject to omission. This is very similar to the
processes that underlie childrens omission of unfooted syllables in lexical items
like banana (> nana), though the two are probably prosodified at different levels of
structure (at the level of the phonological phrase for the article, and the level of the
prosodic word for the unfooted syllable in the lexical item). Interestingly, both
processes tend to disappear around the age of 2;6, at least in English (Demuth
1996; Pater 1997). This strongly suggests that these are more general processes that
operate on childrens phonological representations at the level of the prosodic
word and phonological phrase, where unfooted syllables can only be incorporated
once these forms are permitted in the childs phonological grammar (Demuth
1996; Gerken 1996).
We therefore wondered if the same patterns Gerken (1996) reported for elic-
ited production experiments would be found in childrens spontaneous speech. To
examine this issue we collected data from 6 13-year-old children, audio/video
recording their speech as they interacted with their mothers for approximately one
hour every two weeks over a period of two years. The data were then orthographi-
cally and phonemically transcribed, with sound files and videos attached (see the
Providence Corpus (Demuth, Culbertson, & Alter 2006), CHILDES database,
http://childes.psy.cmu.edu/). We then coded the data for footed vs. unfooted pro-
sodic contexts, and whether the article was produced or not. As in the Gerken
(1996) study, we found that children were much more likely to produce those ar-
ticles that occurred in a footed context, and only started to acquire unfooted arti-
cles several months later (Demuth & McCullough 2009). Thus, it appears that the
elicited imitation procedures tap nicely into childrens phonological abilities in this
domain as well.
Lle and Demuth (1999) took this further, showing that crosslinguistic differ-
ences in the rate at which children acquire articles could be attributed to language-
specific differences in how these are prosodified. In particular, they showed that
articles begin to appear much later in German than in Spanish. German articles
take the form of an independent prosodic word (e.g., das the). This means that the
child who wants to say the ball in German must produce two independent pro-
sodic words. In contrast, Spanish articles are clitics that are prosodified with the
following word (e.g., la+mesa the table), resulting in a wSw prosodic word where
the article is earlier acquired. Since Spanish has many 3- and 4-syllable words (e.g.,
mueca doll, eskalera stairs), childrens early prosodic word representations al-
ready contain two or three syllables by the age of 2 or before (Gennari & Demuth
1997). This led Demuth, Patrolia, Song, and Masapollo (2012) to suggest that the
prosodic structure of the lexicon also plays an important role in determining when
articles will be acquired. Thus, articles in Spanish are prosodically licensed early,
being incorporated into three syllable structures, even at the expense of dropping
a syllable in the lexical item itself (e.g., la+ mueca > a+meca the doll). The result
is that articles are acquired earlier in Spanish (around 1;8 years) compared to a
year later in German. Thus, the prosodic structure of an article, as well as the pro-
sodic structure of the lexicon, can both influence when articles may be acquired.
Connelly (1984) was the first to note that children learning the southern Bantu
language Sesotho tended to omit noun class prefixes on the (mostly) disyllabic
nouns they produced, but not when the nominal root was monosyllabic. Thus, the
same noun class prefix would be either omitted or produced depending on the
syllable count of the nominal root (e.g., (mo)-sadi woman vs. mo-tho person).
This is consistent with the view that noun class prefixes are produced when they
can be prosodified as part of a disyllabic, trochaic foot. Further quantitative analy-
sis showed that this was indeed the case (Demuth & Ellis 2009), and that this ten-
dency disappears around the age of 2;32;6 (Demuth, Machobane & Moloi 2009).
These findings appear to generalize across neighboring Bantu languages (isiXhosa,
Setswana), suggesting that this is an important early stage of development in Bantu
languages more generally, with early variable noun class prefix use due to pro-
sodic rather than semantic or syntactic constraints.
Note that Sesotho noun class prefixes, like Spanish and French articles, pro-
sodically cliticize to the following noun. Sesotho also has penultimate lengthening
at the end of a phonological phrase, somewhat similar to the default lexical penul-
timate stress of Spanish. This raised the question of what would happen in a lan-
guage like French, where the final syllable of a phonological phrase is lengthened,
resulting in an unbounded iambic foot. Would articles (and determiners more
generally) be prosodically licensed in this type of a prosodic context as well? To
explore this issue we collected longitudinal audio/video data from 4 French-
speaking children and mothers from 13 years, resulting in the Lyon Corpus
(Demuth & Tremblay 2008). Like the Providence Corpus, it now resides in the
CHILDES database. Analysis of the data followed similar methods as that used for
the Spanish and Sesotho studies, examining the contexts were articles/determiners
should be used, and the number of syllables that occurred in the following word.
The results showed that, like their Sesotho-speaking counterparts, the French-
speaking children were much more likely to use determiners when the following
Katherine Demuth
word was monosyllabic (e.g., du lait the milk) compared to disyllabic (e.g., la
couronne the crown) (Demuth & Tremblay 2008; see Veneziano & Sinclair (2000)
for similar findings). Thus, the prosodic licensing of determiners appears to be
independent of foot directionality, appearing at an early stage of development
crosslinguistically. Interestingly, there seems to be a universal tendency to produce
determiners first as part of a foot, and only later at a higher level of structure.
Discussion
This paper has reviewed recent findings suggesting that much of the early vari-
ability found in childrens production of segments and morphemes is due to
Prosodic Licensing effects. That is, young children are more likely to use coda
consonants and grammatical morphemes in phonologically unmarked environ-
ments, where they are required by the grammar to meet word-minimality con-
straints, where they can form part of a (disyllabic) foot, or where there is more
time to actually produce an inflectional morpheme, such as in the durationally
longer syllable at the end of a phonological phrase.
It is well known that frequency effects and processing load also influence the
likelihood that a child will use a particular segment, syllable structure or mor-
pheme (Levelt, Schiller & Levelt 2000; Roark & Demuth 2000; Valian 1991). Re-
cent studies manipulating utterance length show that these effects are more often
observed utterance medially (Mealings & Demuth in press-a). We suspect that, in
addition to better semantic transparency, one of the reasons English plural mor-
phology is learned earlier is due to the fact that plurals are not only much more
frequent than English verbal inflectional morphemes, but that nouns (and there-
fore plurals) tend to occur more often in the phonetically more salient phrase-
final position, where the frication on the morpheme is durationally longer (Hsieh
et al. 1999; Song, Demuth, Evans, & Shattuck-Hufnagel 2013). This leads to great-
er perceptual salience, facilitating encoding of the morpheme in the lexicon,
thereby making it easier to retrieve and produce. Thus, although many other fac-
tors (frequency, processing load, etc.) may also contribute to a childs variable use
of a consonant or morpheme, the phonological/prosodic environment is critical
for predicting where this might be more likely to occur. Knowing about the pro-
sodic structure of a language then facilitates making crosslinguistic predictions
about how a particular form will be realized at early stages of phonological devel-
opment. We anticipate these findings will also be highly relevant for assessing
persistent problems of variability in the acquisition of language in other popula-
tions, including early L2 learners/bilinguals, children with SLI, and children with
hearing loss.
What, then, do we make of the 50% use of a segment or morpheme? If it is

systematically used in the easy phonological contexts, this would indicate that the
child does have a representation, but that it may not be as robust as at 75% or 100%
use in obligatory contexts. This suggests that our notion of acquired should more
graded, rather than all or nothing. This would be more consistent with a more
probabilistic, constraint-based type of learner than the more traditional, parame-
ter-setting type of learner. Thus, we can think of 50% as meeting some of the con-
straints, but not others. This is consistent with the notion that a child will be more
likely to use a particular form in a particular context. Thus, all else being equal, we
can expect that more children will produce a particular form in a particular con-
text. Of course, all things are not always equal in spontaneous speech, leading to
the types of variability documented here. Using more controlled experiments,
however, we can explore the nature of these competing constraints, and how they
interact with others in the process of language learning. This is exactly where the
intersection of (for example) processing load and phrase medial position effects
are found, with lower performance at the intersection of these two conditions.
The Prosodic Licensing Hypothesis thus provides a general framework for ex-
ploring the nature of developing grammars across languages and populations. If
children can use a particular segment, syllable structure or morpheme in a pro-
sodically licensed, easy, phonologically unmarked structure, this provides some
assurance to the parent, researcher or clinician that the child has some knowledge
of the phonology/phonotactics and/or syntax/semantics of the form. In the case of
language delayed populations, this provides some evidence that learning to use the
form in other contexts should develop as the childs phonological competence and
working memory increase. If, on the other hand, the use of such forms in the pho-
nologically easy contexts is systematically missing, this may provide evidence that
a different type of intervention is needed.
In sum, learning the phonology of a language is a complex task that takes years
to complete. The Prosodic Licensing Hypothesis provides a framework for explor-
ing how this process develops, leading to new discoveries about the acquisition of
phonology along the way.
Acknowledgements
I thank Dan Dinnsen for taking the initiative to intervene at a critical point in my
linguistics career, thereby making this research possible. I also thank all my won-
derful collaborators over the years, as well as Rebecca Holt for assistance with this
paper. This research has been supported, in part, by funding from the following
grants: ARC CE110001021 and NIH R01HD057606.
Katherine Demuth
References
Brown, R. 1973. A First Language. Cambridge MA: Harvard University Press.

CHILDES database <http://childes.psy.cmu.edu/>
Connelly, M. 1984. Basotho Childrens Acquisition of Noun Morphology. PhD dissertation,
University of Essex.
Demuth, K. 1994. On the underspecification of functional categories in early grammars. In
Syntactic Theory and First Language Acquisition: Cross-Linguistic Perspectives, B. Lust, M.
Suer & J. Whitman (eds) 119134. Hillsdale NJ: Lawrence Erlbaum Associates.
Demuth, K. 1996. The prosodic structure of early words. In Signal to Syntax: Bootstrapping from
Speech to Grammar in Early Acquisition, J. Morgan & K. Demuth (eds), 171184. Mahwah
NJ: Lawrence Erlbaum Associates.
Demuth, K. 2001. Prosodic constraints on morphological development. In Approaches to Boot-
strapping: Phonological, Syntactic and Neurophysiological Aspects of Early Language Acquisi-
tion [Language Acquisition and Language Disorders 2323], J. Weissenborn & B. Hhle
(eds), 321. Amsterdam: John Benjamins.
Demuth, K., Culbertson, J. & Alter, J. 2006. Word-minimality, epenthesis, and coda licensing in
the acquisition of English. Language & Speech 49: 137174.
Demuth, K. & Ellis, D. 2009. Revisiting the acquisition of Sesotho noun class prefixes. In Cross-
linguistic Approaches to the Psychology of Language: Festschrift for Dan Slobin, J. Guo, E.
Lieven, N. Budwig, S. Ervin-Tripp, K. Nakamura & S. Ozalikan (eds), 93104. New York
NY: Psychology Press.
Demuth, K. & McCullough, E. 2009. The prosodic (re)organization of childrens early English
articles. Journal of Child Language 36: 173200.
Demuth, K, Machobane, M. & Moloi, F. 2009. Learning to license null noun class prefixes in
Sesotho. Language 85: 863882.
Demuth, K., Patrolia, M., Song, J.Y. & Masapollo, M. 2012. The development of articles in chil-
drens early Spanish: Prosodic interactions between lexical and grammatical form. Linguis-
tic Interfaces and Language Acquisition in Childhood, J. Rothman & Pedro Guijarro-Fuentes
(eds). First Language 32:1737.
Demuth, K. & Tremblay, A. 2008. Prosodically-conditioned variability in childrens production
of French determiners. Journal of Child Language 35: 99127.
Dodd, B., Holm, A., Hua, Z. & Crosbie, S. 2003. Phonological development: A normative study
of British English-speaking children. Clinical Linguistics & Phonetics 17: 617643.
Gennari, S. & Demuth, K. 1997. Syllable omission in Spanish. In Proceedings of the 21st Annual
Boston University Conference on Language Development, E.M. Hughes & A. Green (eds),
182193. Somerville MA: Cascadilla Press.
Gerken, L. 1996. Prosodic structure in young childrens language production. Language 72:
683712.
Gerken, L. & McIntosh, B. 1993. The interplay of function morphemes and prosody in early
language. Developmental Psychology 29: 448457.
Goad, H., White, L. & Steele, J. 2003. Missing inflection in L2 acquisition: Defective syntax or
L1-constrained prosodic representations? Canadian Journal of Linguistics 48: 243263.
Hsieh, L.I., Leonard, L.B. & Swanson, L.A. 1999. Some differences between English plural noun
inflections and third singular verb inflections in the input: The contribution of frequency,
sentence position, and duration. Journal of Child Language 26: 31543.
Hyams, N. 2007. Aspectual effects on interpretation in early grammar. Language Acquisition 14:
231268.
Ingram, D. 1981. Procedures for the Phonological Analysis of Childrens Language. Baltimore MD:
University Park Press.
Kehoe, M. & Stoel-Gammon, C. 2001. Development of syllable structure in English-speaking
children with particular reference to rhymes. Journal of Child Language 28: 393432.
Kirk, C. & Demuth, K. 2005. Asymmetries in the acquisition of word-initial and word-final
consonant clusters. Journal of Child Language 32: 709734.
Kirk, C. & Demuth, K. 2006. Accounting for variability in 2-year-olds production of coda con-
sonants. Language Learning and Development 2: 97118.
Levelt, C., Schiller, N.O. & Levelt, W.J. (2000). The acquisition of syllable types. Language Acqui-
sition 8: 237264.
Lehiste, I. 1972. The timing of utterances and linguistic boundaries. Journal of the Acoustical
Society of America 51: 20182024.
Lle, C. 2003. Prosodic licensing of codas. Probus 15: 25781.
Lle, C. & Demuth, K. 1999. Prosodic constraints on the emergence of grammatical morphemes:
Crosslinguistic evidence from Germanic and Romance languages. In Proceedings of the
23rd Annual Boston University Conference on Language Development, A. Greenhill, H. Lit-
tlefield & C. Tano (eds), 407418. Somerville MA: Cascadilla Press.
Lin, S. & Demuth, K. 2013. The gradual acquisition of English /l/. In Proceedings of the 37th An-
nual Boston University Conference on Language Development, S. Baiz, E. Goldman & R.
Hawkes (eds), 206218. Somerville, MA: Cascadilla Press.
Marshall, C.R. & van der Lely, H.K.J. 2007. The impact of phonological complexity on past tense
inflection in children with Grammatical-SLI. Advances in Speech Language Pathology 9:
191203.
Mealings, K.T., Cox, F., & Demuth, K. 2013. Acoustic investigations into the delayed acquisition of
the syllabic -es plural in 2-year-olds speech. Journal of Speech, Language, and Hearing Re-
search, 56, 12601271.
Mealings, K.T. & Demuth, K. In press-a. Cluster reduction and compensatory lengthening in the
acquisition of possessive -s. Journal of Child Language.
Mealings, K.T. & Demuth, K. In press-b. The role of utterance length and position in three-year-
olds production of third person singular -s. Journal of Speech, Language, and Hearing
Research.
Miles, K., Cox, F., Yuen, I. & Demuth, K. In submission. The prosodic licensing of coda conso-
nants in early speech: interactions with vowel length.
Nespor, M. & Vogel, I. 1986. Prosodic Phonology. Dordrecht: Foris.
Pater, J. 1997. Minimal violation and phonological development. Language Acquisition 6:
201253.
Prince, A.S. & Smolensky, P. 2004. Optimality Theory: Constraint Interaction in Generative Gram-
mar. Malden MA: Blackwell.
Radford, A. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Basil Blackwell.
Roark, B. & Demuth, K. 2000. Prosodic constraints and the learners environment: A corpus
study. In Proceedings of the 24th Annual Boston University Conference on Language Develop-
ment, S. Catherine Howell, Sarah A. Fish & Thea Keith-Lucas (eds), 597608. Somerville
MA: Cascadilla Press.
Selkirk, E.O. 1984. Phonology and Syntax: The Relation between Sound and Structure. Cambridge
MA: The MIT Press.
Katherine Demuth
Selkirk, E.O. 1996. The prosodic structure of function words. In Signal to Syntax: Bootstrapping
from Speech to Grammar in Early Acquisition, J.L. Morgan & K. Demuth (eds), 187213.
Mahwah NJ: Lawrence Erlbaum Associates.
Smit, A.B. 1993. Phonologic error distributions in the Iowa-Nebraska Articulation Norms Proj-
ect: Consonant singletons. Journal of Speech and Hearing Research 36: 533547.
Song, J.Y., Demuth, K., Evans, K. & Shattuck-Hufnagel, S. 2013. Durational cues to fricative
codas in 2-year-olds American English: Voicing and morphemic factors. Journal of the
Acoustical Society of America, 133: 29312946.
Song, J.Y., Demuth, K., Shattuck-Hufnagel, S. & Mnard, L. 2013. The effects of coarticulation
and morphological complexity on the production of English coda clusters: Acoustic and
articulatory evidence from 2-year-olds and adults using ultrasound. Journal of Phonetics 41:
281295.
Song, J.Y., Sundara, M., & Demuth, K. 2009. Phonological constraints on childrens production
of English third person singular -s. Journal of Speech, Language, and Hearing Research 52:
623642.
Sundara, M., Demuth, K. & Kuhl, P.K. 2011. Sentence-position effects on childrens perception
and production of English 3rd person singular -s. Journal of Speech, Language, and Hearing
Research 54: 5571.
Stites, J., Demuth, K. & Kirk, C. 2004. Markedness versus frequency effects in coda acquisition.
In Proceedings of the 28th Annual Boston University Conference on Language Development,
A. Brugos, L. Micciulla & C.E. Smith (eds), 565576. Somerville MA: Cascadilla Press.
Theodore, R.M., Demuth, K. & Shattuck-Hufnagel, S. 2011. Acoustic evidence for position and
complexity effects on childrens production of plural -s. Journal of Speech, Language, and
Hearing Research 54: 539548.
Theodore, R.M., Demuth, K. & Shattuck-Hufnagel, S. 2012. Effects of articulatory planning factors
on childrens production of plural -s. Journal of the Acoustical Society of America 132: 2000.
Valian, V. 1991. Syntactic subjects in the early speech of American and Italian children. Cogni-
tion 40: 2181.
Veneziano, E. & Sinclair, H. 2000. The changing status of filler syllables on the way to gram-
matical morphemes. Journal of Child Language 27: 461500.
Yava, M. 1988. Padroes na aquisicao da fonologia do portugues. Letras de Hoje 74: 730.
Zamuner, T.S, Gerken L.A., & Hammond, M. 2005. The acquisition of phonology based on in-
put: A closer look at the relation of cross-linguistic and child language data. Lingua 115:
14031426.
Covert contrast in the acquisition
of second language phonology

University of WisconsinMilwaukee
This paper reports results on the acquisition of the English /s/ /z/ phonemic
contrast by native speakers of Spanish. The central finding is that some of the
research participants exhibited a covert contrast between these segments in
their interlanguage productions. Acoustic analysis revealed that four of the
participants produced a statistically reliable distinction between English [s] and
[z], however, this difference was not perceived by the transcribers who were
phonetically trained, native speakers of English. The existence of a stage of covert
contrast in L2 phonology is eminently plausible, given the progressive nature of
phonological acquisition, and brings the learning of second-language contrasts
into conformity with findings in the areas of L1 acquisition and phonologically
disordered speech.
Introduction
Over the last few decades, research on the acquisition of target language (TL) sound
patterns by second language (L2) learners has relied almost exclusively (studies
such as Flege 1987 notwithstanding) on impressionistic, phonetic transcriptions to
document the learners progress in acquiring the TL phonology (Altenberg & Vago
1983; Broselow, Chen & Wang 1998; Carlisle 1998; Eckman 1981; Eckman &
Iverson 1994; Flege 1987; Hammerly 1982; Major 1994; Ritchie, 1968; among many
others). These transcriptions have constituted the data for determining the nature
of the interlanguage (IL) phonological system being acquired by the learners. In
many cases, the transcriptions have shown extensive neutralization of the TL con-
trasts being learned; in other words, the transcribers did not perceive a distinction
in some of the sounds that the research participants were producing.
Until some thirty years ago, there was also this same reliance on phonetic tran-
scriptions in research on child-language (L1) acquisition. However, since seminal
work in this area by Macken and Barton (1980), numerous studies on both normal
acquisition of their native phonology by children and on children with phonologi-
cal disorders have shown that the participants often produce contrasts that are not
perceived by the adult listeners/transcribers (Forrest, Weismer, Hodge, Dinnsen &
Elbert 1990; Gierut & Dinnsen 1986; Macken & Barton 1980; Maxwell & Weismer
1982; Scobbie, Gibbon, Hardcastle & Fletcher 2000). This phenomenon, in which
L1 learners produce a statistically reliable distinction between sounds that is not
perceived by adults, whether phonetically trained or not, is known as covert con-
trast (as opposed to overt contrast, which is perceived by transcribers). To para-
phrase Scobbie (1998), the idea behind covert contrasts is that the phonological
system of a language may be acquired independently of how that system is imple-
mented phonetically. We will have more to say on this matter below.
The purpose of the present paper is to report preliminary findings of an ongo-
ing investigation into the acquisition of L2 phonemic contrasts. The case at hand
centers on the acquisition of the English phonemes /s/ and /z/, as in sip versus zip,
respectively, by native speakers of Spanish. Though [s] and [z] both occur in
Spanish, these sounds are allophones of the phoneme /s/, with [z] occurring only
before voiced consonants within the same word or phrase, as in mi[z]mo same
and la[z] gatas the (female) cats. This distribution motivates a rule (or constraint)
to the effect expressed in (1) below.
(1) Spanish allophonic pattern of s-voicing
/s/ [voice]/___ [consonantal, voice]
The implementation of this process is variable, however, in that whether it takes
place is an apparent function of the rate of speech (Harris 1969; 1983), faster
speech favoring voicing; another aspect of the rules variability is that it is optional
(Hualde 2005). And inasmuch as voice onset time (VOT) is a continuum, the de-
gree or extent of the assimilatory voicing itself has been observed to be gradient
rather than categorical, ranging from partially voiced [s] through weakly voiced
[] to thoroughly voiced [z] (Bradley & Delforge 2006; Garcia 2013; Martnez-Gil
2003).1 As we will suggest below, such variation and gradience appear also to influ-
ence the English productions of our Spanish-speaking research participants, some
of whom produced a covert, rather than overt, contrast between English /s/ and
/z/. We present an acoustic analysis of their utterances showing that four of the
1. In describing the extent of the fricative-voicing phenomenon in Spanish, Martnez-Gil

(2003:57) remarks that I do not know of any compelling evidence suggesting that partial voic-
ing assimilation is a phonological property and not simply a fact of phonetic implementation. In
fact, most available descriptions clearly indicate that the process is gradient, and thus typical of
phonetic phenomena.
Covert contrast
fourteen participants produced a reliable distinction between these fricatives that

was not perceived by the native-speaker transcribers.
The paper proceeds as follows. In the next section, we place our study in con-
text by reviewing some of the literature on covert contrasts in the acquisition of
English by children who acquire their native language without problems (i.e., nor-
mal acquisition), and by children who are phonologically disordered. In accor-
dance with some of our recent work on the acquisition of L2 phonemic contrasts
(Eckman & Iverson 2013; Eckman, Iverson, Fox, Jacewicz & Lee 2011), we then
present a hypothesis about the role played by s-voicing for Spanish speakers learn-
ing English. This is followed by a description of the methodology by which the data
were gathered and the reporting of the results as they bear on the hypothesis. We
conclude with a discussion of our findings in light of the hypothesis, and with a
view toward some pedagogical implications for teaching TL phonemic contrasts.
Background
Covert contrast
Although earlier research had foreshadowed the idea that children acquiring their
native phonology were making statistically significant distinctions that were not
being perceived by adults (Kornfeld & Goehl 1974; Ohala 1974; Smith 1979), the
article by Macken and Barton (1980) is generally cited as being the seminal study
to report the stage of a covert contrast in the acquisition of phonology by children.
Theirs was a longitudinal study of the acquisition of the English voice contrast by
four monolingual children. The authors analyzed the productions of children be-
tween the ages of one year, four months (1;4) and two years, four months (2;4).
Separate frequency distributions were calculated for the phonemically voiced and
voiceless stops at each point of articulation, followed by tests of significance be-
tween the mean voice onset time (VOT) values for the voiced and voiceless conso-
nants. In stops, VOT delay or lag is defined as the time, measured in milliseconds
(ms), between the release of closure and the beginning of vocal cord vibration in
the following vowel (or sonorant consonant). VOT delay is a primary acoustic cue
in determining whether stop consonants are perceived by speakers as voiced or
voiceless (cf. Iverson & Salmons 1995): in phrase-initial position in English, stops
categorized as voiced have a mean VOT value of less than 20 ms. (thus are pho-
netically voiceless, and largely unaspirated), whereas stops categorized as voice-
less show an average VOT lag of 40 ms. or more (thus are prominently aspirated).
Based on analysis of the VOT values in their participants productions of initial
stops, Macken and Barton identified three stages of acquisition. The first was one in
which the children did not produce a voice contrast in any stop consonants, as the
VOT values for both voiced and voiceless consonants fell within the short lag range
of adult speech. In the second stage, the children produced a statistically significant
VOT contrast between the voiced and voiceless stops, but these values all fell with-
in the adult perceptual categories of English voiced stop phonemes. In other words,
the VOT distinction that the children were making was not sufficiently great to be
perceived by adults, but was nevertheless statistically reliable. The presence of such
a covert contrast stage suggested that the children were aware of the voicing con-
trast, although their implementation of VOT was not yet adult-like. The final stage
was one in which the childrens production of a VOT contrast resembled that pro-
duced by adults. Additionally, Macken and Barton found that the children acquired
the VOT contrast at some points of articulation before others.
In the ensuing years, there have been a number of studies on the acquisition of
covert contrasts among phonologically disordered children. One of the earliest
was by Gierut and Dinnsen (1986), who analyzed two children, aged 4;6 and 4;3.
Phonetic transcription of their utterances indicated that both children were pro-
ducing the same kinds of errors. Specifically, according to the transcriptions, both
children failed to make a voice contrast in word-initial stops. Based on an acoustic
analysis of VOT and closure duration, the authors found that one of the children
was producing statistically significant differences between initial voiced and voice-
less stops in both VOT and closure duration measurements, although these dis-
tinctions could not be perceived by adult listeners.
In the following two and a half decades, it seems to have become widely accepted
in research on both the acquisition of L1 phonology and on phonological disorders
that there is a need to move from listener-oriented to speaker-oriented data (Hewlett
1988). Numerous studies investigating stages of covert contrast have been carried
out in both research domains, on a myriad of phonological contrasts involving a
large number of acoustic cues, including, but not limited to, amplitude, differential
vowel duration, formant analysis, pitch and VOT. The reader is referred to Scobbie
(1998) for a thorough listing and review of a large number of such studies.
In sum, the ample research on the acquisition of covert contrasts in phono-
logical acquisition is sufficient to conclude that making such a contrast is a well-
documented, intermediate stage in acquiring phonemic distinctions. We suppose,
therefore, that covert contrasts should be attested in the acquisition of contrasts in
second-language phonology, too.
Allophonic splits
As we have detailed in other work, there is evidence that learning to contrast TL

sounds which are allophones of the same phoneme in the native language (NL)
Covert contrast
involves suppressing the NL allophonic rule and results in implicationally related

patterns of acquisition. We have shown this claim to be supported for the acquisi-
tion of the English /s/ // contrast by native speakers of Korean (Eckman &
Iverson 2013). In the current study, we present evidence that the same pattern of
acquisition occurs in the case of native Spanish speakers acquiring English /s/ and
/z/, and, further, that these patterns are also attested by learners who make the
contrast covertly.
In the case of native speakers of Spanish learning to distinguish [s] and [z]
phonemically, we assume that, in the early stages of English acquisition, the rule
(or implementation strategy) relating [s] and [z] as allophones of /s/ transfers into
the IL grammar and applies to the pronunciation of TL words, causing errors. This
IL rule is subject to the general phonological constraints that pertain to primary-
language grammars, including the derived environment effect (Kiparsky 1982), ac-
cording to which rules effecting the substitution of one phoneme for another apply
only in so-called derived environments.2 In the context of SLA, this means that an
NL rule such as (1) above, when transferred into the IL, begins by applying across-
the-board. The IL then moves through a stage in which the rules application
persists only in the inter-morphemic environment (i.e., applies only when the fol-
lowing voiced consonant is separated from /s/ by an intervening morpheme
boundary, as in seriousness), but is suppressed morpheme internally. Ultimately,
the application of the rule is suppressed by the L2 learner, and does not apply at all
in the IL. Conversely, rules relating allophones of the same phoneme may apply in
both basic and derived environments, without regard for morphological struc-
ture.3 Thus, derived environments consist in portions of words that contain a rep-
resentation to which a rule would apply inter-morphemically, i.e., the segments in
question are separated by a morpheme boundary, whereas basic environments are
found in mono-morphemic words which contain the appropriate segments for
application of the rule.
2. Derived environment refers to a context for the application of a phonological rule where the
crucial representation needed for the rule to be applicable includes a morpheme boundary. An
example of a derived environment in the case of (1) above would be a situation where a mor-
pheme boundary exists between /s/, the segment to which the rule applies, and the following
voiced consonant.
3. Cho (1999; 2001) has shown that, in Korean, the effect of allophonic palatalization on /n/
before /i/ is gradient, with a greater palatalizing effect when the following /i/ begins a new mor-
pheme than when it is in the same morpheme. It may be that the variable voicing of Spanish /s/
is implemented similarly, with generally more penetration of voicing into /s/ before a hetero-
morphemic voiced consonant than before one in the same morpheme. Though we do not have
data on that in Spanish, this parallels the pattern some of the L2 learners show in English, which
then might suggest transfer from Spanish.
The derived environment effect holds implications for IL phonologies in which

an L2 learner must acquire a TL contrast between two segments, e.g., English /s/
and /z/, that are allophones of the same phoneme in the NL. Transfer of the rule
relating these NL allophones, before the TL contrast between them has been estab-
lished in the IL, leads L2 learners to err across-the-board, applying the NL
allophonic rule in all environments where it can be applied, irrespective of mor-
phological structure. For Spanish-speaking learners of English, specifically, frica-
tive voicing would take place (incorrectly) both in mono-morphemic Christmas
and poly-morphemic seriousness. But as learners acquire the TL contrast in some
words (Christmas now with [s] rather than [z], business still with [z]), thereby in-
troducing these sounds into the IL lexicon as phonemes, application of the NL
allophonic rule becomes restricted to derived environments (seriousness still with
[z]).4 Ultimately, the learner may be able to suppress the application of the NL
allophonic rule altogether and thus acquire the contrast in all environments
(seriousness with [s] vs. noiseless with [z]). In view of the derived environment ef-
fect, however, a fourth, logically possible pattern is excluded, viz., that in which the
learner suppresses the application of the NL allophonic rule only in derived envi-
ronments (yielding seriousness with [s], but Christmas with [z]).
Given this background, the specific hypothesis we test here is the following:
(2) Hypothesis
Acquisition of the English /s/ /z/ contrast by Spanish-speaking learners
will be sensitive to morphological structure in a manner consistent with
the derived environment effect.
Two observations about this hypothesis bear mention. The first is that we will
determine whether our participants have acquired the /s/ /z/ contrast either
overtly or covertly. A participant will be credited with an overt contrast in a given
phonological environment (e.g., initially, inter-morphemically etc.) if the partici-
pant produces, on the basis of the phonetic transcriptions, at least 80% target-like
productions for both [s] and [z] in the specified environment.5 If, according to the
transcriptions, the participant fails to reach the 80% threshold on either or both [s]
and [z] in an environment, then we conclude that the participant lacks the relevant
overt contrast in that environment.
4. From a historical perspective, words such as Christmas and business consisted of two mor-
phemes, though there is no evidence that native speakers of English today treat them as consist-
ing of two morphemes. Moreover, the protocol we followed required the participants to add the
suffixes in question, -ness and -less, as they pronounced the word, thus forcing the participants
to make an utterance morphologically composite and therefore, derived.
5. The 80% threshold for acquisition of a structure in L2 acquisition was first used, to the best
of our knowledge, in Cancino et al. 1978, and has been employed extensively since that time.
Covert contrast
The second important aspect of (2) is that evidence of a derived environment

effect is the acquisition of the relevant contrast, either overtly or covertly, in only a
basic environment, or in both a basic and derived environment, but not in only a
derived environment. Thus, all of our participants should evince one of the acqui-
sition patterns outlined above, and none should exhibit the excluded pattern
(contrast only in a derived environment).
Methodology
Stimuli
In order to test the hypothesis in (2), we elicited productions on the /s/ /z/ con-
trast from fourteen L2 learners of English, all of whom were native speakers of
Spanish associated with the University of WisconsinMilwaukee. In an attempt to
enlist participants who had a relatively wide range of English proficiency, we re-
cruited volunteers for the research both from the English as a Second Language
Program and from native speakers of Spanish in the campus community. Thus, the
participants varied widely in age (18 to 49), in how long they had studied English,
in their length of residence in the United States, in their nation of origin and in
their overall command of English. All participants were paid a small fee for their
participation in the project.
Within this context it is important to point out that our hypothesis is indepen-
dent of the level of English proficiency for our research participants. In other
words, the hypothesis simply asserts that, with respect to the /s/ /z/ contrast, the
IL of any Spanish-speaking learner of English will fall into one of the three pre-
dicted stages of acquisition. Therefore, the hypothesis is testable regardless of the
English background of the participants.
The stimuli used to elicit the productions consisted of a set of 90 words, 60 of
which were targets (listed in the Appendix) and 30 of which were fillers. All
are existing lexical items in English, and each target word contained /s/ or /z/ in
one of three different positions in a morphologically basic word, and additionally
in a morphologically derived environment. In the morphologically basic words,
the positions of occurrence for the /s/ or /z/ are initially before a vowel (e.g., sip/
zip), medially following a vowel and before a voiced or voiceless consonant (e.g.,
Christmas/business), and word-finally following a vowel (e.g., pass/buzz). The
morphologically derived environment is the position following a vowel and pre-
ceding a voiced consonant at the juncture of another morpheme, either the suffix
-ness or -less (faceless/noiseless/seriousness).
Several custom programs were written in MATLAB for the purposes of the
present study. A program that controlled the recordings displayed on a computer
screen a set of pictures, clues, and commands, such as Wait or Speak, that were
designed to guide the participant and the experimenter through the elicitation of
each word. Words were elicited, not by giving their spelling, but by displaying an
image depicting the object or idea in question for both basic (picture, say, of a face
to elicit the word face) and derived forms (same picture, but with the cue, without,
appearing on the screen one-half second after appearance of the image to elicit the
word faceless). If participants did not immediately recognize the word or concept
being depicted, they were given on-screen clues, or definitions, and, if need be, a
recorded model of the words pronunciation. The participants were also given a
practice exercise with a different set of words using the same cues in order to ensure
that they could correctly produce the intended derived words with the appropriate
suffix, in this case, -less. The stimuli were presented in a pseudo-randomized order
in that all basic forms were elicited before their related derived forms. The elicita-
tions were recorded directly onto a hard disc drive at the sampling rate of 44.1 kHz.
Participants spoke into a head-mounted microphone at a distance of one inch from
the lips and produced the set of 90 words twice, both during the same session.
Transcriptions
The data were collected at the University of WisconsinMilwaukee and then trans-
ferred to another major, mid-western university via file transfer protocol where
they were transcribed by an assistant who was blind to the hypothesis and unaware
of the intended target segments. The transcriber listened to the utterances in ques-
tion and focused either on a consonant in word-initial position or a word-medial
consonant occurring before the suffixes -less or -ness. The transcribers task was
then to choose from a menu of several choices: (1) [z], (2) [d] voiced alveolar stop,
(3) [s], (4) [t] voiceless alveolar stop, or (5) other. If option (5) was selected, the
transcriber was also required to enter the segment or to make a comment on what
was heard. The completed transcriptions were then returned to UW-Milwaukee
where they were scored.
For the purposes of testing the hypothesis, we consider the basic environment
to be exemplified by two phonological contexts: in words containing [s] and [z] in
word-initial position before a vowel, and in words with those sounds in word-
medial position following a vowel and before a consonant. Derived environments
are found in words in which either [s] or [z] occurs before the suffix -less or -ness,
as in seriousness, noiseless, baseless, etc. 6
6. In frequently occurring morphemes, fossilization on the status quo ante pronunciation may
result in what appears to be the application of the rule in a basic environment (e.g., Christmas
with [z] rather than [s]), which at this stage we take to be lexicalized pronunciations. Support for
this interpretation would come from additional data involving novel words with morpheme-
internal pre-consonantal [s] (e.g., Islip, parsnip) or nonce words of the same type.
Covert contrast
We elicited a total of ten words with word-initial [s] and ten words with word-
initial [z], as well as an additional ten words with word-medial [s] and ten with
word-medial [z]. For the occurrence of [s] and [z] in the derived environment,
participants produced a total of twenty words in which target [s] occurred before
the suffix -less or -ness, and the same number of words in which target [z] occurred
before the same suffixes.7 A participant thus produced a total of 60 target words. In
order for a participants IL to be credited with having a contrast between [s] and
[z] in a given environment, the performance on the productions had to reach the
80% criterial threshold for both [s] and [z] in that environment, as already noted
above. If a participants target-like pronunciations reached the criterial threshold
on only one of the segments in a given environment, or did not reach criterion on
either segment, the participants IL grammar was scored as lacking the contrast in
that environment. For example, a participant had to produce [s] in at least eight of
the ten words in which [s] occurred in initial position before a vowel, and likewise
for [z], in order for the IL to be accorded having the /s/ /z/ contrast in the basic
environment.
Acoustic analysis
As we are also investigating whether any of our participants made a covert contrast
between [s] and [z], we performed an acoustic analysis on the relevant sound files.
We begin by considering the acoustic measures for distinguishing between [s] and
[z] in general, and then proceed to the particular parameters that we employed.
Previous literature has demonstrated the robust effect of voicing on the dura-
tion of fricative noise, with a longer period of frication occurring for voiceless than
for voiced fricatives. For example, in a corpus study, Crystal and House (1988)
found that the duration of voiceless fricatives was overall 47 ms longer than that of
voiced fricatives (97 ms for voiceless fricatives versus 50 ms for voiced). Similarly,
Stevens, Blumstein, Glicksman, Burton and Kurowski (1992) reported that the du-
ration of the voiceless fricative [s] was about 30 ms longer than that of the voiced
fricative [z] in intervocalic position (108 ms for the former, 78 ms for the latter).
The voicing of fricatives is also known to affect the duration of the preceding vow-
el. In English, vowels are typically about 100 ms longer before voiced than before
voiceless obstruents (House 1961; but also see Crystal & House 1988 who report
this difference only in utterance-final position).
7. We also elicited tokens of [s] and [z] in word-final position; however, we are not reporting
these data because none of the subjects even approached the criterial 80% threshold on these
productions.
Another important correlate of the voiced-voiceless contrast for fricatives is

the duration of voicing during the fricative noise. In other words, the amount of
overlap between the fricative noise and the period of voicing has been shown to be
systematically greater for voiced fricatives than for voiceless fricatives. This mea-
sure has been successfully used to examine, for example, the partial devoicing of
[z], which seems to be virtually universal in utterance-final or pre-pausal position
(e.g., Smith 1997). However, it should be noted that fully voiced fricatives with
100% overlap (i.e. fricatives in which the time of vocal cord vibration overlaps
completely with the period of frication) are relatively difficult to produce, and are
uncommon cross-linguistically. As the airflow from the lungs is interrupted by the
constant closing and opening of the glottis for voiced fricatives, it is physiologi-
cally difficult to maintain, at the same time, both voicing and the high velocity of
airflow necessary for the turbulent noise characteristic of a fricative.
The acoustic measure that we used for distinguishing the production of [s]
and [z] by our participants was the percent of voicing that overlapped with the
fricative noise for each token of these segments. For the target fricative consonant
in each word, we calculated the percent of the fricative noise duration during
which the vocal folds were vibrating. On this measure, zero percent indicates that
the frication noise did not overlap at all with the observable vocal fold vibration
(as indicated by periodicity in the waveform and vertical striations in the spectro-
gram). Alternatively, 100 percent indicates that the overlap between frication and
voicing was complete.
It is worth making several points about the acoustic measure that we em-
ployed. First, since some of the [s] and [z] segments that we analyzed occurred in
word-initial position, it was not possible to use the difference in the duration of the
vowel before a voiceless versus voiced fricative as a distinguishing measure. Sec-
ond, the measure that we employed is appropriate for our data in that the target
words were not controlled for the number of segments they contained; rather, the
words were chosen on the basis of how easily and recognizably they could be pic-
tured on a computer screen. Because the duration of segments varies depending
on the number of segments in the word (cf., e.g., Lehiste 1972), we used propor-
tional measurements rather than raw numbers. Thus, for any target segment, we
measured the percent of the fricative noise that overlapped with voicing.
Figure 1 below shows the examples of three different renditions of the same
word zee by three of our Spanish-speaking participants. Two graphs are repre-
sented in each of Figures (1a c), with the top half showing the waveform and the
bottom half showing the spectrogram. The vertical dotted lines in the graphs mark
the beginning of the fricative noise (A), the onset of voicing (B), and the end of the
fricative noise (C). In Figure (1a), the fricative noise of the [z] fully overlaps with
Covert contrast
(a) Full (100%) overlap (b) Partial overlap
0
0
6000
Frequency (Hz)
Frequency (Hz)
6000
3000
3000
0
0 0 0.1 0.2 0.3 0.4 0.5 0.6
0 0.1 0.2 0.3 0.4 0.5 0.6
Time (Sec) Time (Sec)
A B C
A/B C
(c) No (0%) overlap
6000
Frequency (Hz)
3000
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (Sec)
A B/C
Figure 1. Representative waveform and spectrogram for the word zee [zi] produced by
Spanish speakers. A indicates the beginning of fricative noise; B shows the beginning
of voicing; and C marks the end of fricative noise. The interval between B and C is
the fricative noise overlapping with voicing
voicing, as indicated by the simultaneous onset of fricative noise and the voicing,
with voicing continuing through to the onset of the vowel at approximately 0.25
ms. The second example, Figure (1b), shows partial overlap of the fricative noise
and associated voicing, as frication begins at about 0.1 ms and lasts until about
0.25 ms, and voicing begins at 0.15 ms. Figure (1c) exemplifies no overlap between
the fricative noise and voicing, as voicing does not begin until the onset of the
vowel. The acoustic coding was carried out by two trained assistants using Praat
(Boersma & Weenink 2005). Visual information from the spectrogram and wave-
form, as well as auditory information, were used to determine the beginning and
end of fricative noise and voicing.
Results
We report findings first with respect to whether the participants exhibited an overt
contrast between [s] and [z], then whether the participants evidenced a covert
contrast between these segments.
Table 1 shows the participants productions, according to the transcribers, ex-
pressed as a percentage of target-like performance, in words containing [s] and [z]
in the three relevant phonological environments: initially, medially and inter-
morphemically.
Table 2 translates the results from Table 1 into categorical representations as to
whether a given participants IL showed a voice contrast between [s] and [z] in his/
her English productions, again in the three specified phonological environments.
Table 1. Subjects performance, expressed as a percentage of target-like productions,

on [s] and [z] in word-initial position before a vowel, word-medial position following
a vowel and preceding a voiced consonant, and in inter-morphemic position before
the suffix -ness or -less
Subjects Word-initial Word-medial Inter-morphemic
[s] [z] [s] [z] [s] [z]
No contrast
3017 100 0 10 100 20 85
3018 100 0 80 40 50 20
3020 100 50 30 90 25 65
3021 100 0 0 100 5 85
3022 100 40 30 100 5 90
3025 100 0 80 30 80 15
3029 100 60 30 60 60 85
Contrast in basic environment

3019 100 100 0 100 30 95
3026 100 90 40 80 40 80
3027 100 80 20 50 40 95
3028 100 90 80 90 70 40
3030 100 90 30 70 45 100
3031 80 90 20 60 20 100
Contrast medially & in derived environment

3024 100 30 90 80 85 85
Covert contrast
Table 2. Subjects performance, according to whether or not they reached the criterial
threshold of 80% (Yes), or whether they fell below the threshold (No), on [s] and [z]
in initial position before a vowel, medial position following a vowel and preceding
a voiced consonant, and in inter-morphemic position before the suffix -ness or -less
Subjects Contrast initial Contrast medially Contrast inter-morphemically
No contrast
3017 No No No
3018 No No No
3020 No No No
3021 No No No
3022 No No No
3025 No No No
3029 No No No
Contrast in basic environment

3019 Yes No No
3026 Yes No No
3027 Yes No No
3028 Yes Yes No
3030 Yes No No
3031 Yes No No
Contrast medially & in derived environment
3024 No Yes Yes
The findings presented in Table 1 support the hypothesis, which claims that the
presence of the contrast in derived environments implies the contrast in basic en-
vironments, but not vice versa. All of the participants fall into one of the permitted
patterns and none of the participants evinces the excluded pattern. The first set of
participants in the table represents the no contrast group in that none of them
exhibits the /s/ /z/ contrast in any of the three environments. This is because there
are no scores at or above the 80% threshold on both [s] and [z] in any of these en-
vironments. The second group of participants in the table evidences the contrast in
the basic environment, specifically, in word-initial position, or in word-initial and
word-medial positions, but not in the derived (inter-morphemic) environment.
The final participant, 3024, is classified as having the contrast in both basic and
derived environments, with 3024 evincing the contrast in both a basic (word-medial)
and a derived (inter-morphemic) environment. Thus, the first three groupings
of the participants in Table 1 are consistent with the hypothesis in (2): seven par-
ticipants do not have the contrast in any environment; six participants show the
contrast in the basic environment only; and participant 3024 has the contrast in
both the basic and derived contexts. Moreover, none of the participants exhibits the
excluded stage of acquisition, i.e., showing the contrast only in the derived, inter-
morphemic environment while lacking the contrast in the basic environment.
The productions of participant 3024, while consistent with the hypothesis in
maintaining the contrast in both the derived and basic environments, are a bit of
an anomaly, because this participant maintains the contrast in word-medial posi-
tion, but lacks the contrast word-initially. The general expectation among pho-
nologists is that a word-medial contrast would also occur word-initially. We will
have more to say about this participants performance below.
We now focus on results showing that some participants make a covert con-
trast between English [s] and [z]. Findings are reported first for the group of par-
ticipants, as is customary, then for individuals, as is necessary in the context of our
hypothesis. Our hypothesis makes claims about the status of a learners IL gram-
mar, stating that acquisition of the /s/ /z/ contrast will exhibit a derived environ-
ment effect. IL grammars are mental systems whose placement in time and space
is in the mind of individual learners. Therefore, we must test such claims about the
state of an interlanguage grammar using individualized data, simply because there
is no IL grammar of a group of people, at least not one that can be situated in time
and space, just as there is no mind of a group.
Group results
In order to examine whether our native speakers of Spanish were making an acous-
tic distinction between [s] and [z], we compared the percent of the fricative noise
overlapping with voicing for [s] and [z] using paired t-tests. Before running these
t-tests, we examined the distribution of the data. Because the distribution turned
out to be skewed to the right, the data were log-transformed in order to better ap-
proximate a normal distribution. When transforming the data to logarithmic val-
ues, following convention, we first added a constant, 1, to all raw percentages (some
of which were 0, indicating no overlap) so that all percentages were greater than 0.
This procedure was to avoid dealing with the log of 0, which cannot be defined.
The group results for all of the participants for the paired t-tests using the log-
transformed data are shown in Table 3.
The findings suggest that the percent of fricative noise that overlapped with
voicing for [s] significantly differs from that for [z] word-initially and inter-
morphemically, suggesting that [z] is significantly more voiced than [s] in these
positions. However, the difference was not significant in the basic-environment of
word-medial position following a vowel and preceding a voiced consonant, i.e.,
where the fricatives were produced.
Covert contrast
Table 3. Group results. (Note: The mean and SD are based on raw percentages.
The statistical results are based on log-transformed data.)
[s] [z]
Position Mean SD Mean SD Statistics
Word-initial 2.27 5.66 43.90 38.76 t(13) = 5.87, p < 0.00

Word-medial 15.67 15.61 19.03 15.25 t(13) = 1.62, p = 0.13
Inter-morphemic 16.51 18.33 22.51 23.85 t(13) = 2.27, p < 0.05
Individual results
Next we examined which of the individual speakers showed a difference in the

degree of fricative voicing between [s] and [z]. To this end, we ran unpaired t-tests
for each speaker, using values from individual target words for each participant.
As with the group data, the raw percentages were log-transformed before running
the unpaired t-tests. Table 4 compares the results from the acoustic analysis and
phonetic transcriptions in word-initial position.
Table 4. Subjects performance on [s] and [z] in initial position before a vowel.
(Note: The mean and SD are based on raw percentages. The statistical results are based
on log-transformed data.)
Acoustical analysis results Transcription Contrast

results
[s] [z]
Mean SD Mean SD Statistics
3017 .28 .42 .15 .23 t(8) = .48, p = .64 No No

3018 .43 .58 5.88 12.94 t(8) = .66, p = .53 No No
3020 .00 .00 16.13 22.22 t(8) = 2.28, p = .05 No Covert
3021 .18 .34 .66 .66 t(8) = 1.42, p = .19 No No
3022 2.17 1.12 17.47 19.47 t(8) = 1.87, p = .10 No No
3025 .33 .53 5.57 11.88 t(8) = .87, p = .41 No No
3029 1.37 .58 47.96 27.23 t(8) = 5.65, p < .001 No Covert
3019 .00 .00 90.00 22.36 t(8) = 32.77, p < .001 Yes Overt
3026 .00 .00 90.00 22.36 t(8) = 32.77, p < .001 Yes Overt
3027 .56 .29 64.01 39.62 t(8) = 5.94, p < .001 Yes Overt
3028 1.10 1.86 75.29 36.22 t(8) = 8.20, p < .001 Yes Overt
3030 .00 .00 100.00 .00 t(8) = 2296.81, p < .001 Yes Overt
3031 21.59 43.76 85.84 21.75 t(8) = 3.58, p < 0.01 Yes Overt
3024 3.76 3.41 15.62 31.27 t(8) = .44, p = .67 No No

The results show that all 6 participants who reached the criterial threshold of 80%
(as indicated by Yes under Transcription results in Table 2) also showed a sig-
nificant difference in the percent of the fricative noise overlapping with voicing
between [s] and [z]. That is, these participants exhibited differences between [s]
and [z] in word-initial position both in the acoustic analysis and in the phonetic
transcriptions, thereby making the contrast overtly. In addition, there were two
participants who showed significant differences in the percent of the fricative noise
overlapping with voicing, but fell below the threshold of 80%: 3020, 3029. That is,
these participants produced a statistically reliable word-initial distinction between
[s] and [z] in terms of fricative-voicing overlap that was not perceived by tran-
scribers using the 80% criterion, indicating that they made the contrast covertly.
All other Spanish speakers showed no difference between word-initial [s] and [z]
on the basis of both acoustic analysis and phonetic transcriptions, suggesting that
they exhibited no contrast between the two sounds.
We turn now to the participants performance on [s] and [z] in the basic con-
text of medial position following a vowel and preceding a voiced consonant, shown
in Table 5. Twelve participants who did not exhibit the /s/ /z/ contrast in this
environment on the basis of the phonetic transcriptions also did not show a differ-
ence in the percent of the fricative noise overlapping with voicing between [s] and
[z]. However, there were two participants (3024, 3028) who, according to the pho-
netic transcriptions, scored at or above the 80% threshold on both [s] and [z] but
did not show differences acoustically. Though we did not perform additional
acoustic analyses on these participants, we infer that 3024 and 3028 must have
implemented the contrast between [s] and [z] in some other way that was per-
ceived by the transcribers.
Table 6 compares the results from the acoustic analysis and phonetic tran-
scriptions in inter-morphemic position. Although none of the participants (except
for 3024) reached the 80% threshold on the basis of the phonetic transcriptions,
three of the participants (3019, 3029, 3030) made the contrast covertly in this en-
vironment by showing a significant difference in the percent of fricative noise
overlapping with voicing between [s] and [z].
To summarize this section, the results of our acoustic analysis revealed that
four participants (3020, 3029, 3019, and 3030) maintained in their IL a voice con-
trast between [s] and [z] that was not perceived by the research assistants, and
therefore was not transcribed. In other words, these four participants evidenced a
covert contrast in that the percentages of voicing that overlapped with the fricative
noise for [s] and [z] were statistically different in tokens containing [s] and [z]
word-initially and inter-morphemically following a vowel preceding either the
Covert contrast
Table 5. Subjects performance on [s] and [z] in medial position following a vowel,
and before a voiced consonant. (Note: The mean and SD are based on raw percentages.

results
[s] [z]
3017 58.4 34.52 69.69 33.07 t(8) = .86, p = .41 No No

3018 9.43 7.18 10.58 6.99 t(8) = .94, p = .37 No No
3020 6.76 8.13 7.90 5.95 t(8) = .03, p = .98 No No
3021 9.49 4.69 19.34 21.50 t(8) = .26, p = .80 No No
3022 11.29 9.77 11.82 13.10 t(8) = .23, p = .83 No No
3025 2.24 2.10 3.39 3.09 t(8) = .24, p = .82 No No
3029 7.20 3.63 4.53 2.02 t(8) = .72, p = .49 No No
3019 25.84 30.84 62.46 33.05 t(8) = .18, p = .87 No No

3026 6.54 5.71 8.33 6.00 t(8) = 1.47, p = .18 No No
3027 21.26 4.01 19.67 5.04 t(8) = .51, p = .63 No No
3028 7.54 1.37 6.83 2.27 t(8) = .71, p = .50 Yes No
3030 5.26 2.59 21.43 24.51 t(8) = .40, p = .70 No No
3031 55.66 23.38 63.36 13.75 t(8) = .40, p = .70 No No
3024 4.26 2.53 5.75 5.06 t(8) = .38, p = .72 Yes No
suffix -ness or -less. To our knowledge, evidence of such a covert contrast in the
acquisition of second-language pronunciation has not been reported8.
Discussion
Though the findings of our study are subject to certain limitations (segmentally
disparate pairings, acoustic measures used are limited), they nevertheless point to
several interesting and important implications for both second-language acquisi-
tion theory and for L2 pedagogy.
As stated at the outset, the segments [s] and [z] occur in both English and
Spanish: in Spanish as allophones of the same phoneme (/s/), but as contrasting
8. The only exception to this claim that we have encountered is Lim and Oh (2007), a pub-
lished abstract from a conference, which clearly encompasses the idea of covert contrast in SLA,
but has not been widely circulated.
Table 6. Subjects performance on [s] and [z] in inter-morphemic position before

the suffix -ness or -less. (Note: The mean and SD are based on raw percentages.

results
[s] [z]
3017 59.24 34.16 39.55 42.02 t(18) = .75, p = .46 No No

3018 5.83 3.06 4.16 2.68 t(18) = .63, p = .53 No No
3020 11.88 9.92 10.94 7.99 t(18) = 1.28, p = .22 No No
3021 14.22 14.96 12.06 11.17 t(18) = 1.10, p = .29 No No
3022 11.20 6.26 13.45 9.12 t(18) = .47, p = .65 No No
3025 1.44 1.36 3.60 5.97 t(18) = .53, p = .60 No No
3029 9.29 3.69 7.90 4.04 t(18) = 2.08, p = 0.05 No Covert
3019 26.03 23.15 40.02 42.48 t(18) = 2.7, p < .05 No Covert
3026 5.98 4.23 37.31 32.67 t(18) = 1.07, p = .30 No No
3027 17.91 6.30 15.77 4.30 t(18) = .86, p = .40 No No
3028 6.87 1.06 10.5 9.48 t(18) = 1.02, p = .32 No No
3030 7.86 3.08 16.73 24.78 t(18) = 3.19, p < 0.01 No Covert
3031 36.53 35.62 48.14 39.04 t(18) = 1.15, p = .27 No No
3024 5.13 6.26 6.29 4.69 t(18) = .05, p = .96 Yes No
phonemes (/s/ versus /z/) in English. Native speakers of Spanish acquiring English
must therefore implement two major changes in their IL grammar relative to their
NL: first, they must develop the voiced sibilant fricative [z] as a phoneme in con-
trast to /s/, and second, they must suppress imposition of the NL allophonic pat-
tern in the IL. Transferring the NL distribution of [s] and [z] into a learners IL
would cause that learner to err on TL words containing [z] in all environments
except before voiced consonants. Specifically, such learners would err on all TL
words containing [z], except for those where that segment occurs before a voiced
consonant, as in business or prizeless. Thus, learning to contrast /s/ /z/ in the IL
of native speakers of Spanish would ostensibly entail their making an allophonic
split, that is, the learners must separate the NL allophones [s] and [z] into two
phonemes in the IL (Eckman & Iverson 2013). The case at hand, however, is com-
plicated by the fact that the Spanish allophonic pattern produced by (1), which
voices /s/ to [z] before voiced consonants, is optional, or gradiently variable, and
subject to rate of speech (Bradley & Delforge 2006; Harris 1983; Hualde 2005).
Covert contrast
As a consequence, native speakers of Spanish learning the English /s/ /z/

contrast would seem to have a greater possibility of splitting their NL allophones
[s] [z] into separate phonemes by suppressing the pattern resulting from (1) than
would L2 learners who have to suppress an allophonic rule that is obligatory
(Eckman & Iverson 2013). We would therefore expect that those research partici-
pants who have acquired a contrast between /s/ and /z/, in at least one position,
not to impose the NL pattern arising from (1). In other words, we would expect
that those learners who know that the TL contrasts /s/ and /z/ in, say, word-initial
position, would not transfer the NL pattern in (1) to the IL, because imposing this
pattern would neutralize the /s/ /z/ contrast. As we can see from Table 2, how-
ever, this is not the case, as participants 3019, 3026, 3027, 3028, 3030, and 3031
produce the contrast word-initially, yet except for 3019 and 3030, who show the
contrast covertly in inter-morphemic position they continue to follow the NL
pattern by voicing /s/ to [z] before a hetero-morphemic voiced consonant. The
behavior of these six participants would fall into place if it turned out that NL al-
lophonic patterns are not optional when implemented in an IL, or if their elicita-
tions were produced at a speech rate that caused the rule or constraint in (1) to be
invoked consistently. For now, we must leave the question open.
The second point we would like to make related to second-language acquisi-
tion theory concerns the derived-environment effect. As outlined above, the de-
rived-environment effect is a constraint according to which any learner acquiring
an allophonic split will make the relevant contrast in a derived environment only
if that learner also makes the contrast in a basic environment. This constraint ob-
tains for our participants, and interestingly, holds also for those participants
making a covert contrast. Thus, participants 3029, 3019 and 3030 evince a covert
contrast between [s] and [z] in inter-morphemic position. These same participants
also show that contrast in basic environments, either overtly, as in the case of par-
ticipants 3019 and 3030, or only covertly, as with 3029.
Within this context, we should point out an apparent anomaly in the produc-
tions of 3024. Although this participant maintains a contrast in both a basic and a
derived environment, the contrast in the basic environment occurs in word-medial
position but not, as we would expect, also in word-initial position. Because the
transcriptions for this participant ran counter to this widespread expectation that
a voice contrast in word-medial position would entail such a contrast word-initially,
the investigators listened to the sound files of the tokens containing word-initial
[s] and [z] for this participant. Whereas the words containing initial [s] were pro-
nounced by this participant with what was clearly a voiceless sibilant fricative, the
tokens containing initial [z], conversely, in most cases sounded more voiced than
voiceless, and in other instances were clearly voiced.9 However, because our stated
protocol was to accept the transcriptions as they came to us via file transfer proto-
col from the assistants at a Midwestern university, we nevertheless based our
findings on these renderings of the utterances. Further analysis may show that
participant 3024 nonetheless does evince a word-initial contrast between [s] and
[z], along with the noted word-medial contrast.
Pedagogical implications
We now turn to three pedagogical implications of our findings. The first is that the
relative difficulty that L2 learners have in splitting NL allophones into TL pho-
nemes was first pointed out, to our knowledge, by Lado (1957: 15), when he
claimed that this learning situation constituted maximum difficulty. Our findings
can shed some light on the explanation of this difficulty in that the derived envi-
ronment effect constrains learning such that acquisition of a contrast in derived
environments implies that the learner will have the contrast in basic environments.
This allows, of course, for the L2 learners to show the contrast in basic contexts, yet
lack the contrast (that is to say, to continue to err systematically) in derived con-
texts. We can only speculate at this point, but observation of this learning pattern
may have caused linguists and language teachers, including Lado, to take note of
this difficulty.
The second pedagogical implication is that acquisition of a phonemic contrast,
especially one involving an allophonic split, is a function of phonological environ-
ment. The fact that an L2 learner has acquired a TL phonemic contrast in a given
environment does not mean that the learner has also acquired that contrast in
some other environment. Our findings indicate the contrary: a learner may sys-
tematically evidence a contrast in one environment, and just as systematically lack
that contrast elsewhere. Moreover, our results also show that there is a relationship
between the existence of certain kinds of contrast and the environments in which
those contrasts occur.
The final pedagogical consequence of these findings is that learners may make
a phonemic contrast covertly. If our results, which still must be considered pre-
liminary, are viewed in the same light as work on covert contrasts in L1 acquisition
and phonologically disordered speech, then the presence of covert contrasts among
L2 learners points to an intermediate stage of acquisition. On this interpretation,
9. The protocol for the elicitation of the data from the subjects directed us, in cases where
subjects produced more than a single token of the target word, to accept only the last uttered
token. Subject 3024 produced two or more tokens of four of the ten words, and pronounced the
majority of them with a voiced sibilant fricative. If all of these utterances were considered, 3024
would have reached the 80% criterion in the pronunciation of word-initial [z].
Covert contrast
it would be expected that some L2 learners progress from a state of no contrast and
then pass through a stage of covert contrast before ultimately arriving at a (final)
stage of overt contrast. Indeed, an intermediate stage of covert contrast appears to
be necessary, as it takes time to learn, progressively, the phonetic implementation
of target phonemes.
Conclusion
This paper has reported findings indicating that the acquisition of an L2 phonemic
contrast may involve, for at least some learners, an intermediate stage of covert
contrast. Some L2 learners may implement a TL phonemic distinction in a way
that can be reliably measured acoustically, but which is not perceived by native
speakers of the TL, even those who are phonetically trained. The existence of an
intermediate stage of covert contrast in the learning of L2 phonology is eminently
plausible, in view of the progressive nature of this task, and brings the acquisition
of second-language contrasts into conformity with findings of the same phenom-
enon in the areas of L1 acquisition and phonologically disordered speech.
Acknowledgement
This work from was supported in part for the first two authors by a grant from the
National Institutes of Health 1 R01 HD046908-05. The positions expressed in this
paper are those of the authors, and do not necessarily reflect those of NIH. We
thank Carolyn Barry and Alison Garcia for their role in conducting the acoustic
analysis. Any remaining errors are the fault of the authors.
References
Altenberg, E. & Vago, R. 1983. Theoretical implications of an error analysis of second language
phonology production. Language Learning 33: 427447.
Boersma, P. & Weenink, D. 2005. PRAAT: Doing phonetics by computer, Version 4.4.07. <http://
www.praat.org/>
Bradley, T.G. & Delforge, A.M. 2006. Systemic contrast and the diachrony of Spanish sibilant
voicing. In Historical Romance Linguistics: Retrospectives and Perspectives [Current Issues in
Linguistic Theory 274], D. Arteaga & R. Gess (eds), 1952. Amsterdam: John Benjamins.
Broselow, E., Chen, S. & Wang, C. 1998. The emergence of the unmarked. Studies in Second
Language Acquisition 20: 261280.
Cancino, H., Rosansky, E. & Schumann, J. 1978. The acquisition of English negative and inter-
rogatives by native speakers of Spanish, in Second Language Acquisition, E. Hatch (ed).
Rowley MA: Newbury House.
Carlisle, R.S. 1998. The acquisition of onsets in a markedness relationship: A longitudinal study.
Studies in Second Language Acquisition 20: 245260.
Cho, T. 1999. Specification of Intergestural Timing and Gestural Overlap: EMA and EPG Stud-
ies, MA thesis, University of California, Los Angeles.
Cho, T. 2001. Effects of morpheme boundaries on intergestural timing: Evidence from Korean.
Phonetica 58: 129162.
Crystal, T.H. & House, A.S. 1988. Segmental durations in connected-speech signals: Current
results. Journal of the Acoustical Society of America 83: 15531573.
Eckman, F. 1981. On the naturalness of interlanguage phonological rules. Language Learning 31:
195216.
Eckman, F. & Iverson, G. 1994. Pronunciation difficulties in ESL: Coda consonants in English
interlanguage. In First and Second Language Phonology, M. Yavas (ed.), 251266. San Diego
CA: Singular.
Eckman, F., Iverson, G., Fox, R.A., Jacewicz, E. & Lee, S. 2011. Explicit training and implicit
learning of L2 phonemic contrasts. In Implicit and Explicit Conditions, Processes and
Knowledge in SLA Bilingualism, C. Sanz & R. P. Leow (eds), 159174. Washington DC:
Georgetown University Press.
Eckman, F. & Iverson, G. 2013. The role of the native language in the acquisition of L2 phonemic
contrasts. Studies in Second Language Acquisition 35: 126.
Flege, J.E. 1987. The production of new and similar phones in a foreign language: Evidence
for the effect of equivalence classification. Journal of Phonetics 15: 4765.
Forrest, K., Weismer, G., Hodge, M., Dinnsen, D. A. & Elbert, M. 1990. Statistical analysis of
word-initial /k/ and /t/ produced by normal and phonologically disordered children. Clini-
cal Linguistics and Phonetics 4: 327340.
Garcia, A. 2013. Allophonic Variation in the Spanish Sibilant Fricative. PhD dissertation, Uni-
versity of Wisconsin at Milwaukee.
Gierut, J.A. & Dinnsen, D. 1986. On word-initial voicing: Converging sources of evidence in
phonologically disordered speech. Language and Speech 29: 97114.
Hammerly, H. 1982. Contrastive phonology and error analysis. IRAL 20: 1732.
Harris, J. 1969. Spanish Phonology. Cambridge MA: The MIT Press.
Harris, J. 1983. Syllable Structure and Stress in Spanish. Cambridge MA: The MIT Press.
Hewlett, N. 1988. Acoustic properties of /k/ and /t/ in normal and phonologically disordered
speech. Clinical Linguistics and Phonetics 2: 2945.
House, A.S. 1961. On vowel duration in English. Journal of the Acoustical Society of America 33:
11741182.
Hualde, Jos Ignacio. 2005. The Sounds of Spanish. Cambridge: CUP.
Iverson, G. & Salmons, J. 1995. Aspiration and Laryngeal Representation in Germanic. Phonol-
ogy 12: 369396.
Kiparsky, P. 1982. Lexical phonology and morphology. In Linguistics in the Morning Calm, I.S.
Yang (ed.), 391. Seoul: Hanshin.
Kornfeld, J.R. & Goehl, H. 1974. A new twist to an old observation: Kids know more than they
say. In Papers from the Tenth Annual Meeting of the Chicago Linguistic Society, Parasession
on Natural Phonology, 210219. Chicago IL: CLS.
Lado, R. 1957. Linguistics Across Cultures. Ann Arbor MI: University of Michigan Press.
Covert contrast
Lehiste, I. 1972. The timing of utterances and linguistic boundaries. Journal of the Acoustical
Society of America 51: 20182024.
Lim, J.-A. & Oh, M. 2007. Covert contrast in second language acquisition. In Proceedings of the
16th International Conference on Korean Linguistics, International Circle of Korean Linguis-
tics (ed.), 202203. New York NY: SUNY at Binghamton & Cornell University.
Macken, M.A. & Barton, D. 1980. A longitudinal study of the acquisition of the voicing contrast
in American-English word-initial stops, as measured by voice onset time. Journal of Child
Language 7: 4174.
Major, R. 1994. Chronological and stylistic aspects of second language acquisition of consonant
clusters. Language Learning 44: 655680.
Martnez-Gil, F. 2003. Resolving rule-ordering paradoxes of serial derivations: An optimality
theoretical account of the interaction of spirantization and voicing assimilation in Peninsu-
lar Spanish. In Theory, Practice, and Acquisition, P. Kempchinsky & C-E. Pieros (eds), 40
67. Somerville MA: Cascadilla Press.
Maxwell, E.M., & Weismer, G. 1982. The contribution of phonological, acoustic, and perceptual
techniques to the characterization of a misarticulating childs voice contrast for stops. Ap-
plied Psycholinguistics 3: 2943.
Ohala, J.J. 1974. Experimental historical phonology. In Historical Linguistics, II: Theory and De-
scription in Phonology. Proceedings of the 1st International Conference on Historical Linguis-
tics. Edinburgh, 27 Sept. 1973, J.M. Anderson & C. Jones (eds), 353389. Amsterdam:
North Holland.
Ritchie, W. 1968. On the explanation of phonic interference. Language Learning 18:183197.
Scobbie, J.M. 1998. Interactions between the acquisition of phonetics and phonology. In Papers
from the 34th annual regional meeting of the Chicago Linguistic Society, Vol. II, M.C. Gruber,
D. Higgins, K. Olson & T. Wysocki (eds), 343358. Chicago IL: CLS.
Scobbie, J.E., Gibbon, F., Hardcastle, W.J. & Fletcher, P. 2000. Covert contrast as a stage in the
acquisition of phonetics and phonology. In Papers in Laboratory Phonology, V: Language
Acquisition and the Lexicon, M. Broe & J. Pierrehumbert (eds), 194203. Cambridge: CUP.
Smith, B.L. 1979. A phonetic analysis of consonant devoicing in childrens speech. Journal of
Child Language 6: 1928.
Smith, C.L. 1997. The devoicing of /z/ in American English: Effects of local and prosodic con-
text. Journal of Phonetics 25: 471500.
Stevens, K.N., Blumstein, S.E., Glicksman, L., Burton, M. & Kurowski, K. 1992. Acoustic and
perceptual characteristics of voicing in fricatives and fricative clusters. Journal of the Acous-
tical Society of America 91: 29793000.
Appendix
Target words
1. sink 26. grizzly

2. see 27. asthma
3. sick 28. plasma
4. seat 29. business
5. scissors 30. paisley
6. cessna 31. bruise
7. muesli 32. rose
8. Christmas 33. wise
9. Presley 34. news
10. isthmus 35. nose
11. anxious 36. noise
12. serious 37. haze
13. close 38. breeze
14. famous 39. prize
15. base 40. glaze
16. loose 41. seriousness
17. face 42. houseless
18. price 43. anxiousness
19. horse 44. baseless
20. house 45. looseness
21. zebra 46. faceless
22. zero 47. horseless
23. zee 48. closeness
24. zoo 49. famousness
25. zip 50. priceless
51. newsless 56. bruiseless
52. breezeless 57. prizeless
53. hazeless 58. wiseness
54. wiseness 59. noseless
55. roseless 60. glazeless
section 2
Sources of individual differences

in phonological acquisition
Because Dinnsens work often focused on children with phonological disorders, he

has also been interested in the differences among childrens learning patterns. Two
children with very similar phonologies may show very different developmental
trajectories (Dinnsen, Gierut & Farris-Trimble, 2010), and Dinnsen has examined
the extent to which phonological theories can account for these individual differ-
ences. A great deal of his research has been devoted to proving that even the most
unusual developmental trajectories are governed by phonological principles
(Dinnsen, 1984; Dinnsen, Barlow & Morrisette, 1997; Dinnsen, Chin, Elbert &
Powell, 1990; Dinnsen & McGarrity, 2004; Dinnsen, OConnor & Gierut, 2001).
He has demonstrated the application of these principles to changing phonologies,
and he frequently collaborates with speech-language pathologists to apply phono-
logical theory to treatment approaches (Dinnsen, Chin, & Elbert, 1992; Dinnsen,
Chin, Elbert & Powell, 1990; Dinnsen & Gierut, 2008; Elbert, Dinnsen & Powell,
1984; Forrest, Elbert & Dinnsen, 2000; Gierut, Elbert & Dinnsen, 1987; Morrisette,
Dinnsen & Gierut, 2003; Powell, Elbert & Dinnsen, 1991). This work has led to
some important discoveries that have revolutionized treatment programs for chil-
dren with phonological disorders. However, Dinnsen also has shown that a childs
development is influenced by more than just phonological factors.
Section 2 explores external influences on individual differences in phonologi-
cal acquisition patterns. Each paper in this section focuses on a different potential
predictor of phonological learning, illustrating the broader connections between
phonology and other disciplines. Taken together, these three chapters remind the
reader of the many factors (both phonological and otherwise) that contribute to a
childs patterns of acquisition.
Ingram and Dubasik delve into the genetic and environmental influences on
language development in their description of the phonologies of siblings (Sibling
rivalry: Comparing phonological similarity between twin and non-twin siblings).
They find evidence for both genetic and environmental effects, illustrating how
non-phonological factors that are outside of the experimenters control can play a
role in phonological development.
Gierut, Morrisette and Younger examine treatment data to pinpoint how and
when children with phonological disorders generalize what they are learning
(Abstracting phonological generalizations: Evidence from children with disor-
ders). They show how aspects of the treatment process itself (specifically, the
number of treatment trials) correspond to outcomes. That is, apart from phono-
logical considerations about which sounds were treated, simply the number of
trials provided in treatment impacts when children generalize different types of
sounds. In this case, aspects of the treatment protocol that may not be central to
the phonological processes involved can still have a strong impact on learning.
Pisoni explores the relationship between cognitive factors and the develop-
ment of speech perception in children with cochlear implants (Rapid phonologi-
cal coding and working memory dynamics in children with cochlear implants:
Cognitive foundations of spoken language processing). Pisonis summary of the
cognitive characteristics that may predict later outcomes is a reminder that pho-
nology, or even language as a whole, does not reside in a bubble it is heavily in-
fluenced by what else is going on in the brain.
In sum, these three papers situate the study of phonological development
within a larger context and suggest factors with which to predict the outcomes of
that development. As Pisoni notes, Dinnsen has been a pioneer in the application
of phonological theory to clinical populations, and the chapters in this section
reflect the expansion of this pioneering work outside of the bounds of pure pho-
nological theory.
References
Dinnsen, D.A. 1984. Methods and empirical issues in analyzing functional misarticulation. In
M. Elbert, D.A. Dinnsen & G. Weismer (Eds.), Phonological Theory and the Misarticulating
Child (pp. 517). Rockville, MD: ASHA.
Dinnsen, D.A., Barlow, J.A. & Morrisette, M.L. 1997. Long-distance place assimilation with an
interacting error pattern in phonological acquisition. Clinical Linguistics & Phonetics 11:
319338.
Dinnsen, D.A., Chin, S.B., & Elbert, M. 1992. On the lawfulness of change in phonetic invento-
ries. Lingua 86: 207222.
Dinnsen, D.A., Chin, S.B., Elbert, M. & Powell, T. 1990. Some constraints on functionally disor-
dered phonologies: Phonetic inventories and phonotactics. Journal of Speech and Hearing
Research 33: 2837.
Dinnsen, D.A., & Gierut, J.A. 2008. Optimality Theory, Phonological Acquisition, and Disorders.
London: Equinox Publishing Ltd.
Dinnsen, D.A., Gierut, J.A., & Farris-Trimble, A.W. 2010. Comparative markedness and induced
opacity. Language Research, 46: 138.
Section 2. Sources of individual differences in phonological acquisition
Dinnsen, D.A., & McGarrity, L.W. 2004. On the nature of alternations in phonological acquisi-
tion. Studies in Phonetics, Phonology and Morphology 10: 2341.
Dinnsen, D.A., OConnor, K.M., & Gierut, J.A. 2001. The puzzle-puddle-pickle problem and the
Duke-of-York gambit in acquisition. Journal of Linguistics 37: 503525.
Elbert, M., Dinnsen, D.A. & Powell, T. 1984. On the prediction of phonologic generalization
learning patterns. Journal of Speech and Hearing Disorders 49: 309317.
Forrest, K., Elbert, M., & Dinnsen, D.A. 2000. The effect of substitution patterns on phonologi-
cal treatment outcomes. Clinical Linguistics & Phonetics 14: 519531.
Gierut, J.A., Elbert, M. & Dinnsen, D.A. 1987. A functional analysis of phonological knowledge
and generalization learning in misarticulating children. Journal of Speech and Hearing Re-
search 30: 462479.
Morrisette, M.L., Dinnsen, D.A. & Gierut, J.A. 2003. Markedness and context effects in the ac-
quisition of place features. Canadian Journal of Linguistics 48: 329355.
Powell, T., Elbert, M. & Dinnsen, D.A. 1991. Stimulability as a factor in the phonological gener-
alization of misarticulating preschool children. Journal of Speech and Hearing Research 34:
13181328.
Sibling rivalry
Comparing phonological similarity between twin
and non-twin siblings

Arizona State University and Bowling Green State University
Ingram, Dubasik, Liceras & Fernndez Fuertes (2011) developed a measure

of phonological similarity that compares phonological samples across 4 levels
(whole words, word shapes (syllables), consonant inventories, consonant
correctness), using 9 measures in total. The present study used the similarity
measure to compare 4 dyads of children: identical twins, non-identical twins,
non-twin siblings born 2 years apart, non-twin siblings born 5 years apart.
Results indicated that phonological similarity decreased across the 4 dyads,
providing evidence for genetic and environmental effects. The differences
across dyads varied by phonological level, thus, indicating the importance of
conducting multi-dimensional phonological analyses.
Introduction
The notion of similarity underlies all aspects of the study of first language acquisi-
tion. Take, for instance, the efforts to construct theories of language acquisition.
Those who believe in a genetically restricted, strongly constrained language acqui-
sition device, focus on the universality of language acquisition. All children have
the same native endowment, and will acquire specific languages in highly similar
ways. Proposals of the language acquisition device having parameters allow for
some variation (Lightfoot 1991), but the extent to which variation is allowed has
not been resolved. Those taking a more traditional child language perspective
(Ingram 1989) focus more on differences in acquisition, with children capable of
following many different routes in the acquisition of individual languages.
Issues of similarity are at the core of the study of language disorders. Determin-
ing that a child is acquiring language in a typical manner requires conclusions
about similarity in relation to some form of normative information. In speech
assessment, for instance, rates of consonant correctness are often used to determine
similarity (Shriberg 1982). Another approach is to determine whether a childs con-
sonantal substitutions can be traced back to a problem with underlying representa-
tions or a problem with mapping from a correct representation (Weismer, Dinnsen
& Elbert 1981). Efforts to determine similarity between children include the study
reported in Dinnsen, Chin, Elbert and Powell (1990), which established five levels
of consonantal inventories for children with moderate to severe speech disorders.
Another line of research that focuses on differences has been conducted by Dodd
and colleagues to establish differences between children that reflect a phonological
disorder rather than a delay. In cases of delay, the claim is that there is similarity
between older children and younger children with the primary difference being age
(Dodd & McEvoy 1994; Hua & Dodd 2000; McEvoy & Dodd 1992).
The present study is a follow-up to a study on the early phonological develop-
ment of two-year-old bilingual identical twins (Ingram, Dubasik, Liceras &
Fernndez Fuertes 2011). In the earlier study, Ingram and colleagues analyzed
word productions of the twins to determine similarities between the phonological
systems of the twins, as well as similarities within each child with regard to their
English and Spanish. To make these comparisons, a measure of phonological sim-
ilarity was developed that consisted of 9 measures of phonological acquisition
across 4 levels of analysis, that is, whole words, word shapes (syllables), consonant
inventories, and consonant correctness. The measure yielded results indicating
that the twins had very similar phonological systems for both English and Spanish,
but that the English and Spanish for each child were less so. More specifically, Leos
and Simons English were 88% similar, and their Spanish was 94% similar. On the
other hand, Leos English and Spanish were 73% similar, while Simons English and
Spanish were 68% similar. It was concluded that early language separation was tak-
ing place despite a presumed genetic tendency for the children to be highly similar
to each other. Since data from only a single set of twins were analyzed, it was not
possible to conclude whether or not the identical twins were more similar than
non-identical twins, or non-twin siblings. Previous studies have shown that speech
patterns of identical twins are more similar than speech patterns of non-identical
twins (Lewis & Thompson 1992). Other research, however, has focused on how the
speech of twins can be quite different (Clements & Fee 1994; Edwards & Bernhardt
1973; Leonard, Newhoff & Mesalam 1980). The present study was designed to ex-
pand the results from the original study of the bilingual twins by comparing their
English phonological patterns to other children, using the measure of similarity
mentioned above. Data from three dyads of additional children were analyzed.
One dyad was a set of non-identical twins, for whom data were available on the
Childes database (MacWhinney & Snow 1985). Two dyads were non-twin siblings
who differed in the time period between their births. The siblings in one dyad were
Sibling rivalry
2 years apart, while the siblings in the other dyad were 5 years apart. The research
questions were as follows: (1) are the early phonologies of twins more similar than
those of non-twin siblings? (2) are the phonologies of non-twin siblings closer in
age more similar than those further apart in age? and (3) do specific aspects of
phonological acquisition show more similarities than others?
It was hypothesized that the answers to the first two questions would be posi-
tive. No strong prediction was made regarding question 3, though it was antici-
pated that the ways in which the dyads were similar would vary. A further purpose
of the study was to determine the efficacy of the measure of phonological similar-
ity beyond that of the original study.
Method
Participants
The participants in this study consisted of the following four pairs of siblings: iden-
tical twin boys, Leo and Simon, sampled at ages 18 to 20 months; non-identical
twin girls, Jane and Lucy, sampled at ages 17 to 19 months; non-twin siblings born
two years apart, Rachel and Samuel, sampled at age 22 months; non-twin siblings
born 5 years apart, Jennika and Daniel, sampled at 18 and 22 months, respectively.
Leo and Simons samples were collected by Liceras and Fernndez Fuertes as
part of a longitudinal study of bilingual (Spanish & English) language acquisition. A
description of their early phonological acquisition is found in Ingram et al., (2011).
Jane and Lucy were studied by Alan Cruttenden (c.f. Cruttenden 1978) who has
contributed their samples to the Childes Database (MacWhinney & Snow 1985).
Rachel and Samuel are children of the first author whose early phonological devel-
opment is reported in Ingram and Ingram (2011). Jennika and Daniel are children
of the first author and half sister and brother respectively of Rachel and Samuel.
Jennikas early phonological acquisition is reported in Goad and Ingram (1987).
Children in all four dyads were at a similar stage of word acquisition. Samples
were collected approximately during acquisition of the first 100 words. The sample
sizes for each pair of siblings was as follows, Leo and Simon 73 and 67 words, Jane
& Lucy, 63 and 72 words, Rachel and Samuel, 71 and 108 words, Jennika and
Daniel, 72 and 77 words.
Phonological samples
The words used for Leo and Simon were taken from the study by Ingram et al.,
2011, and details on the sampling and transcription are discussed therein. The
children were audio recorded during regular language sampling, and the recordings
were subsequently digitized. The childrens words were transcribed from selected
recordings by two independent transcribers, with differences resolved by a third
transcriber. The words for Jane and Lucy were taken from phonetically transcribed
language samples provided to Childes by Alan Cruttenden, a professional phone-
tician. The Childes website states the transcriptions were done at the time of the
childrens productions and no recordings were made. The words for Rachel and
Samuel were collected by their parents in the form of a parental diary. Only words
were entered into the diary when the parents agreed on the phonetic transcription.
The words for Jennika were from a diary study conducted by her father. The words
from Daniel were from recordings made during regular, longitudinal language
sampling, which were subsequently transcribed by his father.
Phonological analyses
Phonological analyses were conducted on the sample from each child, using the
general approach of the Basic Analysis, presented in Ingram and Ingram (2001),
and adapted in Dubasik and Ingram (2011), Ingram et al. (2011) and Ingram and
Dubasik (2011). The fundamental goal of the Basic Analysis is to provide quantita-
tive, multidimensional analyses of a childs phonological abilities from the word to
the segmental level. Four specific levels are targeted: whole words, word shapes
(syllables), consonant inventories, and consonant correctness. Nine quantitative
scores were determined across the four levels.
Whole Word Measures. At the word level, three- measures of whole word com-
plexity were used. The phonological mean length of utterance (pMLU Targets) is a
measure of the overall complexity of the target words. For each target word, a
point was given to each segment and an additional point to each consonant (e.g.
cat would receive 5 points, while elephant would receive 11 points). The scores
for each word were then averaged across all words. The phonological mean length
of utterance (pMLU Child) is a measure of the overall complexity of the childs
productions. Each segment received 1 point, and each correct consonant received
1 additional point (e.g. cat [ta] would receive 2 points, while [ka] would receive 3
points). The scores for each word were then averaged across all words. Lastly, a
measure of whole word proximity (PWP) was calculated by dividing the pMLU
Child by the pMLU Target (e.g. cat [ta], would be 2/5 or 40%). High PWP scores
(e.g. over 70%) indicate close proximities and indirectly high intelligibility, while
the converse is true for lower proximities (e.g. below 50%).
Word Shape (Syllable) Measures. At the syllable level, two measures were used
to determine the childs individual preferred syllables and overall syllabicity, based
on the childs productions. The proportion of monosyllables is the proportion of
monosyllabic words produced out of the total number of words produced. Previous
Sibling rivalry
research has shown that children often show preferences for either monosyllabic or
multisyllabic words (Ingram 1981). A measure of preferred word shapes was also
used. Previous research such as Stoel-Gammon (1991) has found that the following
word shapes constitute the majority of the childs early word productions: CV, CVC
(in monosyllabic words), and CVCV, CVCVC, VCV (in multisyllabic words). The
number of words that comprised these shapes was determined, and each one that
occurred in at least 10% of the total words was considered a preferred word shape.
Consonant Inventories Measures. The establishment of consonant inventories
provides information regarding the consonants found in the childs productions,
independent of the target phonemes. In the present study, the focus was on the
consonants used by each child separately in word onsets and word codas. A fre-
quency criterion was used (Ingram 1981) to identify two categories of consonants;
(1) Used (in 4 or more words), and (2) Marginal (in 1 to 3 words). Articulation
Scores, AS (Ingram 1981) were derived separately for codas and onsets, by which
each used consonant received 2 points, and each marginal consonant received 1
point. The scores for each word position were summed for a score, resulting in two
scores: CI AS Onsets, and CI AS Codas. For example, the following system of co-
das, [n, (k), (f), s], where parentheses enclose marginal consonants, would receive
a CI AS Coda score of 6.
Measures of Consonant Correctness. The analysis of consonant correctness
identifies how the childs consonants relate to the phonemes in the target words
being expressed. It distinguishes between phonemes used correctly (matches) and
those produced as substituted sounds (substitutions). A match was scored as a
phoneme attempted at least 3 times and produced correctly in over 50% of words
attempted, and was determined separately for word onset consonants and word
coda consonants. The calculation was as follows: matches received 2 points each;
and marginal matches receive 1 point each. For example, a system of matches for
onset consonants as the following (where parentheses indicate a marginal match),
/m, n, b, d, g, f, s, (w)/ would receive a score of 15. Two scores were thus calculated:
CC AS Onsets, and CC AS Codas.
Phonological similarity
Once the Basic Analyses of the childrens productions were completed, a measure-
ment of phonological similarity was determined for the children in each of the
four dyads. For each measure, the scores of each child within the dyad were com-
pared against a predetermined criterion to determine a percentage of similarity
ranging from 100% to 0% (c.f. Ingram et al. 2011). The criteria were determined
through a process of trial and error, since it was undesirable to have criteria that
either showed few differences or few similarities between measures. The goal was
to determine similarity that was neither too strict nor two rigid. For instance, sup-
pose one wanted to compare two children on the basis of their mean length of ut-
terance (MLU), based on Brown (1973). Suppose further that their MLUs were 2.0
and 2.4 respectively. One option would be to assume that they are 100% similar if
they are in the same MLU stage, which in this case would be Browns Stage 2 (2.0
to 2.49). Also, one could deduct 10% for each stage difference, so that a child with
an MLU in Browns Stage 3 (say 3.5) would be 90% similar to these two children.
This would be criteria that are weighted heavily on the side of similarity. The op-
posite would result if the criteria were that 100% similarity required the same
MLU, and a difference of .1 reduced similarity by 10%. The two children would
then be only 60% similar to each other, and 0% similar to the Stage 3 child. This
would be weighted too heavily on the side of difference. The criteria used in the
present study were ones that yielded differences between these two extremes.
The following criterion was used for the pMLU Target measure. The childrens
scores were considered the same (100% similar) if within .3. Differences greater
than that would lose 10%, for each .3 difference. For example, if one childs pMLU
Target score was 6.5, then the other childs score would be considered 100% similar
if it were between 6.2 and 6.8, 90% if it were between 6.9 and 7.2, or 6.1 and 5.8,
etc. The following is a list of the criteria to obtain 100% similarity on each measure,
and the scale used to deduct 10% increments: pMLU Targets and pMLU Child:
100% similarity if scores are within .3, with 10% deducted for each greater dif-
ference; Proximities are 100% same if within .03 of each other, with 10% deducted
for each greater .03 difference; proportions of monosyllables are 100% the same if
within .05 of each other, with 10% deducted for greater each difference; pre-
ferred syllables are 100% the same if they are the same number, with 10% deducted
for each additional difference in number; Articulation Scores for consonant inven-
tories, determined separately for word onsets and word codas, are 100% the same
if they are the same score, with 10% deducted for each additional numerical differ-
ence; this is done similarly for AS scores for correct consonants.
A comparison of the dyads on the 9 measures constituted the primary analysis
of similarity. The final percentages of similarity were also averaged for each of the
4 levels of the phonological analysis to provide an idea of similarity within these,
along with an overall similarity across the levels.
Results
Leo and Simon
The similarity measure results for the identical twins Leo and Simon indicated that
their phonological systems were highly similar, with 7 of the 9 measures showing
Sibling rivalry
Table 1. The measurement of similarity for Leo and Simon
Whole Word Measures Consonant Inventories

pMLU Targets 100% Onsets 80%
pMLU Child 100% Codas 90%
Proximity 90% Mean 85%
Mean 97%
Word Shapes Consonant Correctness

Monosyllables 100% Onsets 90%
Preferred 100% Codas 50% (Simon)
Mean 100% Mean 70%
Overall Similarity = 88%
similarity between 90% and 100% (Table 1, Appendix A). The primary difference
between the two boys involved the production of correct consonants in coda posi-
tion. Simon showed four correct coda consonants, /n, t, k, f/, while Leo had ac-
quired only one, /n/.
Jane & Lucy
Results for the non-identical twin girls, Jane and Lucy, indicated highly similar
phonology, though less so than the identical twin boys (Table 2, Appendix B). Leos
and Simons scores on 7 of the measures were at least 90% similar, while Jane and
Lucys scores were similar on 5 measures. They were most similar with regard to
word shapes, where both girls preferred monosyllabic words (.72 and .70 respec-
tively). They were also relatively similar in terms of correct consonants in onset
position, where Jane had acquired 5 consonants and Lucy 6. In addition to the
Table 2. The measurement of similarity for Jane and Lucy

pMLU Targets 80% Onsets 60% (Lucy)
pMLU Child 100% Codas 90%
Proximity 50% (Lucy) Mean 75%
Mean 73%

Monosyllables 100% Onsets 80%
Preferred 100% Codas 100%
Mean 100% Mean 90%
5 consonants acquired by Jane, Lucy also produced /n/. Turning to differences,

Lucy showed more advanced development in her inventory of onset conso-
nants. Not only did she produce the consonants used by Jane, but also [k], [s],
and [l]. Lucy also had a higher PWP score, at .66 compared to Lucys .54. These
findings suggest that Lucy was more advanced than Jane at this early stage of
acquisition.
Rachel & Samuel
Rachel and Samuel were less similar than children in the previous two dyads, as
evidenced by a lower number of measures (4) that were 90% or higher. Rachel
and Samuel had similar scores on 4 measures as opposed to Leo and Simon who
were similar on 7, and Lucy and Jane on 5 (Table 3, Appendix C). Rachel and
Samuel were both similar in terms of their segmental analyses, in the sense that
their consonant inventories were the most similar level, followed by their number
of correct consonants. The primary difference was that Samuel was able to cor-
rectly produce the phoneme /g/, while this phoneme was lacking in Rachels pho-
nological system. While Samuel was more advanced in this regard, Rachel was
more advanced in whole word productions. Her targets were more complex than
those of Samuel on an average of one additional phoneme per word. Rachels
productions also were more complex, containing nearly an additional segment
per word. This difference is reflected by scores on word shape measures. Samuel
was highly monosyllabic, with a rate of nearly three quarters of his words being
monosyllables, while Rachel showed the opposite preference, with over half of
her words being multisyllabic.
Table 3. The measurement of similarity for Rachel and Samuel

pMLU Targets 60% (Rachel) Onsets 80%
pMLU Child 70% (Rachel) Codas 90%
Mean 73%

Monosyllables 50% (Samuel) Onsets 80% (Samuel)
Mean 75% Mean 90%
Sibling rivalry
Jennika & Daniel
The analysis for Jennika and Daniel (Table 4, Appendix D) showed 5 measures at
90% or above, suggesting that they were more similar than Rachel and Samuel whose
scores on 4 measures were similar. This was not the case, however, when calculating
overall similarity. The overall similarity for Rachel and Samuel was 81% versus 68%
for Jennika and Daniel. Conversely, further examination revealed that Rachel and
Samuel were similar on segmental measures (i.e. consonant inventories and conso-
nant correctness), and differed primarily on their whole word and word shape
characteristics. Jennika and Daniel showed the opposite pattern, by having high pro-
portions of similarity on whole word and word shape measures, but differed in their
consonant inventories and consonant correctness. Daniel was much more advanced
in his acquisition of onset consonants, both in the analyses of consonant inventories
and correct consonants. In the latter, Daniel had acquired 12 onset consonants com-
pared with Jennika who had acquired 6. Daniels more advanced system was the re-
sult of his acquiring the voiceless stop consonants /p, t, k/ and two fricatives /s, /.
Their codas consonants were more similar, with Jennika having acquired one more
coda consonant than Daniel. A noticeable feature of their differences is that Daniel
showed a strong preference for acquiring onset consonants, while Jennikas acquisi-
tion of onset and coda consonants was more balanced.
Similarity across dyads
Overall the analyses of similarity across dyads revealed that the dyads differed from
one another in different ways across the 4 levels examined (Table 5). The measure for
preferred syllables is not included here because it did not yield any differences be-
tween the children. The focus of the analysis was to examine the measures that
Table 4. The measurement of similarity for Jennika and Daniel

pMLU Targets 1000% Onsets 0% (Daniel)
pMLU Child 90% Codas 60% (Jennika)
Mean 90%

Monosyllables 100% Onsets 0% (Daniel)
Mean 100% Mean 50%
Table 5. A comparison of the 4 dyads across 8 measures, with measures below 80% in bold
Summary Leo Simon Jane Lucy Rachel Samuel Jennika Daniel

Measures
pMLU Targets 100% 80% 60% 100%

pMLU Child 100% 90% 70% 90%
Proximity 90% 80% 50% 90%
Monosyllables 100% 100% 50% 100%
CI AS: Onsets 80% 60% 80% 0%
CI AS: Codas 90% 90% 90% 60%
CC AS Onsets 90% 80% 80% 0%
CC AS Codas 50% 100% 100% 100%
Overall 88% 85% 81% 68%
were less than 80% similar (Table 5, bold print). Every dyad had at least one mea-
sure below this criterion with the maximal number being 3 for the non-twin sib-
ling dyads. Further, each level of the analysis had at least one dyad with children
that differed on that level. Conversely, none of the dyads differed in a similar way
to any of the other dyads. In terms of overall similarity, the dyads ranged from
most to least similar from left to right (Table 5). Twins were more similar than
non-twin siblings, productions of the identical twins were more similar than the
non-identical twins, and the siblings closer in age were more similar than the sib-
lings further apart in age.
Lucy compared to all children
Lastly, a comparison of similarity was done between Lucy and the other seven
children (Table 6). This post hoc analysis was suggested by the reviewers to see
what similarities might occur when a child was compared outside of his or her
dyad. Lucy was randomly selected from the subset of children who otherwise had
no genetic relation to any children other than the one in the dyad.
Lucy was found to be more similar to her sister than to six of the other chil-
dren, as measured by the number of measures where similarity was below 80%.
She was, however, more similar to the twin boy Leo, than to her sister. In all seven
comparisons, the areas of similarity (and conversely differences) varied. The spe-
cific differences between Lucy and the other children were as follows: Leo used
more coda consonants; Jane had a lower proximity and fewer onset consonants;
Samuel had a lower proximity and fewer onset and coda consonants; Jennika had
fewer onset consonants, more coda consonants, and more correct coda conso-
nants; Simon had fewer onset consonants, more coda consonants, and more
Sibling rivalry
Table 6. A comparison of Lucy to 6 other children across 8 measures, with measures

below 80% in bold; Difference indicates number of measures below 80% similarity
Summary Leo Jane Jennika Samuel Simon Daniel Rachel

Measures
pMLU Targets 90% 80% 100% 100% 90% 100% 70%

pMLU Child 100% 90% 90% 80% 90% 80% 100%
Proximity 100% 50% 80% 60% 100% 80% 70%
Monosyllables 80% 100% 100% 100% 90% 90% 60%
IA: Onsets 90% 60% 0% 40% 70% 60% 20%
IA: Codas 70% 90% 60% 80% 60% 30% 70%
RA: Onsets 80% 80% 0% 90% 70% 0% 90%
RA: Codas 90% 100% 100% 70% 40% 20% 80%
Difference 1 2 3 3 4 4 5
correct onset and correct coda consonants; Daniel had more onset and coda con-
sonants, and more correct onset and coda consonants; Rachel had a higher pMLU
target score, lower proximity, lower use of monosyllables, and fewer onset and
coda consonants.
Discussion
The findings that identical twins are more similar than non-identical twins support
previous research, though most research examines language measures rather than
phonological measures. Further, the research that has looked at speech has more
often than not pointed out how different twins can be, with one child developing
much more typically than the other. With speech disorders removed, however, the
similarity found in this study is notable. Less has been discussed concerning the
possibility that siblings close in age (within two years in this study) may be more
similar than those farther apart in age. That result is not surprising, however, since
it is reasonable to expect that the environment of siblings closer in age will be more
similar, both in terms of the language addressed to them by their parents and the
topics discussed. The comparison of the speech of half-brothers and half-sisters is
unique. The result of this comparison yielded identifiable obvious similarities.
Perhaps the more important finding was the extent to which the areas of dif-
ference varied across the dyads (and across dyads in the post hoc analysis). The
dyads were highly similar across specific levels, yet varied at each of the four levels
studied. For example, non-identical twins Jane and Lucy, and close in age siblings
Rachel and Samuel were relatively similar in the relational analysis of onsets and
codas, while identical twins Leo and Simon differed in codas, and further in age,
siblings Jennika and Daniel, differed in onsets. An extension of this finding is that
it demonstrates the importance of using a multidimensional approach in assessing
a childs phonological development (Ingram & Dubasik 2011). This point can be
made in relation to the issue of whether or not a child is developing in a typical
fashion or has a speech sound disorder. The results presented here raise the issue
of considering that particular levels of a childs phonological development may be
more typical than others, when compared with some hypothetical typically devel-
oping child. Also, a childs strengths relative to weaknesses may inform speech
sound intervention. One such example would be determining higher vs. lower
whole word proximity scores. A child with a higher proximity score might respond
more favorably to the introduction of new sounds, while a child with a low prox-
imity might be best served building higher proximity with the sounds they already
have. These and other clinical issues are discussed in Ingram and Ingram (2001).
The post hoc analysis found that Lucy was more similar to her sister than to six
of the other seven children. While no predictions were made in this regard in the
original design, it follows from the general conclusions of the study. If genetic and
environmental factors influence similarity, than children who share neither should
overall be expected to be less similar. Why then, was Lucy so similar to Leo? The
answer to that question lies in the complex area of types of phonological learners.
These two children happen to be following highly similar paths of acquisition
across the 4 areas selected for study. The preliminary results of the present study
regarding twins and the role environment do not rule out that children may also
be similar as the result of how they are acquiring their phonological systems.
The results demonstrate the usefulness of a measure of phonological similarity.
In the present study the measure was used for a very specific purpose, that is, to
explore similarity between twins and non-twin siblings. The measure has the poten-
tial, however, to be used in a much broader range of contexts. In Dubasik and In-
gram (2011), the measure was used to compare three dyads of Spanish-speaking
three-year-old children, in which one child was typically developing and the other
exhibited a suspected speech sound disorder. Further, the comparisons were done at
four time points, so that similarity was also compared between different time peri-
ods for each child. This proved to be an effective way to determine which of the four
levels of the Basic Analysis changed for each child over time and which did not.
Since the measure of similarity was only recently developed, more applica-
tions and adaptions are necessary. For one thing, the decisions regarding when
two children are 100% similar on a measure were made by trial and error. Specific
studies may decide to make those comparisons relatively stricter or leaner,
depending on their purpose. It is also possible that it may be preferred to use
the measure as a measure of phonological difference rather than similarity, if
Sibling rivalry
differences are the purpose of the assessment. This adaptation, in fact, was done in
Dubasik and Ingram (2013) where the focus was on how individual child mea-
sures changed over time. From a longitudinal perspective differences represented
gains in phonological development, while similarities did not. Another modifica-
tion to the measure would be to consider further areas to assess, and ones to elim-
inate. In the present study, the measure of preferred syllables did not show any
differences between the children, and was thus dropped from the final compari-
sons. Additional measures for consideration include the possible addition of a
comparison of Percentage of Correct Consonants (PCC) scores (Shriberg 1982),
or comparisons of phonological processes. That said, we conclude that the pre-
liminary usage of a measure of phonological similarity across levels of phonologi-
cal abilities has shown promise so far as an insightful way to assess phonological
development beyond the consideration of segment correctness or phonological
processes solely.
References
Clements, A. & Fee, E.J. 1994. An intra-twin phonological study: Phonologies of a SLI twin and
her normally developing brother. First Language 14: 213231.
Cruttenden, A. 1978. Assimilation in child language and elsewhere. Journal of Child Language 5:
373378.
Dinnsen, D., Chin, S.B., Elbert, M. & Powell, T. 1990. Some constraints on functionally disor-
Research 33: 2837.
Dodd, B. & McEvoy, S. 1994. Twin language or phonological disorder? Journal of Child Lan-
guage 21: 273289.
Dubasik, V.L. & Ingram, D. 2011. Moderately unintelligible Spanish-speaking three-year-olds:
Phonological evolution without intervention. Poster, International Child Phonology Con-
ference, York, England.
Dubasik, V.L. & Ingram, D. 2013. Comparing phonology of dyads of children with typical devel-
opment and protracted development. Clinical Linguistics and Phonetics 27: 705719.
Edwards, M.L. & Bernhardt, B. 1973. Twin speech as the sharing of a phonological system. Ms,
Stanford University.
Goad, H. & Ingram, D. 1987. Individual variation and its relevance to a theory of phonological
acquisition. Journal of Child Language 14: 419432.
Hua, Z. & Dodd, B. 2000. Phonological systems of a set of Putonghua speaking twins. Interna-
tional Journal of Language and Communication Disorders 35: 487506.
Ingram, D. 1981. Procedures for the Phonological Analysis of Childrens Language. Baltimore MD:
University Park Press.
Ingram, D. 1989. First Language Acquisition. Method, Description and Explanation. Cambridge:
CUP.
Ingram, D. & Dubasik, V.L. 2011. Multidimensional assessment of phonological similarity with-
in and between children. Clinical Linguistics and Phonetics 25: 96267.
Ingram, D., Dubasik, V.L., Liceras, J. & Fernndez Fuertes, R. 2011. Early phonological acquisi-
tion in a set of English-Spanish bilingual twins. In Implicit and Explicit Language Learning,
C. Sanz & R. Leow (eds), 195205. Washington DC: Georgetown University Press.
Ingram, D. & Ingram, K. 2001. A whole word approach to phonological intervention. Language,
Speech and Hearing in the Schools 32: 271283.
Ingram, D. & Ingram, K. 2011. Frame/content theory as an account of early phonological acqui-
sition. The Phonetician 103/104: 719.
Leonard, L.B., Newhoff, M. & Mesalam, L. 1980. Individual differences in early child phonology.
Applied Psycholinguistics 1: 730.
Lewis, B. & Thompson, L. 1992. A study of developmental and speech development in twins.
Journal of Speech and Hearing Research 35: 10861095.
Lightfoot, D. 1991. How to Set Parameters: Arguments From Language Change. Cambridge: The
MIT Press.
MacWhinney, B. & Snow, C. 1985. The child language data exchange system. Journal of Child
Language 12: 271295.
McEvoy, S. & Dodd, B. 1992. Communication abilities of 2- to 4-year-old twins. European Jour-
nal of Disorders of Communication 27: 7387.
Shriberg, L.D. 1982. Diagnostic assessment of developmental phonological disorders. In Phono-
logical Intervention, Concepts and Procedures. M. Crary (ed.), 3059. San Diego CA:
College-Hill.
Stoel-Gammon, C. 1991. Normal and disordered phonology in two-year olds. Topics in Lan-
guage Disorders 11: 2132
Weismer, G., Dinnsen, D. & Elbert, M. 1981. A study of the voicing distinction associated with
omitted, final consonants. Journal of Speech and Hearing Disorders 46: 320328.
Sibling rivalry
Appendix A. Basic Analysis for Leo and Simon
Leo Simon
Whole Word Measures

pMLU Target 6.2 6.1
pMLU Child 3.9 4.1
Proximity .63 .67
Word Shapes
Proportion of Monosyllables .59 .63
Preferred Syllables 3 3
(CV, CVC, CVCV)
Consonant Inventories
Onsets m* n m (n)
b* d* g* b* d* g*
f (s) * f* (s) ()
l (j) (w) j
Codas n* n*
(k) (t) k
(f) s* s*
Articulation Score Onsets 18 16
Articulation Score Codas 6 7
Consonant Correctness
Matches in Onsets /m, n, b, d, f, , (l), (j)//m, n, b, d, g, f, s, (w)/
Matches in Codas /n, (f)/ /n, t, k, f/
Note: * = sounds with a high rate of frequency; () = marginal.
Appendix B. Basic Analysis for Jane and Lucy
Jane Lucy
Whole Word Measures

pMLU Target 6.1 5.6
pMLU Child 3.3 3.7
Proximity .54 .66
Word Shapes
(CV, CVC, CVCV)
Onsets m* (n) m n
b* d* g* b* d* g*
p (t) p (t) k
() s
(l) (j) l
Codas n* m (n)
Matches in Onsets /m, b, d, g, l/ /m, n, b, d, g, l/
Matches in Codas /n/ /m/
Note: * = sounds with a high rate of frequency; () = marginal; = a bilabial voiced fricative.
Sibling rivalry
Appendix C. Basic Analysis for Rachel and Samuel
Rachel Samuel
Whole Word Measures

pMLU Target 6.7 5.7
pMLU Child 3.8 3.0
Proximity .56 .52
Word Shapes
(CV, CVC, CVCV)
Onsets m* n* m* n*
b* d* b* d* g*
w w
(l) (h)
Codas (none) (p)

Matches in Onsets /m, n, b, d, w, (l)/ /m, n, b, d, g, w, (h)/
Matches in Codas (none) (none)
Note: * = sounds with a high rate of frequency; ( ) = marginal.
Appendix D. Basic Analysis for Jennika and Daniel
Jennika Daniel
Whole Word Measures

pMLU Target 6.0 5.9
pMLU Child 4.2 4.5
Proximity .70 .75
Word Shapes
(CV, CVC, CVCV)
Onsets m n* m* (n)
b* d* g* b* d* g
p* (t) k*
() (f) s* * h
w (w) (j)
Codas (m) n a (n)

p t* k* (p) t* k*
(s) () (s) (r)
()
Matches in Onsets /m, n, b, d, g, w/ /m, n, b, d, g, p, t, k, s, , h, w, (j)/
Matches in Codas /n, , t, k, r/ /n, p, t, k, /
Note: * = sounds with a high rate of frequency; () = marginal.
Abstracting phonological generalizations
Evidence from children with disorders

Indiana UniversityBloomington
The purpose was to document the trials to induce first generalization in children
with phonological disorders enrolled in treatment. Archival data from 65
preschoolers were examined, with three kinds of generalization documented:
treated, within- and across-class gains in production accuracy. Overall, an
average of 185 trials was sufficient to induce first generalization, but this varied
based on the kind of generalization that occurred. Across-class generalization
required the fewest trials and generalization to the treated sound, the most
trials. Results bear on applied issues associated with lexical diffusion in clinical
treatment and theoretical issues associated with error-driven learning algorithms
and abstraction of phonological generalizations from the input.
Introduction
A metric that is conventionally used to evaluate treatment efficacy for children

with phonological disorders is generalization. Generalization is defined as the
transfer of learning, with an eye toward inducing broad system-wide gains in a
childs phonology following a minimum of treatment. System-wide generalization
is reflected in improved production accuracy of (1) the treated sound in treated
and untreated contexts, (2) untreated (erred) sounds from the same manner class
as the treated sound and (3) untreated (erred) sounds from different manners than
the treated sound. Respectively, these define treated, within-class and across-class
generalization. These three kinds of generalization are conventional to the clinical
literature (Costello & Onstine 1976; Elbert & McReynolds 1985), but are based on
linguistic (Clements & Hume 1995; Dinnsen 1998) and developmental (Dinnsen,
Chin, Elbert & Powell 1990; Rice & Avery 1993) feature hierarchies, where manner
of articulation supersedes properties of place or voice. On behavioral grounds,
manner is also most often implicated in the errored outputs of children with typi-
cal and disordered phonological development (Forrest & Morrisette 1999; Logan
1992). Hence, manner is central to the evaluation of improved production accu-

racy associated with system-wide generalization.
Applied research has largely focused on documenting the ideal conditions that
promote system-wide generalization by children with phonological disorders.
Studies have considered effects of the treated sound (e.g., Dinnsen & Elbert 1984),
treated context (e.g., Elbert & McReynolds 1975) and treated stimuli (e.g., Gierut &
Morrisette 2012). Related work has evaluated the mode (e.g., Shriberg, Kwiatkowski
& Snyder 1989, 1990) and method of treatment delivery (e.g., Dean, Howell, Waters
& Reid 1995; Weiner 1981). Through investigations of this type, evidence-based
clinical practice has advanced in optimizing phonological generalization. However,
despite significant progress, very little is known about how generalization comes
about even though this remains the hallmark of treatment efficacy. This paper be-
gins to explore the issue by identifying the starting point of generalization, as de-
fined by the number of treatment trials to first transfer, and then determining
whether this point varies by the kind of generalization that occurs: treated, within-
or across-class gain. The aim is to depict a potential trajectory of system-wide
phonological generalization. We begin with a summary of the available, albeit lim-
ited clinical evidence about the course of phonological generalization.
Descriptive characterizations of generalization
Typically, phonological generalization occurs in piecemeal fashion, with transfer

of accurate production taking place on a word-by-word basis (Elbert 1984; Leonard
& Brown 1984). The result is partial generalization, with the target sound being
produced correctly in some words but not others, characteristic of lexical diffusion
(Labov 1981). There have been numerous attempts to identify the factors that mo-
tivate diffusion for insight into how word-by-word transfer occurs. One thought is
that frequency of occurrence might predict which words children produce cor-
rectly (Morrisette 1999, 2004), but this seems to hold only for certain phonological
properties (Phillips 1983, 1984). Another idea is that first acquired words might be
first to improve (Tyler & Edwards 1993), but there is evidence to the contrary
(Gierut & Morrisette 2012). Still another view is that imitation of sounds predicts
first transfer, but here too, there are ambiguities and individual differences (Dean
et al. 1995). As it stands, generalization associated with lexical diffusion remains a
mystery. No single variable or set of variables has the predictive power to differen-
tiate the sounds or words that children will generalize.
There is another path of phonological generalization that occurs less often.
Under this scenario, the transfer of accurate sound production takes place across-
the-board, affecting all relevant lexical items (Chen & Wang 1975). Generalization
is presumed to occur in one fell swoop (Ingram 1989). Only a handful of studies
Phonological generalizations
have reported this sort of clinical success (Bedore, Leonard & Gandour 1994;
Compton 1970; Dean et al. 1995; Weiner 1981), which may be due in part to cer-
tain practical challenges associated with sampling. It is not feasible to sample all
relevant lexical items for generalization. Also, longitudinal samples in general are
inherently flawed (Locke 1983) because it is impossible to pinpoint an exact mo-
ment of change. While productions may have appeared to change in one fell swoop,
it is possible that change was occurring all along, but went unobserved between
samples. Further, the reasons why some children show across-the-board change
while others evidence diffusion are not yet known.
Empirical characterizations of generalization
From a complementary quantitative perspective, Elbert, Powell and Swartzlander

(1991) documented the number of exemplars needed to induce phonological gen-
eralization. Retrospective data from 19 preschoolers with phonological disorders
were analyzed. Each child had been taught three sounds in the initial position of
words. Treatment was administered using three sets of exemplars, and continued
until a child reached 90% accuracy of production of the treated sounds in treated
words. Then, a generalization probe was administered, which sampled the treated
sounds in the treated context of untreated words. If a child failed to generalize to
the untreated probe items with 50% accuracy of production, treatment resumed,
but the number of exemplars was increased. Because of additions to the size of the
treated stimulus set, Elbert et al. (1991) were able to establish the minimum exem-
plars for generalization to treated sounds. Results showed that three exemplars
were sufficient to induce generalization for the majority of children. It is notewor-
thy that this effect has been replicated in other domains, including infant percep-
tual learning of stress (Gerken & Bollt 2008), infant categorization of visual
information (Needham, Ducker & Lockhead 2005; Quinn & Bhatt 2005), and
word learning by preschoolers (Xu & Tenenbaum 2007). Together, it appears that
children are able to abstract and form generalizations following exposure to just
three exemplars. There is a further hint that the number of exemplars to general-
ization may be independent of the learning task. If true, the significance for under-
standing the process of generalization is remarkable, but there are at least two
missing pieces that warrant attention.
One gap is associated with the distinction between type and token. Type refers
to the substantive stimuli that comprise a category or class, whereas token refers to
how often those same stimuli are presented in the input. From the aforementioned
studies, three types may be requisite for generalization, but it is not known what
the minimum number of tokens might be to achieve generalization (Elbert et al.
1991). An evaluation of tokens to generalization would round out the existing data
on stimulus types.
Another gap is associated with breadth of generalization. Elbert et al. (1991)
defined generalization as 50% accuracy of production of the treated sound in the
treated context. This yields a very stringent and narrow view of transfer. The expec-
tation of 50% accuracy of production does not take into account that even the
smallest gain reflects a clinically significant practical improvement for children
with phonological disorders (Bain & Dollaghan 1991). Moreover, examination of
the treated sound to the exclusion of gains in other erred sounds and contexts pre-
cludes insight to system-wide phonological generalization. Further, we do not know
whether the trials to induce treated, within- and across-class generalization vary,
yet the answer to this might lend insight to a possible trajectory of generalization.
The purpose of this paper was to address these gaps in a retrospective descrip-
tive study. The goal was to establish the minimum tokens to induce first general-
ization in children with phonological disorders enrolled in treatment. Tokens were
defined as the number of treatment trials to the first transfer of production accu-
racy. Three kinds of generalization treated, within- and across-class gains were
tracked over time. It was predicted that treated sounds would generalize first, re-
quiring the fewest trials to first generalization. Within-class generalization was
expected next, followed by across-class generalization as accuracy gradually
spreads across a childs phonology. As will be seen, the trajectory of first phono-
logical generalization took a rather different course.
Methods
Data were drawn from the Developmental Phonology Archive of the Learnability
Project. This is a compendium of longitudinal production data from 242 children
with phonological disorders enrolled in experimental studies of treatment. The ar-
chive, participant characteristics, phonological samples, treatment procedures, gen-
eralization measures, reliability and fidelity of data have all been described in detail
elsewhere (Dinnsen & Gierut 2008); relevant points are briefly recapitulated below.
Participants and phonemic inventories
Archival data were consulted to identify children who met five inclusionary crite-
ria: (1) treatment of one erred sound, (2) use of a traditional format of instruction
(Van Riper 1978), (3) use of real words as stimuli in treatment, (4) evidence of
phonological generalization, and (5) enrollment in a study utilizing the multiple
baseline design. Those who were taught more than one sound using minimal pairs
were excluded to eliminate a potential inflation of treatment trials due to presumed

increases in task demands (Elbert et al. 1991). Those who were taught nonwords
were excluded to eliminate a potential reduction in treatment trials due to the ap-
parent benefits of novel stimuli (Gierut, Morrisette & Ziemer 2010). Those who
failed to generalize did not inform the question of interest. The multiple baseline
is the most conventional experimental design of phonological treatment studies
(Gierut 2008a), thereby lending generality to the anticipated findings. In all, archi-
val data from 65 of 242 children met the inclusionary criteria. The mean age
of children included in the study was 4 years; 6 months (SD=9 months; range:
3;06;1). The group was comprised of 35 boys and 30 girls.
All 65 children performed within typical limits on batteries of developmental
skills; however, all were identified as having a phonological disorder based on the
results of standardized testing and structured probe data (Gierut 2008b). With
respect to the phonology, the mean number of consonants excluded from chil-
drens phonemic inventories was 9 (SD=2; range: 516). Velar stops, fricatives, af-
fricates and liquids encompassed the full range of inventory exclusions. These gaps
were established by conventional linguistic analyses (Dinnsen 1984), such that
consonants excluded from the inventory were produced with 0% accuracy across
contexts with an absence of corresponding minimal pairs.
Treatment stimuli and protocol
Each child was taught one sound excluded from the inventory. The sound chosen
for treatment was based on prior experimental assignments. Of 65 children, 6 were
taught a velar stop, 30 a fricative, 2 an affricate and 27 a liquid. The treated sound
was taught in initial position of real words that were picturable and familiar to
children (Walley & Metsala 1992). Stimulus words used in treatment were unique
to a childs experimental assignment, but these items were held constant for the
duration of training. The mean number of treated words for a given child was 7
(SD = 1, range: 610).
Treatment followed an established experimental protocol (Gierut 2008a) that
closely resembled that seen in applied settings. As noted, treatment was imple-
mented in the context of a multiple baseline experimental design, which staggers
childrens enrollment to control for maturation and to provide systematic and di-
rect replications of learning (McReynolds & Kearns 1983). There were two se-
quential phases of instruction: imitation of the treated sound in treated words
following a model and spontaneous production of the treated sounds in treated
words. Instruction was provided 3 times weekly in 1-hour individualized sessions.
The number of trials in a given session ranged from 40150 as dictated by the
experimental protocol and design. A child advanced through the phases of
treatment based on preset criteria. Specifically, imitation continued for a total of

seven 1-hr sessions or until the child reached 75% accuracy of production over
2 consecutive sessions, whichever occurred first. Spontaneous training continued
for a total of nine 1-hr sessions or until 90% accuracy of production was achieved
over 3 consecutive sessions, whichever occurred first. Throughout treatment, a
child was provided 1:1 feedback about production accuracy, and accordingly, each
response was scored as correct or incorrect. From these data, it was possible to
tally the number of treatment trials session-by-session as a dependent measure.
The total number of trials entered into analysis was 58,812.
Measure of phonological generalization
Generalization was measured using a structured probe that sampled all target
English sounds across relevant contexts and multiple exemplars in a spontaneous
naming task (Gierut 2008b: 4351). Probes were reserved exclusively as the test for
generalization, and feedback about production accuracy was never provided.
Probes were administered longitudinally to track gains in production accuracy,
with data collected at baseline and continuing on a variable ratio schedule of 3 ses-
sions to completion of treatment. The mean number of probes administered to a
given child was 9 (SD = 3, range: 313). Crucially, the longitudinal administration
of probes aligned with the longitudinal delivery of treatment. Thus, it was possible
to determine how many cumulative treatment trials had been provided to each
probe point.
Probe data were audio-recorded and phonetically transcribed by a listener
blind to childrens experimental assignments; interjudge transcription reliability
was established at 92% agreement (Gierut 2008b: 53). Transcriptions were used to
compute percentages of accurate production of all sounds excluded from a childs
phonemic inventory. Any gain in accuracy of production over baseline was taken
as evidence of first generalization; this was the second dependent measure. The
total number of probe productions entered into analysis was 40,714 words.
Data analysis
Data from each of 65 children were examined individually. All consonants ex-
cluded from a childs phonemic inventory were sorted into three categories reflect-
ing treated, within- and across-class generalization. To illustrate, Child 235
excluded /f v s z t d l/ from the pretreatment phonemic inventory and was
taught /s/ in treatment. In the sort for Child 235, /s/ comprised the category of
treated sound generalization; /f v z / comprised the category of within-class
generalization as these were untreated fricatives; /t d l/ comprised the category

of across-class generalization as untreated affricates and liquid.
After the sort, production accuracy of each sound in a category was computed
using a childs longitudinal probe data. The number of correct responses was tal-
lied and divided by the total responses relative to baseline for each sound and each
probe sample. Using accuracy data, the first points of generalization were then
identified. First generalization was operationally defined as the first observed in-
crease in percent accuracy over baseline, without regard for the amount of per-
centage gain. This definition is consistent with our purpose in establishing the
minimum trials to any first observed accurate production. This stands apart from
Elbert et al.s (1991) stringent requirement of 50% accuracy of production as in-
dicative of generalization. It also stands apart from criteria used to establish the
amount of generalization at post-treatment for purposes of comparing the efficacy
of one intervention to another (Elbert, Dinnsen & Powell 1984; Gierut & Morrisette
2011). The concern herein was not to guarantee that all children achieved the same
level of accuracy, to measure how much generalization took place or to compare
interventions. Rather, the interest was in documenting the point of first observed
generalization in the context of the delivery of treatment. The aim was thus similar
to that of longitudinal studies of typical phonological development (e.g., Ferguson
& Farwell 1975; Smith 1973; Stoel-Gammon & Cooper 1984), where single words
are examined over time for the first apparent change in production accuracy. The
point of first generalization was determined separately for treated, within- and
across-class gains in production accuracy.
After the first points of generalization were established, a childs treatment tri-
als were overlaid on the data. Cumulative trials were counted to first generalization
in the treated, within-class and across-class categories. Thus, for each child, (1)
production accuracy of treated and untreated sounds excluded from the inventory
was computed for each probe sample, (2) three points of first generalization were
identified corresponding to treated, within- and across-class gains, and (3) total
treatment trials to each of the three points of first generalization were tallied.
The collective data were then aggregated for the 65 children and group means
were computed. Mean treatment trials and mean percent accuracies were gener-
ated for the three points of first generalization. From this, it was possible to estab-
lish how many treatment trials were required before each kind of first phonologi-
cal generalization was observed for the group.
Reliability of coding
Reliability was established for the tally of treatment trials and computation of ac-
curacies. For each of 65 children, data from two sessions were randomly selected
for reliability purposes, yielding 130 samples. An independent judge recounted

the number of treatment trials administered in each of those 130 samples, and
compared that count against the originals. Another independent judge recomput-
ed percent accuracies of probe productions for each of the 130 samples, and these
were compared against the originals. There was 99.96% agreement in the tally of
trials and 99.72% agreement in calculations of accuracy.
Results
Results are organized to establish trials and accuracy at first generalization, indepen-
dent of the type of generalization observed. Trials and accuracy associated with
treated, within- and across-class generalization are then reported to inform a pos-
sible trajectory of phonological gain. The course of generalization taken by individ-
ual children was documented as further instantiation of the proposed trajectory.
Trials and accuracy at first generalization
The aggregated data revealed that children were provided an average of 905 total
trials in treatment (SD = 398, range: 2721,700). First generalization was observed
after 185 mean trials (SD = 169; range: 641,272). At the point of first generaliza-
tion, the mean number of sounds that showed gains in accuracy was 3 (SD = 1,
range: 16), bearing in mind that, on average, 9 sounds were excluded from chil-
drens phonemic inventories. Also, at first generalization, sounds that showed im-
provement were produced with 8% mean accuracy (SD = 7; range: <139). It is
necessary to keep in mind that accuracy associated with first generalization oc-
curred after roughly 12 hours of treatment, and did not reflect upper levels of
performance documented at post-treatment and beyond. There was no significant
correlation between trials to first generalization and production accuracy, r(63) =
-.18, p = .15. It is of mention that gains observed at the point of first generalization
held over time for a majority of children. 64 of 65 children sustained first general-
ization accuracy over an average of 5 consecutive probe samples (SD = 3, range:
114). For the remaining child, improvements at first generalization were tran-
sient and likely due to the lack of session-by-session progress during training. The
overall findings thus suggest that first generalization was initiated early in the de-
livery of treatment, positively affected about a third of the sounds produced in
error, and held longitudinally for the majority of children.
Correlational analyses were completed to determine possible contributing
variables to first generalization, with consideration of chronological age, phone-
mic inventory size and number of treatment exemplars. It is possible, for example,
that younger children might have required more trials to first generalization. Per-
haps too the size of a childs phonemic inventory influenced first generalization,
such that those with fewer gaps might have required fewer trials. Likewise, there
may be a relationship between types and tokens, such that children who were ex-
posed to more treatment stimuli might have achieved first generalization in fewer
trials. Results showed no significant correlations between any of these factors and
trials to first generalization r(63) = -.12.03., p >.35. There were also no significant
correlations between these variables and production accuracy r(63) = -.16.17,
p >.18. Thus, trials to and accuracy at first generalization did not appear to be gov-
erned by the child (age), phonological (inventory) or treatment (stimuli) variables
that were examined herein.
Differential first generalization
A more fine-grained analysis of trials and accuracy was completed to determine

whether the point of first generalization varied depending on the kind of phono-
logical generalization that occurred. Specifically, 58 of 65 children evidenced gen-
eralization to the treated sound. The mean number of trials to first generalization
of the treated sound was 358 (SD = 266; range: 641,272) and corresponding ac-
curacy was 26% (SD = 11; range: 650). There was no correlation between number
of treatment trials and accuracy of the treated sound at first generalization, r(56) =
.03, p = .80.
61 of 65 children had the opportunity to generalize to untreated sounds of the
same manner class as the treated. Of these, 46 children showed within-class gen-
eralization. The mean number of trials at first generalization within-class was 276
(SD = 263; range: 641,356) and corresponding accuracy was 14% (SD = 12; range:
150). There was no significant correlation between mean trials and accuracy at
first generalization within-class, r(44) = -.17, p = .27.
63 of 65 children had the opportunity to generalize across-class to untreated
sounds from manners other than the treated. Of these, 60 showed across-class
generalization. The mean number of trials at first generalization across-class was
221 (SD = 184; range: 641,080) and corresponding accuracy was 10% (SD = 10;
range: <150). The correlation between mean trials and accuracy at first general-
ization across-class was again not significant, r(58) = -.06, p = .65.
Trajectory of generalization
Differential trials to first generalization hint of a possible trajectory of phonologi-

cal gain: across-class > within-class > treated sound (see Vihman & Keren-Portnoy
2011 for a similar observation in typical development). This follows from the
aforementioned results where across-class generalization was initiated after 221

mean trials, within-class after 276 mean trials and treated sound gains after 358
mean trials. To evaluate whether these differences to first generalization were sta-
tistically reliable, pair-wise t-tests for dependent samples were conducted. Results
showed a significant difference in trials to first treated versus first across-class gen-
eralization, t(52) = 2.95, p = .005. There was no significant difference in trials to
first treated versus first within-class generalization, t(39) = 1.78, p = .08. Likewise,
trials to first within- versus first across-class generalization were not statistically
significant, t(41) = .99, p = .33. This indicates that childrens first generalization to
the treated sound and across-class were reliably distinct as endpoints of the pro-
posed trajectory. First within-class generalization, on the other hand, overlapped
with these endpoints based on the data at hand.
A complementary way to evaluate the proposed trajectory of generalization is
to examine its instantiation by individual children. The data from 65 children were
considered in documenting which kind of generalization occurred first for a given
child; the results are shown in Figure 1.
A first observation is that a subset of children evidenced one and only one
kind of transfer at first generalization (Figure 1, panel a); 31 of 65 children showed
this pattern. As can be seen, the majority made across-class gains and the minor-
ity, treated sound gains at first generalization. This is consistent with the hypoth-
esis that the first kind of change is likely to be across-class transfer.
A second observation is that some children showed more than one kind of
first generalization concurrently (Figure 1, panel b); 34 of 65 children showed this
pattern. For this subgroup, the breadth of first generalization was broader, but as
before, the majority showed across-class generalization (in combination with some
other kind of gain) at the point of first transfer. This agrees with the hypothesis that
across-class generalization defines the starting point of transfer. However, it also
suggests that the trajectory of generalization may not be sequential, but rather si-
multaneous for some children. From the present data, it was not possible to estab-
lish whether the simultaneous emergence of different kinds of generalization was
real or simply a function of the probe schedule. If probes had been administered
more frequently than every third session, it might have been possible to tease this
apart; however, this would have introduced other new concerns associated with
teaching to the test.
Nonetheless, observed individual differences in breadth of first generalization
are of interest for added reasons. It might be thought, for example, that children ex-
hibiting narrow generalization (Figure 1a) required fewer trials than those with
broader gains (Figure 1b). A t-test comparison revealed no statistically significant
difference in trials relative to breadth of first generalization, t(63) = 1.14, p = .26. The
patterns in Figure 1a resulted from 210 mean trials (SD = 224, range: 641,272),
20
18
16
14
Number of children
12
(a)
10
2
0
Across Within Treated
20
18
16
14
Number of children
12
(b) 10
6
4
2
0
Across+within Across+within Across+treated Within+treated
+treated
Figure 1. Patterns of first generalization as instantiated by 65 children. Panel (a), top,

plots the number who evidenced one kind of first generalization; Panel (b), bottom, plots
the number who evidenced more than one kind of first generalization concurrently.
whereas those in Figure 1b resulted from 162 mean trials (SD = 94, range: 64492).
Thus, treatment trials were not predictive of breadth of first generalization. It
might also be thought that breadth of first generalization was related to produc-
tion accuracy. If first generalization is narrow (Figure 1a), accuracy might be
greater than if it were broad (Figure 1b). A t-test comparison showed this to be the
case, but in the reverse direction, t(63) = -5.44, p < .001. Mean production accu-
racy for children with the narrow pattern of first generalization in Figure 1a was
4% (SD = 3%, range: <112%). By comparison, those who showed broader gain in
Figure 1b had mean production accuracy of 12% (SD = 8%, range: 339%) at first
generalization. Thus, a broader scope of first generalization was associated with
greater accuracy of production; moreover, these benefits came without the need
for additional treatment trials.
Discussion
This research aimed to establish the minimum trials to induce first generalization in
an effort to better understand how system-wide transfer takes place for children with
phonological disorders enrolled in treatment. The main finding was that relatively
few treatment trials are needed to start the process of generalization, but that differ-
ent kinds of generalization require different numbers of trials. This revealed a possi-
ble trajectory of first generalization that was supported by aggregated and individual
childrens data. The results have applied potential to inform the course of generaliza-
tion through lexical diffusion as it unfolds in clinical treatment. There is theoretical
potential for extending linguistic and psycholinguistic theories that make claims
about how children abstract generalizations from the input in language learning.
Applied considerations
It has long been noted that treatment leads to partial generalization, but this study
adds substance to that observation by identifying where in the phonology those gen-
eralizations will take place. Predictably, the first words susceptible to lexical diffusion
will contain sounds from manners other than the treated. We return to the example
of Child 235 to illustrate. Recall that this child was taught fricative /s/, having ex-
cluded /f v s z t d l/ from the inventory. Based on the present results, we might
expect that /t/, /d/ and/or /l/ would be most likely to generalize first because these
sounds are not from the treated fricative category. Indeed, Child 235 showed first
generalization to /d l/ following 144 trials; transfer to treated /s/ occurred later fol-
lowing 204 trials. For the future, prospective longitudinal studies are needed to
validate such predictions given the inherent limitations of retrospective work. Longi-
tudinal studies might also clarify the emergence of within-class generalization, which
remains undifferentiated based on the present findings. Longitudinal studies have a
further advantage in that it may be possible to track generalization to the point of
mastery. This would complement the evaluation of first generalization as herein. It is
possible that different kinds of generalization will accelerate, plateau, decline or in-
teract along the way to a childs achieving adult-like phonological skills.
Results of the present study offer some clinical guidelines for probe adminis-
tration. Recall that treated, within- and across-class generalization were observed
following 358, 276 and 221 mean trials, respectively. Given this, one recommenda-
tion is to initiate probing for generalization early in the delivery of treatment, per-
haps after about 200 trials. This contrasts with the often-conventional practice of
reserving probes for post-treatment evaluations of learning. Another recommen-
dation bears on probe structure. Given our results, early probes might weight the
sampling of untreated sounds by emphasizing manners other than the treated.
This too contrasts with the often-conventional practice of probing only the treated
sound to determine if goals have been met.
At least two other questions can be raised about the current study relative to
the clinical literature. A first relates to our finding that the number of stimuli used
in treatment was not related to trials to first generalization. This suggests that types
(exemplars) and tokens (trials) may have independent effects on phonological
generalization. While intriguing, the available literature offers little insight on the
issue because of an apparent omission. Consider that, in Elbert et al. (1991), chil-
dren were presented with increasing numbers of exemplars in treatment and the
trials afforded in practice increased accordingly. Thus, types and tokens both var-
ied. In our study, a child was exposed to a fixed stimulus set for the duration of
treatment, but trials fluctuated. Here, types were held constant, but tokens varied.
To our knowledge, the final logical possibility tokens held constant and types
varied has not yet been evaluated in phonological treatment (cf. Hamann,
Apoussidou & Boersma 2012, for an adult simulation using an artificial grammar).
It is possible that future research will clarify this relationship between types and
tokens through orthogonal manipulations of the two variables.
A second question relates to our finding that trials and accuracy were not re-
lated. At first glance, this might seem to imply that increased production practice
does not lead to greater generalization accuracy. A more cautious interpretation
keeps in mind that our interest was in first generalization as gleaned from a pro-
duction-based treatment protocol. It is possible that trials and accuracy are linked,
but the relationship may only be revealed well after the completion of treatment. It
is also likely that production practice is not the sole contributor to phonological
generalization. Childrens perceptual (Rvachew 1994) and metalinguistic (Dean
et al. 1995) skills are other probable complements to production practice in treat-
ment. In future research, the interface between trials and accuracy needs to be
explored for other treatment protocols (Williams 2012).
Theoretical implications
These results bear more generally on linguistic and psycholinguistic models of

language acquisition. Current models are built on the assumption that children
pay attention to regularities in the input (Jusczyk 1997 for perception; Stoel-
Gammon 2011 for production). They then use this statistically robust information
to abstract generalizations about linguistic structure. This presumably enables

their acquisition of that structure.
On the linguistic side, the focus has been on speech production and the cor-
respondence between language input and a childs own unique output (e.g., opti-
mality theory; see Bernhardt & Stemberger 1998; Dinnsen & Gierut 2008 for
application to phonological disorders). One assumption is that language learning
is error-driven (Tesar & Smolensky 1998). The more opportunities a child has to
recognize that his/her erred output does not match the target, the more likely the
error will be suppressed. Various algorithms have been put forth to model error-
driven learning, with computational simulations being offered as primary evi-
dence in support of such proposals (Boersma & Hayes 2001; Boersma & Levelt
2000; Farris-Trimble 2009; Jarosz 2010; Pater 2008). It is curious, however, that
behavioral evidence from children directly engaged in phonological learning has
been lacking in these endeavors. The present study begins to contribute toward
this end, with methodological benefits but also challenges.
Specifically, children with phonological disorders enrolled in treatment pro-
vide an ideal testing ground for evaluating assumptions about the role of the input,
error-driven learning and the necessity of evaluating output mismatches. Con-
sider that, in treatment, the amount of input responsible for damping error pat-
terns may be empirically established. Also, in treatment, children have multiple
opportunities for production with corrective feedback to presumably draw out the
necessary comparisons between the target language and their own outputs. The
present findings contribute to the available computational evidence in two key
ways. First, we found that the mean aggregated trials to first generalization were
185. This is considerably less than that reported in computational simulations of
developmental phonological effects (e.g., 100,000 trials in a conventional compu-
tational run). Second, we found no correlation between input trials and improved
accuracy of production. Yet, this is the crux of error-driven learning. While these
differences will need to be reconciled, they underscore the importance of securing
behavioral evidence about input trials to phonological generalization in verifica-
tion and refinement of linguistic models of acquisition. Data from children with
phonological disorders enrolled in treatment may provide one such venue.
On the psycholinguistic side, the emphasis has been on speech perception and
identification of particular input structures that guide childrens language learning
(Gerken, Wilson & Lewis 2005; Newport & Aslin 2004; White, Peperkamp, Kirk &
Morgan 2008). From this vantage, an emerging literature is beginning to explore
the process of generalization (Adriaans & Kager 2009; Gerken 2004; Maye, Weiss
& Aslin 2008). One hypothesis is of particular interest. Specifically, Gerken (2006)
proposed that when a child is faced with linguistic input that has more than one
structural description, the learning task is more complex and computationally
challenging. Yet, this linguistic experience is ultimately most valuable because a

child must learn a broader formal system, i.e., a system that defines a more diverse
range of input structures.
This hypothesis might be extended as a potentially viable account of our find-
ings of individual differences and breadth of first generalization. Recall that a sub-
group of children simultaneously showed more than one kind of first generalization;
those who did also had greater production accuracy. Notably, the pattern of broad
generalization was not at the expense of increased treatment trials for the children of
this study. If Gerken (2006) is correct, then possibly the advantage of broad general-
ization to production accuracy is due to the problem space that children inferred
from the input of treatment. Those who showed broad generalization may have been
attuned to structural resemblances between the treated sound and other untreated
aspects of the phonology. From the outset, these children may have noticed multiple
and/or discrepant phonological patterns relative to their existing knowledge. This
might have contributed to their mapping out a larger problem space. Their corre-
sponding solution to that larger problem might have been more in keeping with
system-wide generalization. Thus, treatment may have cast a floodlight on the whole
phonology to enable broad generalization (cf. spotlighting in typical acquisition,
Peters & Strmqvist 1996). While speculative, it remains to be determined why some
children might define the phonological problem space more broadly than others,
even when faced with the same quality or quantity of input. Continued psycholin-
guistic research on generalization in language learning is likely to offer new explana-
tory accounts that attend to some of these details.
Conclusion
This study offered preliminary evidence about how system-wide generalization

may evolve in the context of treatment for children with phonological disorders.
The findings have applied potential for the administration and structure of clinical
probes, and theoretical relevance to contemporary linguistic and psycholinguistic
models of phonological acquisition. While there remain many unanswered ques-
tions, the research represents an initial effort to better understand the process of
generalization as the hallmark of phonological treatment efficacy.
Authors note
Research reported in this publication was supported by the National Institute on

Deafness and Other Communication Disorders of the National Institutes of Health
under Award Number R01DC001694. The content is solely the responsibility of

the authors and does not necessarily represent the official views of the National
Institutes of Health. We would like to thank Dan Dinnsen for his pioneering in-
sights to the study of phonology, phonological acquisition and disorders. His
impact on the science of the Learnability Project is immeasurable, and his good
humor has kept us in stitches, especially in those terminal cases. Portions of this
paper were presented at the 2009 American Speech-Language-Hearing Associa-
tion Convention, New Orleans. Address for correspondence: Judith A. Gierut,
Department of Speech and Hearing Sciences, 200 South Jordan Avenue, Indiana
University, Bloomington, IN 47405-7002, USA, Email: gierut@indiana.edu
References
Adriaans, F. & Kager, R. 2009. Adding generalization to statistical learning: The induction of
phonotactics from continuous speech. Journal of Memory and Language 62: 311331.
Bain, B.A. & Dollaghan, C.A. 1991. The notion of clinically significant change. Language, Speech
and Hearing Services in Schools 22: 264270.
Bedore, L.M., Leonard, L.B. & Gandour, J. 1994. The substitution of a click for sibilants: A case
study. Clinical Linguistics & Phonetics 8: 283293.
Bernhardt, B.H. & Stemberger, J.P. 1998. Handbook of Phonological Development from the Per-
spective of Constraint-based Non-linear Phonology. San Diego CA: Academic Press.
Boersma, P. & Levelt, C. 2000. Gradual constraint-ranking learning algorithm predicts acquisi-
tion order. Proceedings of the Child Language Research Forum 30: 229237.
Boersma, P. & Hayes, B. 2001. Empirical tests of the gradual learning algorithm. Linguistic In-
quiry 32: 4586.
Chen, M. & Wang, W.S.-Y. 1975. Sound change: Actuation and implementation. Language 51:
255281.
Clements, G.N. & Hume, E.V. 1995. The internal organization of speech sounds. In The Hand-
book of Phonological Theory, J.A. Goldsmith (ed.), 245306. Cambridge MA: Blackwell.
Compton, A. 1970. Generative studies of childrens phonological disorders. Journal of Speech
and Hearing Disorders 35: 315339.
Costello, J. & Onstine, J. 1976. The modification of multiple articulation errors based on distinc-
tive feature theory. Journal of Speech and Hearing Disorders 41: 199215.
Dean, E.C., Howell, J., Waters, D. & Reid, J. 1995. Metaphon: A metalinguistic approach to the
treatment of phonological disorder in children. Clinical Linguistics & Phonetics 9: 119.
Phonological Theory and the Misarticulating Child, M. Elbert, D.A. Dinnsen & G. Weismer
(eds), 517. Rockville MD: American Speech-Language-Hearing Association.
Dinnsen, D.A. 1998. On the organization and specification of manner features. Journal of Lin-
guistics 34: 125.
Research 33: 2837.
Dinnsen, D.A. & Elbert, M. 1984. On the relationship between phonology and learning. In Pho-
nological Theory and the Misarticulating Child [ASHA Monographs 22], M. Elbert, D.A.
Dinnsen & G. Weismer (eds), 5968. Rockville MD: ASHA.
Dinnsen, D.A. & Gierut, J.A. (eds) 2008. Optimality Theory, Phonological Acquisition and Disor-
ders. London: Equinox.
Elbert, M. 1984. The relationship between normal phonological acquisition and clinical inter-
vention. In Speech and language: Advances in Basic Research and Practice, N.J. Lass (ed.),
111139. New York NY: Academic Press.
Elbert, M., Dinnsen, D.A. & Powell, T. 1984. On the prediction of phonologic generalization
learning patterns. Journal of Speech and Hearing Disorders 49: 309317.
Elbert, M. & McReynolds, L.V. 1975. Transfer of /r/ across contexts. Journal of Speech and Hear-
ing Disorders 40: 380387.
Elbert, M. & McReynolds, L.V. 1985. The generalization hypothesis: Final consonant deletion.
Language and Speech 28: 281294.
Elbert, M., Powell, T. & Swartzlander, P. 1991. Toward a technology of generalization: How many
exemplars are sufficient? Journal of Speech and Hearing Research 34: 8187.
Farris-Trimble, A. 2009. Weighted constraints and faithfulness cumulativity in phonological
acquisition. In Proceedings of the 33rd annual Boston University Conference on Language
Development, J. Chandlee, M. Franchini, S. Lord & G.-M. Rheiner (eds), 151162. Somerville
Ferguson, C.A. & Farwell, C.B. 1975. Words and sounds in early language acquisition: English
initial consonants in the first fifty words. Language 51: 419439.
Forrest, K. & Morrisette, M.L. 1999. Feature analysis of segmental errors in children with pho-
nological disorders. Journal of Speech, Language, and Hearing Research 42: 187194.
Gerken, L. 2004. Nine-month-olds extract structural principles required for natural language.
Cognition 93: B89B96.
Gerken, L. 2006. Decisions, decisions: Infant language learning when multiple generalizations
are possible. Cognition 98: B67B74.
Gerken, L. & Bollt, A. 2008. Three exemplars allow at least some linguistic generalizations: Im-
plications for generalization mechanisms and constraints. Language Learning and Develop-
ment 4: 228248.
Gerken, L., Wilson, R. & Lewis, W. 2005. Infants can use distributional cues to form syntactic
categories. Journal of Child Language 32: 249268.
Gierut, J.A. 2008a. Fundamentals of experimental design and treatment. In Optimality Theory,
Phonological Acquisition and Disorders, D.A. Dinnsen & J.A. Gierut (eds), 93118. London:
Equinox.
Gierut, J.A. 2008b. Phonological disorders and the Developmental Phonology Archive. In Opti-
mality Theory, Phonological Acquisition and Disorders, D.A. Dinnsen & J.A. Gierut (eds),
792. London: Equinox.
Gierut, J.A. & Morrisette, M.L. 2011. Effect size in clinical phonology. Clinical Linguistics &
Phonetics 25: 975980.
Gierut, J.A. & Morrisette, M.L. 2012. Age-of-word acquisition effects in treatment of children
with phonological delays. Applied Psycholinguistics 33: 121144.
Gierut, J.A., Morrisette, M.L. & Ziemer, S. 2010. Nonwords and generalization in children with
phonological disorders. American Journal of Speech-Language Pathology 19: 167177.
Hamann, S., Apoussidou, D. & Boersma, P. 2012. Modeling the formation of phonotactic re-
strictions across the mental lexicon. In Proceedings from the annual meeting of the Chicago
Linguistics Society 45, M. Ryan Bochnak, P. Klecha, A. Lemieux, N. Nassiar, J. Urban & C.
Weaver (eds), 193206. Chicago IL: Chicago University Press.
Ingram, D. 1989. Phonological Disability in Children. London: Cole and Whurr.
Jarosz, G. 2010. Implicational markedness and frequency in constraint-based computational
models of phonological acquisition. Journal of Child Language 37: 565606.
Jusczyk, P.W. 1997. The Discovery of Spoken Language. Cambridge MA: The MIT Press.
Labov, W. 1981. Resolving the Neogrammarian controversy. Language 57: 267309.
Leonard, L.B. & Brown, B.L. 1984. Nature and boundaries of phonologic categories: A case study
of an unusual phonologic pattern in a language-impaired child. Journal of Speech and Hear-
ing Disorders 49: 419428.
Locke, J. 1983. Phonological Acquisition and Change. New York NY: Academic Press.
Logan, J.S. 1992. A Computational Analysis of Young Childrens Lexicons. Bloomington IN:
Speech Research Laboratory, Indiana University.
Maye, J., Weiss, D.J. & Aslin, R.N. 2008. Statistical phonetic learning in infants: Facilitation and
feature generalization. Developmental Science 11: 122134.
McReynolds, L.V. & Kearns, K.P. 1983. Single-subject Experimental Designs in Communicative
Disorders. Baltimore MD: University Park Press.
Morrisette, M.L. 1999. Lexical characteristics of sound change. Clinical Linguistics & Phonetics
13: 219238.
Morrisette, M.L. 2004. Recursive change in phonological acquisition. In From Sound to Sense:
50+ Years of Discoveries in Speech Communication, J. Slifka, S. Manuel & M. Matthies (eds),
6772. Boston MA: MIT.
Needham, A., Ducker, G. & Lockhead, G. 2005. Infants formation and use of categories to seg-
regate objects. Cognition 94: 215240.
Newport, E.L. & Aslin, R.N. 2004. Learning at a distance, I: Statistical learning of non-adjacent
dependencies. Cognitive Psychology 48: 127162.
Pater, J. 2008. Gradual learning and convergence. Linguistic Inquiry 39: 334345.
Peters, A.M. & Strmqvist, S. 1996. The role of prosody in the acquisition of grammatical mor-
phemes. In Signal to Syntax: Bootstrapping from Speech to Grammar in Early Acquisition,
J.L. Morgan & K. Demuth (eds), 215232. Mahwah NJ: Lawrence Erlbaum Associates.
Phillips, B.S. 1983. Lexical diffusion and function words. Linguistics 21: 487499.
Phillips, B.S. 1984. Word frequency and the actuation of sound change. Language 60: 320342.
Quinn, P.C. & Bhatt, R.S. 2005. Learning perceptual organization in infancy. Psychological Sci-
ence 16: 511515.
Rice, K. & Avery, P. 1993. Segmental complexity and the structure of inventories. Toronto Work-
ing Papers in Linguistics 12: 131153.
Rvachew, S. 1994. Speech perception training can facilitate sound production learning. Journal
of Speech and Hearing Research 37: 347357.
Shriberg, L.D., Kwiatkowski, J. & Snyder, T. 1989. Tabletop versus microcomputer-assisted speech
management: Stabilization phase. Journal of Speech and Hearing Disorders 54: 233248.
Shriberg, L.D., Kwiatkowski, J. & Snyder, T. 1990. Tabletop versus microcomputer-assisted
speech management: Response evocation phase. Journal of Speech and Hearing Disorders
55: 635655.
Smith, N.V. 1973. The Acquisition of Phonology: A Case Study. Cambridge: CUP.
Stoel-Gammon, C. 2011. Relationships between lexical and phonological development in young

children. Journal of Child Language 38: 134.
Stoel-Gammon, C. & Cooper, J.A. 1984. Patterns of early lexical and phonological development.
Journal of Child Language 11: 247271.
Tesar, B. & Smolensky, P. 1998. Learnability in optimality theory. Linguistic Inquiry 29: 229268.
Tyler, A.A. & Edwards, M.L. 1993. Lexical acquisition and acquisition of initial voiceless stops.
Van Riper, C. 1978. Speech Correction: Principles and Methods, 6th edition. Englewood Cliffs NJ:
Prentice-Hall.
Vihman, M. & Keren-Portnoy, T. 2011. The role of production practice in lexical and phono-
logical development A commentary on Stoel-Gammons Relationships between lexical
and phonological development in young children. Journal of Child Language 38: 4145.
Walley, A.C. & Metsala, J.L. 1992. Young childrens age-of-acquisition estimates for spoken
words. Memory & Cognition 20: 171182.
Weiner, F.F. 1981. Treatment of phonological disability using the method of meaningful mini-
mal contrast: Two case studies. Journal of Speech and Hearing Disorders 46: 97103.
White, K.S., Peperkamp, S., Kirk, C. & Morgan, J.L. 2008. Rapid acquisition of phonological al-
ternations by infants. Cognition 107: 238265.
Williams, A.L. 2012. Intensity in phonological intervention: Is there a prescribed amount? Inter-
national Journal of Speech-Language Pathology 14: 456461.
Xu, F. & Tenenbaum, J.B. 2007. Word learning as Bayesian inference. Psychological Review 114:
245272.
Rapid phonological coding and working
memory dynamics in children with cochlear
implants
Cognitive foundations of spoken language processing
David B. Pisoni
Indiana University, Bloomington & Indiana University School of Medicine
Past research on cochlear implants in deaf children has been narrowly focused
on speech and language outcomes and efficacy of cochlear implantation as
a medical treatment for profound hearing loss. Little, if any, basic or clinical
research has investigated the underlying neurobiological and cognitive factors
that are responsible for the enormous individual differences and variability in
the effectiveness of cochlear implants in this unique clinical population. Our
research over the last twenty years has demonstrated that a small number of
core elementary cognitive processes such as rapid phonological coding, working
memory dynamics, processing speed, inhibition and sequence learning are
strongly associated with traditional endpoint clinical speech and language
outcome measures. These foundational cognitive processes reflect the global
coordination, integration and functional connectivity of multiple underlying
brain systems used in speech perception, production and spoken language
processing. Our findings on the underlying sources of variability in spoken
language processing provide a new theoretical framework that can help to
understand and explain the enormous individual differences in speech and
language outcomes following cochlear implantation. The results of our research
have direct clinical implications for improving the diagnosis, treatment and early
identification of young profoundly deaf children who may be at high risk for
poor speech and language outcomes following cochlear implantation.
Introduction
There are many ways that individuals can influence a particular field of study and
have their research findings and theoretical ideas produce lasting transformative
effects on the direction of future research and clinical application. Very often,
David B. Pisoni
simply discovering a new empirical finding, developing a novel experimental

methodology or introducing a fundamentally different theoretical framework or
conceptualization can have significant effects on research and practice. This is es-
pecially true in the clinical specialties like speech-language pathology or clinical
audiology which draw very heavily on basic science from many different allied
disciplines such as Linguistics, Psychology, Engineering and Neuroscience.
In 1984 Dan Dinnsen and two of his colleagues from Speech and Hearing
Sciences at Indiana University, Mary Elbert and Gary Weismer, published a highly
influential ASHA Monograph entitled Phonological Theory and the Misarticulat-
ing Child. This seminal monograph on phonological disorders in children along
with a series of other publications has transformed the field of phonological disor-
ders in children and has significantly and permanently impacted the development
of several highly effective novel clinical interventions that are now used routinely
to treat children with various types of phonological disorders. The most important
and distinctive feature of this monograph was the proposal of a fundamentally
new theoretical perspective that phonological disorders in children should be
viewed within the broader context of language, and language acquisition and pho-
nological disorders reflect underlying processes that are linguistic in nature and
are not due to sensory-motor delays related to the development of speech motor
control and speech articulation.
The proposal that phonological disorders in otherwise typically-developing
normal-hearing children were due to underlying linguistic processes was a very
radical idea at the time in the field of speech-language pathology. I remember at-
tending Judith Gieruts PhD defense in 1985 and explaining to Professor Charles
Watson, then chair of Indiana Universitys Speech and Hearing Department, that
the observed atypical/disordered surface forms displayed in the articulation of
children with phonological disorders reflected irregularities and disturbances in
the underlying linguistic system that the child had acquired and were not due to
some idiopathic sensory-motor reorganization or disturbances in peripheral mo-
tor control processes.
Recourse to a linguistic account of phonological disorders in functional mis-
articulators, as these children were called years ago, was completely new to
Professor Watson who was trained at Indiana University back in the 1950s to think
like a behaviorist-oriented experimental-sensory psychologist. At the time this
event took place, Watson was unaware of the revolution that had already taken
place in the field of linguistics a decade or so earlier and how it fundamentally
changed our views about language and language behavior. At the same time Dan
and his colleagues had been pioneering the translation of linguistic principles of
phonological theory to the study of phonological disorders in children.
In many ways, I feel a strange sense of dj vu myself right now being im-
mersed in something like a timewarp back to the early 1980s because the field that
I have been working in, cochlear implants in deaf children, is undergoing a very
similar transformation in thinking at the present time. For the last twenty years,
Ive been trying to convince clinicians and researchers who work with profoundly
deaf children and adults that the ear is connected to the brain and that individual
parts of the brain work together as a functionally integrated system. As I explain to
the Neurotologists the CI surgeons, the clinical audiologists, speech-language
pathologists and assorted hearing scientists who work with deaf children the ear
is not like the muffler in your car. It cant simply be replaced by a neural prosthe-
sis, a bionic ear. Not only is the ear connected to the brain, but more importantly,
the brain is connected to the ear via many reciprocal neural connections and path-
ways forming a complex highly coordinated embodied information processing
system linking brain, body and world (Clark 1997).
My primary mission has been to document that an additional portion of the
variance observed in the conventional speech and language outcomes obtained
after cochlear implantation beyond the usual demographic, medical, educational
and device variables reflects the inseparable contribution of several elementary
core cognitive information processing operations. The enormous individual differ-
ences and variability in speech and language observed in profoundly deaf children
after they receive a cochlear implant are not due only to hearing and early sensory
coding of speech signals by the auditory periphery; in fact, on closer examination,
most profoundly deaf children who receive cochlear implants turn out to have
other comorbid neuropsychological conditions and often display subtle subclini-
cal delays and/or disturbances in several more basic underlying domain-general
cognitive processes that are not strictly sensory in nature. These differences reflect
fundamental aspects of how the entire human information processing system
adapts and reorganizes itself in response to highly degraded and underspecified
acoustic-phonetic information supplied to the brain by the current generation of
cochlear implants. In the sections below, I will provide a brief summary of the
major findings of our research program to illustrate how several novel experimen-
tal methodologies and theoretical ideas from the field of cognitive psychology have
provided fundamentally new insights and understanding of individual differences
and variability in deaf children with cochlear implants (Pisoni 2000).
One of the long-term goal of our research program is to understand, explain
and predict the enormous variability in speech and language outcomes in deaf
children who have received CIs as a medical treatment for profound deafness. In-
dividual differences and variability in speech and language outcomes are signifi-
cant clinical problems that have not been addressed adequately in the past. Little,
if any, solid progress has been made in understanding the core neurobiological
David B. Pisoni
and cognitive mechanisms and information processing operations that are re-
sponsible for the variability observed in speech and language outcomes following
cochlear implantation. Almost all of the research in the field of CIs has been con-
cerned primarily with investigating the efficacy of cochlear implantation as a
treatment for profound deafness, focusing almost exclusively on demographic, de-
vice, medical and educational variables as predictors of outcome and benefit.
With the average age of implantation steadily decreasing to 12 months and
below, the ability to reliably predict the effectiveness of CIs from behavioral mea-
sures obtained from infants and young children prior to implantation is absolutely
critical to providing appropriate intervention and habilitation. Understanding
sources of variability in speech and language outcomes following cochlear implan-
tation is a very complex and challenging clinical problem that requires a
collaborative multidisciplinary approach to research from scientists working in
neuroscience, cognitive psychology, developmental science and cognitive neurop-
sychology. These problems are not the property of only one academic discipline;
they are complex and involve interactions among many variables.
Core findings on speech and language outcomes after cochlear implantation
Large individual differences in outcome and benefit
Although CIs work well for many profoundly deaf children, they do not always
provide optimal benefits to all children who receive them. Some children do ex-
tremely well with their implants and display near-typical age-appropriate speech
and language skills on a wide range of traditional clinical tests when administered
under quiet listening conditions in the clinic or laboratory. In contrast, other deaf
children often struggle for long periods of time after they receive their CIs and
frequently never achieve comparable levels of speech and language performance
or verbal fluency. If we can identify the reasons why a good user is performing so
well, we should be able to use this basic fundamental knowledge to help poorer-
performing children improve their speech and language skills and achieve their
potential to derive optimal benefits from their CIs.
What is a cochlear implant and how does it work?
A cochlear implant is a surgically implanted electronic device that serves as a sen-

sory aid or auditory prosthesis for hearing-impaired patients who have bilateral
severe-to-profound sensorineural hearing loss (See Figure 1). Cochlear implants
provide electrical stimulation to the surviving spiral ganglion cells of the auditory
1
3
2
4
Figure 1. Internal and external components of a multichannel cochlear implant system,

consisting of a microphone (1) that picks up sound from the environment, a speech
processor (2) that converts sound into electrical signals, a surgically implanted receiver,
(3) and an internal electrode array inserted in the cochlea (4) that sends electrical pulses
to each electrode tonotopically. (Source: Courtesy of Cochlear Americas.)
nerve, bypassing the damaged hair cells of the inner ear to restore the sense of
hearing in both adult and pediatric populations. Cochlear implants provide deaf
adults and children with access to sound from the auditory sensory modality but
they do not restore normal hearing. Based on data reported by the NIH in 2011,
219,000 patients have received CIs worldwide; 42,600 adults and 28,400 children
currently have CIs in the USA.
The current generation of multichannel cochlear implants all have two com-
ponents: an internal receiver which is connected to an electrode array that is in-
serted in the cochlea and an external signal processing unit that transmits digital
signals to the internal receiver using telemetry. The external processing unit con-
sists of a microphone that picks up sound from the environment and a signal pro-
cessor that codes frequency, amplitude and time and compresses the signal to
match the dynamic range of the ear. Cochlear implants do a good job at encoding
temporal and amplitude information. Depending on the manufacturer, different
place coding schemes are used to encode, represent and transmit frequency infor-
mation in the signal.
For postlingually profoundly deaf adults, a cochlear implant provides a trans-
formed electrical signal to an already fully developed auditory system and intact
mature language processing system. These patients have already acquired spoken
language under typical listening conditions so we know their central auditory
David B. Pisoni
system and brain are functioning normally. However, in the case of a congenitally
deaf child, a cochlear implant provides novel electrical stimulation through the
auditory sensory modality. The electrical stimulation from a CI provides access to
sound and an opportunity to perceive speech and develop spoken language for the
first time after a period of auditory deprivation. Congenitally deaf children have
not been exposed to sound and speech, and therefore they do not develop spoken
language normally. Although the brain and nervous system of profoundly deaf
children continue to develop in the absence of normal auditory stimulation, sub-
stantial converging evidence suggests that cortical reorganization has already tak-
en place during the period of sensory deprivation before implantation. As a result,
many speech and language processing skills are delayed and may develop in an
atypical manner after implantation.
Preimplant predictors of speech and language outcomes
Until recently, clinicians and researchers have been unable to find reliable preim-
plant predictors of outcome and benefit with a cochlear implant (see, however,
Bergeson & Pisoni 2004; Horn, Pisoni, Sanders & Miyamoto 2005). The failure to
identify reliable preimplant predictors is a theoretically significant finding because
it suggests that many complex interactions take place between the newly acquired
sensory capabilities of a child after a period of auditory deprivation, attributes of
the language-learning environment and various interactions with parents and
caregivers that the child is exposed to immediately after receiving a cochlear im-
plant. More importantly, the lack of preimplant predictors of outcome and benefit
also makes it difficult for clinicians to reliably identify those children who may be
at high risk for poor outcomes with their CI at a time in neural, cognitive and lin-
guistic development when interventions can be made to modify and improve their
language processing skills.
Age of implantation
Age of implantation is one of the primary demographic factors that has been found
to influence almost all outcome measures of performance. Children who receive
an implant at a young age consistently perform better on a wide range of conven-
tional behavioral measures than children who are implanted at older ages (Kirk,
Pisoni & Miyamoto 2000; Niparko et al. 2010). Length of auditory deprivation or
duration of deafness is also related to outcome and benefit. And children who have
been deaf for shorter periods of time before implantation also do much better
on a wide variety of clinical measures than children who have been deaf longer.
Both findings, age of implantation and duration of deafness before implantation,
demonstrate the contribution of neural plasticity and the role of sensitive periods
in sensory, perceptual, cognitive and linguistic development and establish the close
links between neural development and behavior, especially the development of
hearing, speech and spoken language skills (Konishi 1985; Marler & Peters 1988).
Communication mode: Experience- and activity-dependent learning
The nature of the early linguistic and educational environment that a deaf child is
immersed in after receiving a CI has been found to affect performance on a wide
range of speech and language outcome measures (Geers et al. 2003; Kirk 2000).
Children who are educated in auditory-oral environments that emphasize oral
communication (OC) methods and the development of fluent speaking and listen-
ing skills consistently do better on conventional behavioral tests of speech percep-
tion, open-set spoken word recognition and spoken language comprehension. In
contrast, children who are placed in total (or simultaneous) communication
(TC) environments are exposed to and encouraged to use both manual signs along
with spoken language; these children acquire manual signs using English syntax as
well as spoken language, but they typically do more poorly on receptive and ex-
pressive language tests that rely on the use of highly automatized phonological
processing skills than children who are educated using auditory-oral methods.
Very few deaf children with CIs learn ASL along with spoken language. Children
who use auditory-oral communication modes also produce more intelligible
speech and show significant gains in expressive language.
Product vs. process measures
Almost all of the previous clinical research on CIs has focused on the effects of a
small number of medical, device and demographic variables using traditional
product outcome measures based on assessment tools developed by clinical au-
diologists and speech-language pathologists (see Kirk 2000). These assessments
relied exclusively on accuracy measures of performance in conventional behav-
ioral tasks assessing speech perception, spoken word recognition, sentence com-
prehension, vocabulary knowledge and speech intelligibility. Although rarely
discussed explicitly in the literature, all of these clinical outcome measures are the
final product or endpoint of a large number of complex multiparametric sen-
sory, perceptual, cognitive and linguistic processes that contribute to the observed
variation among CI users. Until recently, no one had obtained process measures
of performance to examine the more basic underlying elementary information
processing mechanisms used to perceive and produce spoken language in this
clinical population (Pisoni & Cleary 2004; Pisoni 2005). Our recent findings,
David B. Pisoni
summarized briefly in the sections below using a variety of novel process measures
of performance, have provided important new fundamental knowledge and un-
derstanding about the neuro-cognitive basis of individual differences and variabil-
ity in profoundly deaf children with CIs.
Learning and memory processes
Finally, when all of the outcome and demographic measures are considered
together, the available evidence strongly suggests that the underlying sensory, per-
ceptual and cognitive abilities for speech and language emerge after implanta-
tion. Successful outcomes with a cochlear implant are also due, in large part, to
learning and memory processes and to exposure to a rich language model in the
childs language learning environment (Clark 2003). Because outcome and benefit
with a cochlear implant cannot be predicted reliably from traditional behavioral
measures obtained before implantation, any improvements in performance ob-
served after implantation must be due to sensory, cognitive and linguistic processes
that are closely linked to learning and maturational changes in brain development
(see Sharma, Dorman, & Spahr 2002).
The information processing approach to cognition
To understand, explain and predict variability in outcome and benefit, it is neces-

sary to situate the problem of individual differences in this clinical population in a
much broader theoretical framework that fully acknowledges that variability in
brain-behavior relations is a natural consequence of biological development of all
living systems (Sporns 1998). The enormous variability observed in a wide range
of outcome measures following implantation may not be unique to this particular
clinical population but may reflect instead more general underlying sources of
variability observed in speech and language processing in healthy typically-
developing normal-hearing adults and children (Cicchetti & Curtis 2006). More-
over, it is very likely that the sources of the individual differences observed in
speech and language outcomes in deaf children with CIs also reflect variation in
both domain-general and domain-specific cognitive processes (Geers & Moog
1987; Ullman & Pierpont 2005).
In order to investigate the sources of variability in performance and under-
stand the neural and cognitive processes that underlie variation in outcome and
benefits following implantation, it has been necessary to broaden the outcome
measures to assess a much wider range of behaviors and information process-
ing skills beyond just the traditional clinical audiological speech and language
assessment measures that have been routinely used in the past (Pisoni 2000). Fur-
thermore, a childs failure to obtain optimal benefits and achieve age-appropriate
speech and language milestones from his/her CI may not be due directly to the
functioning of the CI itself but may reflect complex multiparametric interactions
among a number of contributing factors (Geers, Brenner & Davidson 2003). We
have adopted the general working assumption in our research that many profound-
ly deaf children who use CIs, especially the low-performing users, may have other
neural, cognitive and affective sequelae resulting from a period of auditory depriva-
tion combined with a delay in language development before implantation. Thus,
our hypothesis is that the enormous variability observed in speech and language
outcomes is not only due to hearing or the early sensory encoding of speech but
also reflects the contribution of other neurocognitive factors related to how sensory
information is encoded, stored, and retrieved from memory (see Figure 2).
Brain-behavior relations
Our approach to the problems of variability in outcome and benefit following co-
chlear implantation is motivated by recent findings and theoretical developments
that suggest that deafness and hearing impairment cannot be viewed in isolation
as a simple sensory impairment (Pisoni et al. 2008, 2010). The enormous variability
Information storage
Short term
Registration Long term
and working
(Encoding) memory
memory
Sensing Action
Information
information functions
Input processing and Output
[Receiving] [Physical
decisions
control]
Flow of Information Processing in the nervous system
Figure 2. Simplified schematic view of the basic flow of information within the human
operator/observer involving sensory encoding, information storage and processing and
action or control. The general systems approach shown here linking early sensory
processes, short-term/working memory and long-term storage is applicable to a wide
range of behavioral tasks used to assess performance and identify sources of individual
differences in different populations
David B. Pisoni
in outcome and benefit reflects numerous complex neural and cognitive processes
that depend heavily on functional connectivity of multiple brain areas working
together as a complex integrated system (Luria 1973). As W. Nauta (1964) pointed
out almost 50 years ago, no part of the brain functions on its own, but only through
the other parts of the brain with which it is connected (p. 125). Except for our
recent studies, measures of system-wide brain coordination and functional con-
nectivity reflecting cognitive processes that are critical for robust highly adaptive
speech and language behaviors have not been routinely obtained in deaf children
with CIs. We believe this is a promising new direction to pursue in clinical research
on individual differences in profoundly deaf children who have received CIs.
Domain-general cognitive factors
Our recent work with preimplant visual-motor integration (VMI) tests that use
only visual patterns requiring reproduction and construction processes has un-
covered significant correlations with a range of conventional clinical speech and
language outcome measures obtained from deaf children following implantation
(Horn, Davis, Pisoni & Miyamoto 2004). These findings obtained using nonverbal
visual information processing tasks suggest that the underlying elementary cogni-
tive information processes used in open-set speech perception and spoken lan-
guage processing tasks may be much more general in nature drawing heavily on
resources from multiple neural systems and brain circuits linking sensory and mo-
tor systems that are not modality- or even domain-specific. Similarly, our findings
on working memory capacity, verbal rehearsal speed and scanning of verbal infor-
mation in short-term memory (STM), visual sequence memory, implicit learning
of probabilistic statistical patterns and talker discrimination all suggest that an
important additional source of variance in speech and language outcomes in deaf
children with CIs is associated with domain-general cognitive processes that in-
volve executive attention, cognitive control and self-regulation (Blair & Razza-
Peters 2007; Oberauer, Sub, Wilhelm & Wittmann 2003; Oberauer 2005). There is
now good agreement among cognitive scientists that these cognitive control pro-
cesses rely critically on global system-wide executive attention processing opera-
tions that reflect organization-integration, coordination, functional connectivity
and close interactions of multiple neural circuits and subsystems that are widely
distributed across numerous areas of the brain (Sporns 2003).
Executive function and cognitive control processes
Converging evidence also suggests that a set of domain-general executive control

processes may also underlie a substantial portion of the variance found in speech
and language outcomes of deaf children with CIs. Executive control processes are
defined here as a broad cluster of related information processing operations that
involve cognitive control, attentional focus, concentration and the construction of
holistic symbolic internal representations from individual parts or components.
Executive control reflects the demands for sustained, purposeful, organized infor-
mation processing in the service of producing a unified coordinated response to
complex multidimensional stimuli. These executive control abilities are mediated
by neural circuits in the prefrontal cortex, which has extensive functional connec-
tivity with several other brain regions (McNab & Klingberg 2008). Executive
control shows rapid development during early to middle childhood years, with
reliable differences beginning to emerge in toddlers and persisting through adult-
hood (Isquith, Crawford, Espy & Gioia 2005). These executive function domains
are related to a variety of important outcomes ranging from behavioral adjustment
to critical areas of quality of life (Barkley 1997; 2006). We have argued recently that
executive control abilities are an inseparable component of speech-language de-
velopment because of strong bidirectional reciprocal relations between the devel-
opment of spoken language processing skills and the development of cognitive
control abilities (Kronenberger et al. 2013).
The unifying hypothesis motivating our approach to individual differences
and variability in outcomes is that many deaf children who use CIs may display
delays and/or dysfunctions in several cognitive information processing domains
in addition to their primary hearing loss and language delay. Moreover, many deaf
children with CIs may not show age-appropriate scores on a variety of neuropsy-
chological tests that on the surface appear to have little, if anything, directly to do
with domain- and/or modality-specific aspects of hearing, speech perception and
spoken language processing, but may reflect instead domain-general cognitive
processes. Variability in these basic elementary information processing skills may
ultimately be responsible for the unexplained variability and individual differences
in audiological outcome measures of CI benefit that are routinely observed at all
cochlear implant centers around the world.
Detection and discrimination vs. categorization and classification
It is important to emphasize strongly here that we are not only interested in

whether deaf children with CIs are able to hear that is, whether they can de-
tect and discriminate the sensory attributes of simple and complex acoustic sig-
nals after receiving their cochlear implant. More importantly, we are interested in
how deaf children process, that is, how they encode, store and retrieve the sen-
sory information that they are able to hear. We are interested in the nature of the
phonological and lexical representations that deaf children construct from highly
David B. Pisoni
degraded underspecified acoustic signals delivered by the CI and how do they use
this coarsely-coded impoverished information about speech in a range of differ-
ent language and cognitive processing tasks. If we have a better understanding of
the underlying causal mechanisms of action, we can modify the ways in which
the central auditory, cognitive and linguistic processes are carried out in deaf
children with CIs, especially processes related to learned attention, working
memory capacity and processing speed, in principle, it should then be possible to
provide significant benefits to the low-performing children with CIs using novel
interventions.
Analysis of The Stars The extraordinary CI users
We began our investigation of individual differences and sources of variability in

speech and language outcomes by analyzing a set of clinical data obtained from a
group of exceptionally good CI users (Pisoni 1997). These deaf children, often re-
ferred to in the field as the Stars, acquire spoken language quickly and easily after
implantation and show a developmental trajectory that parallels normal-hearing
children (see Svirsky, Teoh & Neurburger 2004). Clinical outcome measures of
speech perception and language in these exceptionally good users (Pisoni 1997;
Pisoni, Cleary, Geers & Tobey 2000) were all found to be strongly inter-correlated
with each other, suggesting the existence of a common underlying source of vari-
ance related to phonological processing of speech (Pisoni & Cleary 2003; Pisoni
2005). Unfortunately, the nature of the outcome measures obtained from these
children prevented us from moving forward in new directions.
The traditional clinical outcome measures that we examined in this early study
were all product or endpoint measures of performance that reflect a long series
of sensory, perceptual and cognitive processes. Process measures designed to as-
sess how well a child uses the sensory information provided by his/her cochlear
implant were not included in any of the standard clinical protocols at the time
these data were collected, so it was impossible to assess information capacity, pro-
cessing speed, memory and learning, inhibition or organizational-integrative pro-
cesses, cognitive factors which may be central for determining which children will
become good CI users and which children will be at high risk for poor outcomes.
New data collected from another group of pediatric CI users a few years later pro-
vided the first solid evidence that individual differences in short-term working
memory capacity, processing speed and sequence learning might be responsible
for the variation in performance, and may contribute to the differences in speech
perception and spoken language processing skills observed in prelingually deaf
children with CIs (Pisoni & Cleary 2004).
Process measures of outcome and benefit
Verbal short-term and working memory capacity
Differences in the information processing capacity of immediate memory of pedi-

atric CI users have been found to underlie variability in a large number of different
cognitive tasks. In a large group of 8- and 9-year old CI users (N=176), we found
that both forward and backward digit spans were significantly shorter than the
digit spans obtained from a group of age-matched normal-hearing children as well
as normative data from the WISC III (see Pisoni & Cleary 2003). Both forward and
backward digit spans reflect differences in rapid phonological coding and storage
of order information and suggest that pediatric CI users may have fundamental
limitations in processing capacity of working memory compared to age-matched
normal hearing children. Moreover, we also found that children who were en-
rolled in oral communication (OC) programs had longer forward and backward
digit spans than children enrolled in total communication (TC) programs, sug-
gesting direct links between linguistic experience and processing mechanisms.
These results indicate that early experience and activities in an educational
environment that emphasizes auditory-oral language skills are associated with lon-
ger forward digit spans and increased information processing capacity of working
memory. However, without additional converging measures, it is difficult to iden-
tify precisely which information processing mechanisms are actually affected by
early auditory and linguistic experience and which ones are responsible for the
increases in forward digit spans observed in these children. Correlational analyses
of the data also revealed that pediatric CI users who had longer forward digit spans
also had higher scores on spoken word recognition tests even after confounding
variables such as chronological age, communication mode, duration of deafness,
duration of device use, age at onset of deafness and number of active electrodes
were removed. These correlations suggest that measures of immediate memory
capacity are fundamentally linked with speech perception (Pisoni & Cleary 2003).
Verbal rehearsal speed
In typically-developing children and adults, verbal rehearsal speed (VRS) is

strongly correlated with immediate memory capacity as indexed by forward digit
spans as well as measures of spoken word recognition (see Pisoni & Cleary 2004).
Estimates of the childrens verbal rehearsal speed were also obtained in this study
by measuring speaking rates from spoken sentences (see Pisoni & Cleary 2003).
The 168 pediatric CI users that we examined displayed slower verbal rehearsal
speeds than a group of age-matched normal hearing control children. Moreover,
David B. Pisoni
verbal rehearsal speed was found to be negatively correlated with forward digit
span; children who produced sentences with longer durations spoke more slowly
and, in turn, displayed shorter forward digit spans (r = .59; p < .001). It is impor-
tant to emphasize that the relations we observed between digit span and speaking
rate were selective in nature and were found only for the forward digit spans. Sen-
tence duration was also found to be significantly correlated with spoken word rec-
ognition scores, suggesting that verbal rehearsal may be the common information
processing mechanism that underlies the variability and individual differences
observed in these outcome measures: if deaf children with CIs take longer to ini-
tially encode phonological information in short-term memory and also have slow-
er verbal rehearsal processes, they will necessarily have shorter digit spans at the
time of recall (Baddeley, Thomson & Buchanan 1975; Hulme, Thomson, Muir &
Lawrence 1984).
Scanning of information in short-term memory
Scanning is a term that refers to how fast an individual can access and retrieve
information from short-term memory. We also investigated retrieval dynamics
and scanning of verbal information in STM using the same digit span recall task
but now we measured the pause durations between digits in the vocal response
protocols (see Burkholder & Pisoni 2003). We found that although deaf children
with CIs could correctly recall all the items from three- and four-digit lists, their
retrieval speed was on the average three times slower than the retrieval speed of a
group of age-matched normal-hearing children suggesting less robust perceptual
encoding and slower active maintenance of phonological representations in verbal
working memory. We also found that early linguistic experience and activities in-
fluenced the rate of STM scanning; children who were fully immersed in auditory-
oral (OC) educational programs displayed faster STM scanning rates than children
who were enrolled in total communication (TC) programs. The links observed
between phonological encoding and STM scanning speed suggest that childrens
working memory processes are flexible and can be influenced by learning, experi-
ence and processing activities in the language learning environment. Additional
measures of digit span and verbal rehearsal speed have been obtained from 112 of
the original children who returned after more than ten years of CI use (see Geers
et al. 2011; Pisoni et al. 2011). The results of these recent analyses revealed strong
correlations among the measures over time and suggest that both memory span
and verbal rehearsal speed are reliable predictors of the development of speech
and language in this clinical sample after ten years of CI use.
Nonword repetition: Phonological decomposition and reassembly
To gain further insights into the underlying cognitive and linguistic processes that
are responsible for the variability in speech and language outcomes following
implantation, we used a nonword repetition task to investigate how deaf children
with CIs make use of sublexical phonological knowledge (Carter, Dillon & Pisoni
2002; Cleary, Dillon & Pisoni 2002; Dillon, Cleary Pisoni, & Carter 2004; Dillon
Burkholder, Cleary & Pisoni 2004). Non-word repetition is strongly correlated with
vocabulary development and other language learning milestones in normal hearing
children (Gathercole & Baddeley 1990; Gathercole, Hitch, Service & Martin 1997).
In our study, 88 pediatric CI users were asked to listen to nonsense words that con-
formed to English phonology and phonotactics such as ballop, bannifer, or
altupatory and immediately repeat back what they heard. Several linguistic mea-
sures of their performance on this task were obtained and then correlated with
performance measures of open-set word recognition using the Lexical Neighbor-
hood Test (LNT), Forward Digit Span, Speech Intelligibility, Speaking Rate, Word
Attack, and Rhyme Errors. The Word Attack and Rhyme Errors were obtained from
an isolated single word reading task. As summarized in Table I, the consonant and
vowel transcription scores as well as the perceptual ratings of the nonwords by a
separate group of normal-hearing English-speaking adults were all very strongly
correlated with the clinical outcome measures, suggesting that a common set of
phonological representations and information processing operations are used
across a wide range of language processing tasks, even single word reading tasks
that used visual presentation. Although nonword repetition at first glance appears
Table I. Partial Correlations Between Nonword Repetition Scores and Several

Conventional Speech and Language Outcome Measures
Consonants Vowels Accuracy Ratings

(N = 76) (N = 76) (N = 76)
LNT (Easy Words) +.83*** +.78*** +.76***

LNT (Hard Words) +.85*** +.71*** +.70***
MLNT +.77*** +.74*** +.77***
Forward Digit Span +.60** +.62** +.76***
Speech Intelligibility +.91*** +.88*** +.87***
Speaking Rate .84*** .81*** .85***
Word Attack (Reading) +.75*** +.72*** +.78***
Rhyme Errors (Reading) .63** .68** .54*
*p < .05, **p < .01, ***p < .001
[Based on data reported in Cleary, Dillon & Pisoni 2002; Carter, Dillon & Pisoni 2002; Dillon, Cleary, Pisoni
& Carter 2004; Dillon, Burkholder, Cleary & Pisoni 2004]
David B. Pisoni
to be a simple information processing task involving reproduction of an ordered

sequence of phonological segments, it is actually a very complex multiparametric
psycholinguistic process that requires the child to perform well on each of the in-
dividual component processes involving speech perception and perceptual encod-
ing, phonological decomposition, active verbal rehearsal and maintenance in
working memory, retrieval of an ordered temporal sequence of phonemes, phono-
logical reassembly and phonetic implementation in speech production and speech
motor control. Moreover, the nonword repetition task requires the active use of
organizational-integrative processes linking these individual component process-
es together to produce a unitary coordinated motor output response.
The findings from our nonword repetition studies provide additional support
for the hypothesis that an important source of variability in clinical outcomes is
related to the use of the rapid automatic phonological processing skills that are
needed to construct robust, stable and reliable phonological representations of
spoken words in working memory which are then used in a wide range of different
language processing tasks. Additional nonword repetition data were obtained
more recently from these children after ten years of CI use (Casserly & Pisoni
2013). Strong correlations were found between the initial nonword repetition
scores obtained in elementary school and the follow up data obtained in high
school suggesting stability over time. More importantly, like the digit span and
verbal rehearsal speed measures described earlier, nonword repetition skills were
found to be a robust predictor of future performance on a wide range of conven-
tional speech and language outcome measures obtained after more than ten years
of CI use in this clinical sample (Casserly & Pisoni 2013).
Executive function, inhibition and cognitive control processes
Our research also suggests that several aspects of executive function may be dis-
turbed in deaf children with CIs. Executive function is an umbrella term used in
Cognitive Psychology that includes several different information processing do-
mains such as attention, cognitive control, working memory, inhibition, planning,
reasoning and problem solving. Many researchers believe that executive function
involves using prior knowledge and experience to predict future events and
modulate the current contents of immediate memory and conscious attention
(Goldman-Rakic 1988). Executive function also plays an important role in recep-
tive and expressive language processes via top-down feedback and control of in-
formation processing activities in a wide range of behavioral tasks many of which
are used routinely in the clinic to assess speech and language outcomes and ben-
efits (Kronenberger et al. 2013).
Inhibition processes in speech perception
A large body of research has shown that normal-hearing listeners encode and pro-
cess attributes of speech that provide indexical or vocal source information about
the talker (e.g., gender, age, regional background, emotional state, etc.). Although
normal hearing listeners have very little difficulty identifying whether two sen-
tences were produced by two different talkers, deaf children with CIs have much
more difficulty in this simple task (Cleary & Pisoni 2002). If the linguistic content
of the two sentences is identical (i.e., fixed-sentence context), children with CIs
can perform the task better than chance (67% correct), although their scores are
significantly lower than normal hearing childrens scores (100% correct). However,
when the linguistic content of the two sentences differs (i. e., varied-sentence
context), children with CIs are no longer able to do this simple voice discrimina-
tion task with short meaningful English sentences. Their performance did not dif-
fer from chance (58% correct) and they were significantly impaired compared to a
younger group of normal hearing subjects (89% correct).
These findings on inhibition and cognitive control processes in a simple speech
discrimination task are theoretically significant because they suggest that children
with CIs have considerable difficulty inhibiting irrelevant semantic information in
tasks requiring active cognitive control. When both sentences were spoken by the
same talker in the fixed-sentence condition, the child can simply judge whether
the voice is the same in both conditions because active cognitive control of atten-
tion to a competing response is not mandatory. However, in the varied-sentence
condition, the child must be able to ignore and inhibit a competing response to the
differences in meaning in the two sentences in order to make a reliable judgment
about the speakers voice. An examination of the errors produced in the varied-
sentence condition showed that the CI users displayed a significant response bias
to incorrectly respond different more often than same for pairs presented in
this condition. In contrast, the normal-hearing children showed no evidence of
any response bias at all in this condition. The differences observed in talker dis-
crimination performance across these two sentence conditions suggest that basic
sensory discriminative capacities related to audibility of the signal are not the pri-
mary factor that underlies the differences observed in performance in this same-
different discrimination task. Rather, performance is influenced by differences in
controlled attention reflecting the ability to actively use cognitive control strate-
gies to encode, maintain, monitor and manipulate representations of the talkers
voice in working memory and inhibit responses to the meaning of the two sen-
tences (see also Cleary, Pisoni & Kirk 2005).
David B. Pisoni
Cognitive control and self-regulation
To investigate executive functioning and cognitive control processes in deaf chil-

dren with cochlear implants, we have also carried out several studies using a par-
ent report behavior rating inventory known as the BRIEF (Gioia, Isquith, Guy &
Kenworthy 2000; Pisoni et al. 2008). The BRIEF is filled out by parents, teachers or
caregivers to assess a childs executive functioning and self-regulation skills in ev-
eryday real-world environments. The rating inventory has been shown to be useful
in evaluating children with a wide spectrum of developmental and acquired cog-
nitive conditions (Gioia et al. 2000). The BRIEF contains nine clinical rating scales
designed to assess inhibition, shifting of attention, emotional control, working
memory, planning and organization, among others. Scores from these nine scales
are then used to construct three indices: a behavioral regulation index (BRI), a
metacognition index (MI), and a global executive composite score (GEC).
Our analyses of data obtained from deaf children with CIs revealed elevated
scores on all three composite indices of the BRIEF. Although none of the scores fell
within the clinically significant range, the three composite scores were all signifi-
cantly higher for the group of deaf children with CIs than a comparison group of
age-matched normal-hearing children (Pisoni et al. 2008; Pisoni, Conway,
Kronenberger, Henning & Anaya 2010). An analysis of the BRIEF clinical scales
revealed that children with CIs showed elevated scores on shifting, emotional con-
trol, initiation, working memory, planning and organization, and organization of
materials compared to the normal hearing children.
These parent rating data of everyday real-world behaviors using the BRIEF
suggest that a period of profound deafness and associated language delay before
cochlear implantation not only affect basic cognitive information processing
skills such as working memory, planning and organization, but they also affect
the development of self-regulation and emotional control skills, processes not
typically considered to be comorbid with deafness and sensory deprivation. The
scores on the BRIEF provide additional converging evidence for our general hy-
pothesis that multiple information processing systems are linked together in de-
velopment and that disturbances resulting from deafness and language delays are
not domain-specific restricted only to hearing, speech perception and processing
spoken language.
Theoretical and clinical implications
Our recent findings suggest that many deaf children with CIs also have comorbid
disturbances and delays in several basic elementary cognitive processes that are
inseparable components of the core information processing system used in

speech perception and spoken language processing, and these disturbances are
secondary to their profound hearing loss and delay in language. A period of au-
ditory deprivation during early critical developmental periods before implanta-
tion followed by exposure to degraded and underspecified acoustic information
from a CI affects cognitive development in a variety of ways. Differences result-
ing from both deafness and subsequent neural reorganization of multiple brain
systems may be responsible for the enormous variability observed in speech and
language outcome measures following implantation. Without knowing what spe-
cific underlying biological and neurocognitive factors are responsible for the
enormous individual differences in outcomes, it is difficult to develop appropri-
ate and efficacious approaches to habilitation and therapy after a child receives a
cochlear implant.
More importantly, deaf children who perform poorly after receiving a CI are
an extremely heterogeneous group and they differ from each other in numerous
ways, reflecting dysfunction of multiple brain systems associated with congenital
deafness, a period of profound hearing loss, language delays as well as impover-
ished and sparsely-coded acoustic inputs from the CI. In addition, it seems very
unlikely that an individual child will be able to achieve optimal speech and lan-
guage benefits from his/her CI without researchers and clinicians having a better
understanding of precisely why an individual child is having delays and distur-
bances and which particular cognitive domains and information processing sys-
tems underlie these specific problems.
Early knowledge of which children may be at high risk for poor outcomes fol-
lowing implantation is critical for medical decisions about habilitation, interven-
tion and assessment of alternative treatments for deafness and hearing impairment.
Understanding the reasons for the enormous variability in outcome and benefit
following cochlear implantation provides new knowledge and information for
surgeons, audiologists and speech-language pathologists who are concerned with
clinical problems in deaf children, as well as neuroscientists and biomedical engi-
neers who are currently developing the next generation of CIs and speech process-
ing strategies.
Research on CIs also has important implications for research and theory in
cognitive and linguistic development. Above and beyond the pressing immediate
clinical issues of helping deaf children learn to speak and listen like normal-hearing
typically-developing children and reach their full intellectual potential, deaf chil-
dren with CIs also serve as a unique model system to study the effects of brain
plasticity, experience- and activity-dependent learning and neural development in
speech perception and language development.
David B. Pisoni
Acknowledgements
Preparation of this chapter was supported by 2 R01 grants: NIDCD 5R01DC000111

and 1R01DC009581. I thank Luis Hernandez for his help and support and Terren
Green for her editorial work.
References
Baddeley, A.D., Thomson, N., & Buchanan, M. 1975. Word length and the structure of short
term memory. Journal of Verbal Learning and Verbal Behavior 14: 575589.
Barkley, R.A. 1997. ADHD and the Nature of Self-control. New York NY: Guilford.
Barkley, R.A. 2006. Attention-Deficit/Hyperactivity Disorder: A Handbook for Diagnosis and
Treatment, 3rd edn. New York NY: Guilford.
Bergeson, T. & Pisoni, D.B. 2004. Audiovisual speech perception in deaf adults and children fol-
lowing cochlear implantation. In Handbook of Multisensory Integration, G. Calvert, C.
Spence & B.E. Stein (eds), 749772. Cambridge MA: The MIT Press.
Blair, C. & Razza-Peters, R. (2007). Relating effortful control, executive function, and false belief
understanding to the emerging math and literacy ability in kindergarten. Child Develop-
ment 78, 647663.
Burkholder, R.A. & Pisoni, D.B. (2003). Speech timing and working memory in profoundly deaf
children after cochlear implantation. Journal of Experimental Child Psychology 85: 6388.
Carter, A.K., Dillon, C.M. & Pisoni, D.B. 2002. Imitation of nonwords by hearing impaired chil-
dren with cochlear implants: Suprasegmental analyses. Clinical Linguistics & Phonetics 16:
619638.
Casserly, E.D., & Pisoni, D.B. (2013). Nonword repetition as a predictor of long-term speech and
language skills in children with cochlear implants. Otology & Neurotology 34(3): 460470.
Cicchetti, D. & Curtis, W.J. 2006. The developing brain and neural plasticity: Implications for
normality, psychopathology, and resilience. In Developmental Psychopathology: Develop-
mental Neuroscience, Vol. 2, 2nd edn, D. Cicchetti & D. Cohen (eds). New York NY: Wiley.
Clark, A. 1997. Being There: Putting Brain, Body, and World Together Again. Cambridge MA: The
MIT Press.
Clark, G.M. 2003. Speech (sound) processing. In Cochlear Implants: Fundamentals and Applica-
tions, 421432. New York NY: Springer.
Cleary, M. & Pisoni, D.B. 2002. Talker discrimination by prelingually-deaf children with co-
chlear implants: Preliminary results. Annals of Otology, Rhinology, & Laryngology Supple-
ment-Proceedings of the 8th Symposium on Cochlear Implants in Children 111: 113118.
Cleary, M., Dillon, C.M. & Pisoni, D.B. 2002. Imitation of nonwords by deaf children after co-
chlear implantation: Preliminary findings. Annals of Otology, Rhinology, & Laryngology
Supplement-Proceedings of the 8th Symposium on Cochlear Implants in Children 111: 9196.
Cleary, M., Pisoni, D.B. & Kirk, K.I. 2005. Influence of voice similarity on talker discrimination
in normal-hearing children and hearing-impaired children with cochlear implants. Journal
of Speech, Language, and Hearing Research 48: 204223.
Dillon, C.M., Burkholder, R.A., Cleary, M. & Pisoni, D.B. 2004. Nonword repetition by children
with cochlear implants: Accuracy ratings from normal-hearing listeners. Journal of Speech,
Language, and Hearing Research 47: 11031116.
Dillon, C.M., Cleary, M., Pisoni, D.B. & Carter, A.K. 2004. Imitation of nonwords by hearing-
impaired children with cochlear implants: Segmental analyses. Clinical Linguistics and Pho-
netics 18: 3955.
Gathercole, S. & Baddeley, A. 1990. Phonological memory deficits in language disordered chil-
dren: Is there a causal connection? Journal of Memory and Language 29: 336360.
Gathercole, S.E., Hitch, G.J., Service, E. & Martin, A.J. 1997. Phonological short-term memory
and new word learning in children. Developmental Psychology 33: 966979.
Geers, A.E & Moog, J.S. 1987. Predicting spoken language acquisition in profoundly deaf chil-
dren. Journal of Speech and Hearing Disorders 52: 8494.
Geers, A., Brenner, C., & Davidson, L. 2003. Factors associated with development of speech
perception skills in children implanted by age five. Ear and Hearing 24: 24S35S.
Geers, A., Nicholas, J., & Sedey, A. 2003. Language skills of children with early cochlear implan-
tation. Ear and Hearing 24: 46S58S.
Geers, A., Brenner, C., & Tobey, E.A. 2011. Long-term outcomes of cochlear implantation in
early childhood; Sample characteristics and data collection methods. Ear & Hearing 32:
2S12S.
Gioia, G.A., Isquith, P.K., Guy, S.C. & Kenworthy, L. 2000. BRIEF: Behavior Rating Inventory of
Executive Function Professional Manual. Lutz FL: Psychological Assessment Resources.
Goldman-Rakic, P.S. 1988. Topography of cognition: Parallel distributed networks in primate
association cortex. Annual Reviews of Neuroscience 11: 137156.
Horn, D.L., Davis, R.A.O., Pisoni, D.B. & Miyamoto, R.T. 2004. Visuomotor integration ability
of pre-lingually deaf children predicts audiological outcome with a cochlear implant: A first
report. International Congress Series 1273: 356359.
Horn, D., Pisoni, D., Sanders, M. & Miyamoto, R. 2005. Behavioral assessment of pre-lingually
deaf children prior to cochlear implantation. Laryngoscope 115: 16031611.
Hulme, C., Thomson, N., Muir, C. & Lawrence, A. 1984. Speech rate and the development of
short-term memory span. Journal of Experimental Child Psychology 38: 241253.
Isquith, P.K., Crawford, J.S., Espy, K.A., & Gioia, G.A. 2005. Assessment of executive function in
preschool-aged children. Mental Retardation and Developmental Disabilities Research Re-
views 11: 209215.
Kirk, K.I., Pisoni, D.B., & Miyamoto, R.T. 2000. Lexical discrimination by children with co-
chlear implants: Effects of age at implantation and communication mode. In Cochlear Im-
plants, S.B. Waltzman & N.L. Cohen (eds), 252254. New York NY: Thieme.
Kirk, K.I. 2000. Challenges in the clinical investigation of cochlear implant outcomes. In Co-
chlear Implants. J.K. Niparko et al. (eds), 225259. Philadelphia PA: Lippincott, Williams &
Wilkins.
Konishi, M. 1985. Birdsong: From behavior to neuron. Annual Review of Neuroscience 8: 125170.
Kronenberger, W.G., Pisoni, D.B., Henning, S.C. & Colson, B.G. 2013. Executive functioning
skills in long-term users of cochlear implants: A case control study. Journal of Pediatric
Psychology. (Advance online publication).
Luria, A.R. 1973. The Working Brain. Harmondsworth: Penguin.
Marler, P. & Peters, S. 1988. Sensitive periods for song acquisition from tape recordings and live
tutors in the swamp sparrow. Melospiza georgiana. Ethology 77: 7684.
McNab, F. & Klingberg, T. 2008. Prefrontal cortex and basal ganglia control access to working
memory. Nature Neuroscience 11: 103107.
David B. Pisoni
Nauta, W.J.H. 1964. Discussion of Retardation and facilitation in learning by stimulation of

frontal cortex in monkeys. In The Frontal Granular Cortex and Behavior, J.M. Warren & K.
Akert (eds), 125. New York NY: McGraw-Hill.
Niparko, J.K., Tobey, E.A., Thal, D.J., Eisenberg, L.S., Wang, N.Y., Quittner, L. & Fink, N.E.,
CDaCI Investigative Team. 2010. Spoken language development in children following co-
chlear implantation. The Journal of the American Medical Association 303(15): 1498 1506.
Oberauer, K., Sub, H.M., Wilhelm, O. & Wittmann, W.W. 2003. The multiple faces of working
memory: Storage, processing, supervision, and coordination. Intelligence 31: 167193.
Oberauer, K. 2005. Binding and inhibition in working memory: Individual and age differences
in short-term recognition. Journal of Experimental Psychology: General 134: 368387.
Pisoni, D.B. (1997). Looking at the Stars: A first report on the intercorrelations among mea-
sures of speech perception, intelligibility, and language development in pediatric cochlear
implant users. Vth International Cochlear Implant Conference, New York, May 13.
Pisoni, D.B., Cleary, M., Geers, A.E. & Tobey, E.A. 2000. Individual differences in effectiveness
of cochlear implants in prelingually deaf children: Some new process measures of perfor-
mance. Volta Review 101: 111164.
Pisoni, D.B. 2000. Cognitive factors and cochlear implants: Some thoughts on perception, learn-
ing, and memory in speech perception. Ear and Hearing 21: 7078.
Pisoni, D.B. & Cleary, M. 2003. Measures of working memory span and verbal rehearsal speed
in deaf children after cochlear implantation. Ear and Hearing 24: 106S120S.
Pisoni, D.B. & Cleary, M. 2004. Learning, memory, and cognitive processes in deaf children fol-
lowing cochlear implantation. In Springer Handbook of Auditory Research: Auditory Prosthe-
sis, SHAR Volume X, F.G. Zeng, A.N. Popper & R.R. Fay (eds), 377426. Berlin: Springer.
Pisoni, D.B. 2005. Speech perception in deaf children with cochlear implants. In Handbook of
Speech Perception, D.B. Pisoni & R.E. Remez (eds), 494523. Oxford: Blackwell.
Pisoni, D.B., Conway, C.M., Kronenberger, W.G., Horn, D.L., Karpicke, J. & Henning, S.C. 2008.
Efficacy and effectiveness of cochlear implants in deaf children. In Deaf Cognition: Founda-
tions and Outcomes, M. Marschark & P. Hauser (eds), 52101. Oxford: OUP.
Pisoni, D.B., Conway, C.M., Kronenberger, W.G., Henning, S.C. & Anaya, E. 2010. Executive
function, cognitive control and sequence learning in deaf children with cochlear implants.
In Oxford Handbook of Deaf Studies, Language and Education, M. Marschark & P. Spencer
(eds). Oxford: OUP.
Pisoni, D.B., Kronenberger, W.G., Roman, A.S. & Geers, A.E. 2011. Measures of digit span and
verbal rehearsal speed in deaf children after more than 10 years of cochlear implantation.
Ear and Hearing 32: 60S74S.
Sharma, A., Dorman, M.F. & Spahr, A.J. 2002. A sensitive period for the development of the
central auditory system in children with cochlear implants: Implications for age of implan-
tation. Ear and Hearing 23: 532539.
Sporns, O. 1998. Biological variability and brain function. In Consciousness and Human Identity,
J. Cornwell (ed.), 3856. Oxford: OUP.
Sporns, O. 2003. Network analysis, complexity, and brain function. Complexity 8: 5660.
Svirsky, M.A., Teoh, S.W. & Neurburger, H. 2004. Development of language and speech percep-
tion in congenitally, profoundly deaf children as a function of age at cochlear implantation.
Audiology & Neuro-Otology 9: 224233.
Ullman, M.T. & Pierpont, E.I. 2005. Specific language impairment is not specific to language:
The procedural deficit hypothesis. Cortex 41: 399433.
section 3
Cross-linguistic approaches to phonological

acquisition
While much of Dinnsens research has characterized individual childrens phonol-

ogies, he has also shown interest in cross-linguistic typologies and implicational
relationships among phonological elements. The goal of this area of inquiry is to
find and account for commonalities across languages (or across individuals), for
instance, a language that contains affricates also contains fricatives. Dinnsens work
recognizes the fundamental importance of cross-linguistic universals to the study
of development: if it is true that all languages that have affricates also have frica-
tives, then teaching affricates to a child who produces neither type of sound should
also induce learning of fricatives at no cost. Research by Dinnsen and colleagues
has supported this type of prediction (Dinnsen, Chin, Elbert & Powell, 1990;
Dinnsen, Elbert & Weismer, 1980; Dinnsen, Gierut, Morrisette, Green & Farris-
Trimble, 2011; Dinnsen & OConnor, 2001a,b). Predictions of this sort have a
broader impact on our understanding of the structure of language as a whole. By
studying different children, different dialects, or different languages, Dinnsen has
been able to discover broader generalizations about how phonology is acquired
and organized, and to characterize differences among sound systems, both
developing and fully-developed (Dinnsen, 1977; Dinnsen & Charles-Luce, 1984;
Dinnsen & Chin, 1993; Dinnsen & Eckman, 1975; Dinnsen, Elbert & Weismer,
1980; Dinnsen & OConnor, 2001a,b).
Section 3 includes three papers that use a cross-linguistic approach in the
study of phonological acquisition.
Yava takes a typological approach to the development of s-clusters, which are
particularly problematic for most phonological theories (What guides children's
acquisition of #sC clusters? A cross-linguistic account). His findings suggest that
while certain patterns of accurate s-cluster acquisition seem universal, the inac-
curate productions (as might be displayed by children with phonological disor-
ders) are more variable and language-specific.
Barlow and Pruitt-Lord examine the phonology/morphology interaction in two
dialects of English; because these dialects differ in their morphology, the authors hy-
pothesize that these morphological differences also play a role in childrens phono-
logical development (The role of phonological context in childrens overt marking of
-s in two dialects of American English). They find that dialectal and phonological
factors interact in determining overt plural and third-personal singular markings.
Keel discusses the relationship between phonology and morphology in the ac-
quisition and attrition of minority dialects, drawing comparisons between the de-
velopment of a minority dialect and its parent language (German settlement
varieties in Kansas: Some unusual phonological and morphological developments
with the approach of language death). While this chapter is probably the furthest
in scope from Dinnsens own research, it illustrates both the broad timescale of
development (development and change of a dialect, as opposed to development
and change of a single childs phonological system) and the importance of examin-
ing non-standard phonologies (be it minority dialect or disordered phonology).
Each of these three papers emphasizes the importance of looking across lan-
guages or dialects to find fundamental truths about how phonological systems are
acquired and change over time.
References
Dinnsen, D.A. 1977. A functional explanation of dialect difference. Language Sciences 46: 17.
Dinnsen, D.A. & Charles-Luce, J. 1984. Phonological neutralization, phonetic implementation
and individual differences. Journal of Phonetics 12: 4960.
Dinnsen, D.A., & Chin, S.B. 1993. Individual differences in phonological disorders and implica-
tions for a theory of acquisition. In F.R. Eckman (Ed.), Confluence: Linguistics, L2 acquisi-
tion and speech pathology (pp. 139154). Amsterdam: John Benjamins.
Research 33: 2837.
Dinnsen, D.A., & Eckman, F.R. 1975. A functional explanation of some phonological typologies.
In R.E. Grossman, L.J. San & T.J. Vance (Eds.), Functionalism (pp. 126134). Chicago:
Chicago Linguistic Society.
Dinnsen, D.A., Elbert, M., & Weismer, G. 1980. Some typological properties of functional mis-
articulation systems. In W. O. Dressler (Ed.), Phonological 1980 (pp. 123128). Innsbruck:
Innsbrucker Beitrge zur Sprachwissenschaft.
Dinnsen, D.A., Gierut, J.A., Morrisette, M.L., Green C.R. & Farris-Trimble, A.W. 2011. On the
interaction of deaffrication and consonant harmony. Journal of Child Language 38: 380403.
Dinnsen, D.A., & OConnor, K.M. 2001a. Typological predictions in developmental phonology.
Dinnsen, D.A., & OConnor, K.M. 2001b. Implicationally-related error patterns and the selec-
tion of treatment targets. Language, Speech and Hearing Services in Schools 32: 257270.
What guides childrens acquisition
of #sC clusters?
A cross-linguistic account
Mehmet Yava
Florida International University
This chapter is an overview of the cross-linguistic patterns in the acquisition of

two-member s-clusters. Results of several studies on three Germanic languages
(English, Dutch, and Norwegian) and three non-Germanic languages (Hebrew,
Croatian, and Polish) were evaluated. Productions from typically developing
children as well as children with phonological disorders were looked at with
respect to cluster reductions and correct productions. While the examination of
the former revealed a common behavior among all six languages, the latter point
to language-specific effects.
Introduction
In any discussion on the acquisition of consonant clusters, sonority plays an im-

portant role because the Sonority Sequencing Principle (hereafter SSP) states that
a segment constituting a sonority peak is preceded and/or followed by a sequence
of segments with progressively decreasing sonority values. Therefore, the SSP is a
significant force in languages and has been invoked in explaining various phe-
nomena related to syllable structure. Among these are syllable structure (Hooper
1976; Prince & Smolensky 2004; Selkirk 1984; Smolensky 2006; Steriade 1990;
Venneman 1972), syllable contact (Gouskova 2001, 2004), Syllable markedness
and misperceptions (Berent, Lennertz & Smolensky 2011), stress assignment
(de Lacy 2007), and reduplication (Morelli 1999; Parker 2002; Pinker & Birdsong
1979; Steriade 1990). Sonority of consonants is also correlated with production
accuracy in cluster acquisition in typical first language phonological acquisition
(Barlow 2005; Ohala 1999; Pater 2004; Yava et al. 2008), second language phonol-
ogy (Broselow, Chen & Wang 1998; Broselow & Finer 1991; Broselow & Xu 2004;
Yava 2011), developmental phonological disorders (Barlow 2001; Gierut 1999;
Mehmet Yava
Yava 2010), aphasia (Romani & Calabrese 1998; Stenneken, Bastiaanse, Huber &
Jacobs 2005), and speech errors (Stemberger & Treiman 1986).
Several sonority scales have been proposed by scholars. The proposals vary in
details but all share the same basic ordering: vowels are placed at the higher end of
sonority, obstruents are at the bottom, and sonorant consonants in the middle. In
this chapter, we adopt the 10 point scale proposed by Hogg and McCully (1987),
given below.
SI (Sonority index)
Low vowels 10
Mid vowels 9
High vowels 8
Flaps 7
Laterals 6
Nasals 5
Vd. fricatives 4
Vs. fricatives 3
Vd. stops 2
Vs. stops 1
It is commonly agreed on that the sonority difference between the cluster mem-
bers translates into their relative complexity (markedness). Firstly, clusters that
have falling (negative) sonority (e.g. Hebrew /taim/ two (fm.) from 3 to 1 in so-
nority) are more marked than level sonority clusters (e.g. Hebrew /bgadim/ clothes
leveled at 2 in sonority) which, in turn, are more marked than rising sonority clus-
ters (e.g. blue [blu], rising from 2 to 6 in sonority). The existence of a more marked
cluster in a language implies the existence of a less marked cluster. Thus, if a lan-
guage allows negative sonority clusters, it also allows level sonority and rising so-
nority clusters. Also, in rising sonority canonical clusters, the ones that have a
smaller sonority distance are more complex (marked) than those with a greater
difference. Accordingly, /fl/ (e.g. fly) which has a sonority distance of 3 between its
members is more marked than /pj/ (e.g. pure) which has a greater distance from
C1 to C2. This is related to the minimal sonority distance principle. For example,
in Spanish, an onset such as /pn/ is ungrammatical while /pl/ is acceptable since
laterals are higher in sonority than nasals (Harris 1983). In English, and some
other Germanic languages such as Dutch and Norwegian, some #sC clusters (/sp/,
/st/, sk/) are the only combinations that do not follow the SSP in that the sonority
falls rather than rises from the first member to the second. There are other pecu-
liarities of #sC clusters. For example, English and Dutch violate the principle that
prohibits homorganic clusters (/pw/, /bw/ are disallowed, but /st/, /sn/, /sl/ are
not), and violate the generalization that disallows obstruent +obstruent clusters
Cross-linguistic acquisition of #sC clusters
true cluster (complex onset) adjust cluster

Onset nucleus coda Onset nucleus coda
X X X X X X X
b l i k s p i k
Figure 1. Metrical trees illustrating true and adjunct clusters
(/kt/, /fp/ are disallowed, but /sp/ /sk/ are not). In addition, in English, obstruent
+ nasal clusters are permitted only in #sC clusters (e.g. /sm/, /sn/). Such pecu-
liarities have led some scholars to suggest a special adjunct status for #sC clusters.
(Davis 1990; Fikkert 1994; Giegerich 1992; Kenstowicz 1994, Trommelen 1984).
Under this view, the /s/ in #sC clusters is not syllabified directly under the onset
position, but is a direct dependent of the syllable. Thus, we have two categories of
cluster types: true clusters (complex onsets), and adjunct clusters, as shown in
Figure 1.
There are, however, divergent views on which s-clusters should be treated as
adjuncts. For English, in general, only the negative sonority sequences (i.e. /s+stop/)
are considered to have the adjunct status. However, in some studies (Gierut 1999;
Yava & Someillan 2005) /s+nasal/ sequences evoke the adjunct status too. There
is indeterminacy in other languages too. In Dutch (Fikkert 1994) and German
(Hall 1992), only /s+stop/ sequences are treated as adjuncts, while in Italian (Davis
1990), /s+stop/, /s+nasal/ and /sl/ are treated similarly. In addition to their inde-
terminacies in developmental data, s+sonorant C clusters create further ambi-
guities. For example, the canonical obstruent + sonorant C clusters rise in
sonority, but like /s + obstruent/ clusters, some in this group (e.g. /sn/) do not re-
spect the place constraints holding of /obstruent + sonorant/ clusters. As a result,
it seems that we have to deal with the following 3 types of clusters: a) canonical
(obst. + son) clusters, b) /s/ + stop/ clusters, and c) /s/+son/ clusters. Cluster re-
pair patterns observed in different language contact situations (L2 acquisition and
loan word adaptation) have the potential to answer the question regarding the
status of /s+sonorant/ clusters. However, as Figure 2 shows, through their behav-
ior, it is not possible to come to any non-controversial grouping.
Japanese (Lovins, 1974) and Turkish (Yava 2003) employ epenthesis (e.g. star
[star] [star], train [tren] [tiren]), and Iraqi Arabic (Broselow, 1992) employs
Mehmet Yava
Can. Cl. /s/+son. C. /s/+stop

Japanese, Turkish E E E
(Lovins 1974), (Yava_ 2003)
Iraq. Arabic P P P
(Broselow 1992)
Persian E P P
(Karimi 1987)
Sindhi, Bengali, E E P
Egyp. Arabic
(Broselow 1992)
E= Epenthesis
P= Prothesis
Figure 2. Cluster repair patterns in various languages (the preferences are shown in boxes)
prothesis for all three types of clusters (e.g. floor [flr] [iflor], snow [sno]
[isno], study [stdi] [istadi]). The situation is different in other languages. While
in Persian (Karimi 1987), epenthesis is restricted to canonical clusters, and
s-clusters of any kind receive prothesis (e.g. drink [drnk] [dirink], snow
[sno] [esno], struggle [strgl] [estragl]), in Egyptian Arabic, Bengali, and
Sindhi (Broselow 1992) s+stop clusters are singled out for prothesis and the re-
maining clusters show epenthesis (e.g. Egyptian Arabic: floor [flr] [filor], slide
[slad] [silad], study [stdi] [istadi]). All the above do not lead us to a con-
sensus about how /s+sonorant C/ clusters should be interpreted.
As one can see from all the above, there are a variety of unresolved issues sur-
round #sC clusters. Cross-linguistic investigations in acquisition of different types
of #sC can help elucidate the issues and can help the extent to which sonority is a
factor in the explanations. With this in mind, in the following, we are going to have
a cross-linguistic look at the patterns emerging in the acquisition of #sC clusters.
Results of various studies done with data from typically developing children
(hereafter TD) and children with phonological disorders (hereafter PD) coming
from three Germanic languages (English, Dutch, Norwegian), and three non-
Germanic languages (Hebrew, Croatian, and Polish) will be examined. The moti-
vation for the inclusion of data from T.D. and P.D. groups is to see if the same
constraints are in operation in both populations or if there may be differences. We
are going to try to answer the following questions: Do production patterns of
#sC clusters depend on the sonority restrictions? More specifically, in correct
productions, do children show greater accuracy for the unmarked targets than the
marked ones? Also, in the erroneous production of a specific nature, namely the
cluster reductions, do we see any hierarchical patterns supported by sonority con-
straints? If the answers to these questions are negative, then what other factor(s)
can explain the patterns observed? Do the results point at a common behavior
among languages or suggest language-specific explanations?
Participants and procedures
Before we present the results, a few words regarding the subjects and procedures
will be helpful. The data examined here are culled from nine separate studies of the
languages in question. The participants in the typically developing group were the
following in numbers, genders, and age groups in the respective languages. English
(Yava & Core 2006): 40 (22 girls and 18 boys), between the ages 2;5 and 4;2, mean
age 3;1; Hebrew (Ben-David 2006): 40 (20 boys and 20 girls), between the ages
1;10 and 3;0, mean age 2;6; Dutch (Gerrits & Zumach 2006): 45 (22 girls and 23
boys) between the ages 2;2 and 3;6, mean age 3;0; Norwegian (Kristoffersen &
Simonsen 2006): 27 (12 girls and 15 boys) between the ages 1;9 and 3;0, mean age:
2;6; Croatian (Mildner & Tomic 2010): 41 (20 boys and 21 girls) between the ages
2;4 and 4;1, mean age: 3;5; Polish (Yava & Marecka in press): 25 (13 boys and 12
girls), between the ages 2;9 and 4;3, mean age: 3;5. The subjects were chosen from
the participants in these studies who had the mastery of single onset /s/, the sec-
ond members of the clusters in question in singleton mode, and showed some
signs of acquisition of #sC clusters without having completed the acquisition of all
#sC clusters. This criterion excluded children who produced all #sC clusters cor-
rectly and those who produced all targets incorrectly. All of the individual studies
involved a picture-naming task, appropriate to individual languages. Each childs
productions were recorded and transcribed by individuals trained in the use of
International Phonetic Alphabet. Reliability of the transcriptions was achieved via
consensus between independent judges. All attempts at the production of words
that in the target language involve #sC clusters were analyzed for accuracy of pro-
duction as well as the reductions with respect to potential differences across sub-
types of #sC groups.
In children with phonological disorders, the above criteria and the procedures
were also followed. In addition, these participants passed an audiological screen-
ing test, and none of them showed signs of any oral motor anomalies. The distribu-
tion of numbers and genders was the following. English (Yava & McLeod 2010):
30 (20 boys and 10 girls) between the ages 3;7 and 4;10, with a mean age of 4;1;
Dutch (Gerrits 2010): 24 (3 girls and 21 boys) between the ages 3;0 and 5;8, with a
Mehmet Yava
mean age of 4;3; Hebrew (Ben-David, Ezrati & Stuhlman 2010): 30 (18 boys and
12 girls), between the ages of 3;5 and 5;2, with a mean age of 4;4; Croatian (Mildner
& Tomic 2010): 30 (22 boys and 8 girls) between the ages 3;8 and 7;0, with a mean
age of 5;4; Polish (Yava & Marecka in press): 24 (20 boys and 4 girls) between the
ages 3;2 and 6:10, with a mean age of 4;9.
Correct renditions
We will start our exposition with the correct renditions of the targets in the six
languages mentioned above. If, as mentioned above, sonority distance between C1
and C2 is an indicator of markedness, then we can expect the following order in
the correct rendition: in descending order /sw/, /sl/, /sN/, and /sT/. First, we look
at the three Germanic languages (English, Dutch, and Norwegian). The results are
shown in Figure 3. (The y-axis shows percent correct.)
A brief look at the results from English (T.D. as well as P.D), shows that the
sonority distance between C1 and C2 is an influential factor in that the percentage
of accurate renditions increases as we go from more marked to less marked targets.
However, when we look at the results of Dutch and Norwegian, it is not possible to
say the same thing, as the percentage of correct renditions is lower in /sN/ than
/sT/ targets, despite the fact that the former displays a rising sonority as opposed
to the latters negative sonority. Specifically, in Dutch TD children, the percentage
goes down from /sT/ (65%) to /sN/ 62%; similarly, we see a drop from 33% to 25%
in PD children. The situation in Norwegian is also parallel to that of Dutch. While
the correct renditions for /sT/ reach 75%, they are lower, 69%, for /sN/ targets. In
all three languages, /s+lateral/ and /s+glide/ (i.e., /s+approximant) targets have
expectedly higher accurate responses. Although sonority based markedness can-
not explain the results because of the discrepancy created by /s+nasal/ clusters, an
alternative dimension based on the continuancy of C2 seems to account for the
patterns prevalent in the three languages, as shown in Figure 4 below.
The correct renditions are always higher in /s/+[+continuant] (i.e. /sl/, /sw/)
targets than in /s/+ [continuant] (i.e. /sT/, /sN/) targets both in TD and PD
children. In English, the jump is 17 points in TD, and a whopping 43 points in PD
children. In Dutch, it is 18 points in TD and 21 points in PD. In Norwegian TD,
the jump is a comparable 18 points.
Turning to the 3 non-Germanic languages (Hebrew, Croatian, and Polish),
however, we see a very different picture. As shown in Figure 5, the production ac-
curacy does not increase via sonority difference between C1 and C2.
The results dont change when we redesign the tables based on the continu-
ancy of C2. Quite the contrary, /s/ + [-continuant] targets enjoy better accuracy
than the others, as shown in Figure 6.
English (T.D.) English (P.D.)

40 children; mean age 3:1 30 children; mean age 4:1
100 100 90
90 90 83
80 76 80
70 70
56 58 59
60 51 60
50 50
40 40 31
30 30
20 20
10 10
0 0
sT sN sl sw sT sN sl sw
(Yava_ and Core 2006) (Yava_ and McLeod 2010)
Dutch* (T.D.) Dutch* (P.D.)

100 100
90 81 90
80 80
65
70 62 70
60 60
49
50 50
33
40 40
25
30 30
20 20
10 10
0 0
sT sN sl sT sN sl
*no/sw/ *no/sw/
(Gerrits & Zumach 2006) (Gerrits 2010)
Norwegian* (T.D.)
27 children; mean age 2:6
100
89
90 82
75
80
69
70
60
50
40
30
20
10
0
sT sN sl s
*no/sw/
(Kristoffersen & Simonsen 2006)
Figure 3. Percentages of correct productions in different #sC targets in Germanic languages

Mehmet Yava
100 100 100

86
90 90 90 81
80 80 80
67
70 70 70 63
60 60 60
50
50 50 43 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
/s/ + [cont] /s/ + [+cont] /s/ + [cont.] /s/ + [+cont.] /s/ + [cont] /s/ + [+cont]
English T.D. English P.D. Dutch T.D.
100 100
90 90 81
80 80
70 70 63
60 60
49
50 50
40 40
28
30 30
20 20
10 10
0 0
/s/ + [cont.] /s/ + [+cont.] /s/ + [cont] /s/ + [+cont]
Dutch P.D. Norwegian T.D.
Figure 4. Percentages of correct productions based on [ continuant] nature of C2

in Germanic languages
It is a legitimate question to ask why the two groups of languages show such a
drastic difference. The answer may lie in the frequency effects. The two-member
initial cluster inventories of the three Germanic languages English, Dutch, and
Norwegian (Appendix A) show that they have very few clusters that do not follow
the SSP; in fact, only their /s+stop/ clusters have negative sonority, in addition to
one cluster with level sonority, /sx/, in Dutch. The three non-Germanic languages
(Appendix B), in sharp contrast, have rather rich cluster inventories with more
lenient attitudes towards Sonority Sequencing. Not only do they have a multitude
of negative sonority and level sonority clusters, they also have several obstruents
as C1 that are followed by a nasal [-continuant] as C2. The children that are raised
in the ambiance of such languages would probably produce the marked clusters
more frequently (i.e. neural pathways involved in these productions are strength-
ened and stabilized as a result of continuous activation) (Beckman & Edwards
1999; Zamuner, Gerken & Hammond 2004).
Hebrew* (T.D.) Hebrew* (P.D.)

100 100
90 90
80 80
70 70
60 54 60
52
46 48
50 50
37
40 40
30
30 30 22
16
20 20
10 10
0 0
sT sN sl sr sT sN sl sr
*no/sw/
(Ben-David 2006) (Ben-David, Ezrati & Stulman 2010)
Croatian* (T.D.) Croatian* (P.D.)

100 100
90 90
80 71
80 73
66 69
70 70
60 56 61 59
60 60
50 50
40 40
30 30
20 20
10 10
0 0
sT sN sl s sT sN sl s
*no/sw/ *no/sw/
(Mildner and Tomi 2010)
Polish** (T.D.) Polish** (P.D.)

100 100
90 90
80 80
70 70
60 60
50 50
40 36 40
34
30 30
20 16 18
20 20 11
10 10
0 0
sT sN sw sT sN sw
**no/sl/ *no/sl/
(Yava & Marecka, in press)
Figure 5. Percentages of correct productions in different #sC targets in non-Germanic

Languages
Mehmet Yava
100 100 100

90 90 90
80 80 80
68
70 70 70
58
60 60 53 60
47
50 50 50
40 33 40 40
30 30 30
19
20 20 20
10 10 10
0 0 0
/s/ + [cont.] /s/ + [+cont.] /s/ + [cont] /s/ + [+cont] /s/ + [cont] /s/ + [+cont]
Hebrew T.D. Hebrew P.D. Croatian T.D.
100 100 100

90 90 90
80 71
80 80
70 70 70
60
60 60 60
50 50 50
35
40 40 40
30 30 30
20 17
20 20 20 11
10 10 10
0 0 0
/s/ + [cont.] /s/ + [+cont.] /s/ + [cont] /s/ + [+cont] /s/ + [cont.] /s/ + [+cont.]
Croatian P.D. Polish T.D. Polish P.D.
Figure 6. Percentages of correct productions based on [continuant] nature of C2

in non-Germanic languages
Patterns in reductions
When children do not produce consonant clusters accurately, several different ren-
ditions are possible. Although epenthesis (e.g. blood [bld] [bld]) and coales-
cence (e.g. swim [swm] [fm], whereby the labiality of /w/ and the frication of
/s/ are combined) are found, the reduction to one member is by far the most com-
mon process. In the reductions of canonical /obstruent + approximant/ clusters,
the situation is generally quite predictable via sonority. Generally, the more sono-
rous member of the cluster is deleted so as to leave the remaining structure to
provide a higher jump in sonority from the retained least sonorous member to the
high sonority nucleus. This is in accordance with the Sonority Dispersion Princi-
ple, Clements (1990). For example, the realizations creme [krim] [kim], blue
[blu] [bu] are commonplace, whereas the alternatives [rim] and [lu], whereby
the least sonorous member of the cluster eliminated, are not found in typical de-
velopment. This formula of retaining the lower sonority and eliminating the high-
er sonority member also works rather well for /s+stop/ clusters (e.g. spider [spad]
[pad] not [sad], skate [sket] [ket] not [set]). When we look at /s+sonorant
C/ targets, however, we see a different picture. While the overall tendency in the
reduction of /s+approximant/ (i.e. /sl/, /sw/) clusters is to retain the lower sonority
member, /s/, /s+nasal/ clusters generally do not follow this route.
There have been two major attempts in literature related to the reduction in #sC
clusters. Very briefly stated, Factorial Typology advanced by Pater and Barlow (2003)
predicts that if an #sC cluster retains the more sonorous C2 instead of the /s/, the
reduction to the more sonorous segment will occur for all other #sC clusters. Thus,
if swim [swm] is reduced to [wm], then sleep [slip] should reduce to [lip].
Goad and Rose (2004) proposed the Headedness approach for the explana-
tions of #sC reductions. The idea in this approach is to retain the head of the clus-
ter and delete the non-head. In stage I, children choose the head of a cluster via
sonority (the lower sonority item], thus, sleep [slip] [sip]). In stage II, children
learn that /s/ is an adjunct and C2 is the head and thus the latter is retained (sleep
[lip]). In the following we will look at the patterns exhibited in the cross-
linguistic data we are considering. Figure 7 gives a summary of what is found in
both TD and PD groups. (For greater details see Yava 2013.)
Before we discuss these results, two points in the figure deserve attention.
Firstly, although the percentages given for Dutch /sN/ reflect the combination of
/sm/ and /sn/ targets, there was a significant discrepancy; as detailed in Yava
(2013), /sm/ targets show C2 preferences much more emphatically than /sn/ tar-
gets. Secondly, the use of /s/ in Hebrew and /s/ in Norwegian and Croatian, in-
stead of the missing /sw/, is because of their approximant status in the respective
languages, and should not be taken as a claim of absolute equivalence in sonority.
Eng. Dutch** Norw.**** Heb.*** Croat.**** Polish*
/sT/ TD C2 (83/17) C2 (82/18) C2 (88/12) C2 (85/15) C2 (100/0) C2 (84/16)

PD C2 (90/10) C2 (95/5) C2 (85/15) C2 (97/3) C2 (94/6)
/sN/ TD C2 (84/16) C2 (71/29) C2 (97/3) C2 (70/30) C2 (85/15) C2 (63/37)

PD C2 (96/4) C2 (75/25) C2 (68/32) C2(100/0) C2 (60/40)
/sl/ TD C1 (85/15) C1 (63/37) C1 (100/0) C1 (68/32) C1 (100/0) --- ---

PD C1 (90/10) C1 (67/33) C1 (85/15) C1 (75/25)
/sw/ TD C1 (73/27) --- --- C1 (67/33) C1 (92/8) C1 (100/0) C1 (60/40)

PD C1 (53/45) C1 (89/11) C1(100/0) C1 (78/22)
* No /sl/; ** No /sw/; *** No /sw/, /sr/ was looked at instead; **** No /sw/, /s/ was looked at instead
Figure 7. Summary of reduction patterns. Percentage points are given in the order of the
preferred retained consonant over the other (e.g. English /sT/ in P.D: C2 retained 90%, C1
retained 10%)
Mehmet Yava
As expected, data from children (TD as well as PD) strongly support the deletion
tendencies in /s+stop/ targets. The predominance of the retention of the less sono-
rous member, i.e., C2, in /sT/ targets is not specific to English-speaking children
and is repeated in other languages. However, this sonority based account runs into
problems when we look at the results obtained in /s+sonorant C/ clusters. In all of
these clusters, we have sonority rises from C1, /s/, to C2 in these clusters and this
should support the retention of /s/ and the deletion of the higher sonority C2s.
While this prediction is successful in the majority of cases with /s+approximant/
clusters, it fails rather miserably in /s+nasal/ clusters. Here, again, the pattern seems
to be shared by all six languages and in both TD and PD groups. It is interesting to
note that the predominance of C2 retention in /s+nasal/ targets is very similar to
that found in /s+stop/ targets. In fact, in some groups such as English PD, Norwe-
gian TD, and Croatian PD groups the preference of C2 retention over C1 in
/s+nasal/ targets is even more drastic than that of revealed by /s+stop/ targets. Such
behavior of /s+nasal/ targets clearly separates them from the other /s+sonorant C/
targets and suggests that they be paired with /s+stop/ targets. This pattern exhib-
ited by /s+nasal/ targets does not lend itself to an explanation based on sonority;
targets that are rising in sonority (sonority index changes from 3, for /s/, to 5, for
nasal) are expected to behave like other rising sonority targets (/s+ approximant/)
whereby the tendency found is more generally to delete the more sonorous C2.
Nor does this emphatic behavior of /s+nasal/ clusters have any explanation through
the factorial typology or the headedness approach, the two approaches that have
dealt with the deletions in /s/-clusters. Once again, as detailed in Yava (2013),
continuancy of C2 seems to be the factor in the pairing up of /s+stop/ and /s+nasal/
clusters. When C2 is [-continuant], a very strong tendency of the deletion of
the lower sonority /s/ is exhibited by all six languages. Since the remaining
/s+approximant/ clusters whereby C2 is [+continuant], the tendency is retaining
the lower sonority member, /s/, this suggests that sonority is an ancillary force and
is activated when by members of the cluster are [+continuant]. Data from Spanish-
English bilingual children give support to this view. Contrary to what we obtain in
monolingual English speaking children, in both typically developing children and
in children with phonological disorders, we see a very emphatic preference of sleep
[slip] [lip] (cf. Monolingual English [slip] [sip]). This is likely due to the very
different (very non-continuant) nature of the Spanish lateral.
Conclusions
In this chapter, we looked at the cross-linguistic patterns of the acquisition of #sC

clusters. We examined the results of the studies on six different languages with
data from typically developing children and children with phonological disorders.
We evaluated the results in childrens correct renditions, and in the reduction pat-
terns in incorrect productions. We wanted to see if the patterns were explainable
through sonority. The findings did not lend themselves to an explanation via so-
nority. For the accurate renditions, in the three Germanic languages (English,
Dutch, and Norwegian), there was a significant increase by children between
[continuant] C2 (e.g. /sT, sN/) and [+continuant] C2 (i.e. /s+approximant/). On
the other hand, in the three non-Germanic languages (Hebrew, Croatian, and
Polish) which have a rich inventory of clusters with negative sonority and level
sonority as well as clusters with [continuant] C2, showed no difference between
[continuant] C2 and [+continuant] C2. These differences in the two groups of
languages give support to the role of language-specific input frequency in chil-
drens acquisition patterns that are based on the patterns in the input language
whereby the input triggers the default settings to be set or reset.
The reduction patterns that are observed in the inaccurate productions, on the
other hand, presented a different picture. While in all six languages, variability oc-
curred both within and across children, a rather uniform picture emerged for the
preferences. Data from typically developing children and children with phono-
logical disorders reveal a binary grouping of #sC clusters: these are a) /sT and /sN/,
and b) /sl/ and /sw/. We see that in the reduction of the targets in the former group,
the retained consonant was predominantly C2, while the opposite tendency
(i.e. retaining C1) is observed when the reduction is involved the latter group.
Here, again, the binary split relies on the continuancy of C2. Stops and nasals
(i.e. [continuant]) have the common property of completely blocking the airflow
through the center of the vocal tract, as opposed to /s/+approximant/ clusters
which are [+continuant]. Overall, the two aspects of cross-linguistic #sC cluster
acquisition we examined suggest that both language-specific factors and universal
tendencies are at work.
References
Barlow, J.A. 2005. Sonority effects in the production of consonant clusters by Spanish-speaking
children. In Selected Proceedings of the sixth Conference on the Acquisition of Spanish and Por-
tuguese as a First and Second Language, D. Eddington (ed.), 114. Somerville MA: Cascadilla.
Barlow, J.A. 2001. A preliminary typology of initial clusters in acquisition. Clinical Linguistics
and Phonetics 15: 913.
Beckman, M. & Edwards, M.L. 1999. Lexical frequency effects on young childrens imitative
Productions. In Papers in Laboratory Phonology V, M.B. Broe & J.B. Pierrehumbert (eds),
208218. Cambridge: CUP.
Mehmet Yava
Ben-David, A. 2006. On the acquisition of Hebrew #sC onsets. Journal of Multilingual Commu-
nication Disorders 4: 205217.
Ben-David, A., Ezrati, R. & Stuhlman, N. 2010. Acquisition of /s/ clusters in Hebrew speaking
children with phonological disorders. Clinical Linguistics and Phonetics 24(3): 210223.
Berent, I., Lennert, T. & Smolensky, P. 2011. Syllable markedness and misperceptions: Its a two-
way street. In Handbook of Syllable, C.E. Cairns & E. Raimy (eds). Leiden: Brill.
Broselow, E. 1992. Language transfer and universals in second language epenthesis. In Language
Transfer in Language Learning [Language Acquisition and Language Disorders 6], S.M.
Gass & L. Selinker (eds), 7186. Amsterdam: John Benjamins.
Broselow, E. & Xu, Z. 2004. Differential difficulty in the acquisition of second language phonol-
ogy. International Journal of English Studies 4: 135163.
Broselow, E., Chen, S. & Wang, C. 1998. The emergence of the unmarked in second language
phonology. Studies in Second Language Acquisition 20: 261280.
Broselow, E. & Finer, D. 1991. Parameter setting in second language phonology and syntax.
Second Language Research 7: 3539.
Clements, G.N. 1990. The role of sonority cycle in core syllabification. In Papers in Laboratory
Phonology, 1: Between the Grammar and Physics of Speech, J. Kingston & M.E. Beckman
(eds), 283333. Cambridge: CUP.
Davis, S. 1990. Italian onset structure and the distribution of il and lo. Linguistics 28: 4355.
De Lacy, P. 2007. The interaction of tone, sonority, and prosodic structure. In The Cambridge
Handbook of Phonology, P. de Lacy (ed.), 281307. Cambridge: CUP.
Fikkert, P. 1994. On the Acquisition of Prosodic Structure. PhD dissertation, University of Leiden.
Gerrits, E. 2010. Acquisition of /s/ clusters in Dutch-speaking children with phonological disor-
ders. Clinical Linguistics and Phonetics 24(3): 199209.
Gerrits, E. & Zumach, A. 2006. The acquisition of #sC clusters in Dutch. Journal of Multi- lingual
Communication Disorders 4: 218230.
Giegerich, H.J. 1992. English Phonology. Cambridge: CUP.
Gierut, J. 1999. Syllable onsets: Clusters and adjuncts in acquisition. Journal of Speech, Language,
and Hearing Research 42: 708726.
Goad, H. & Rose, Y. 2004. Input elaboration, head faithfulness, and evidence for representation
in the acquisition of left-edge clusters in West Germanic. In Constraints in Phonological
Acquisition, R. Kager, J. Pater & W. Zonneveld (eds), 109157. Cambridge: CUP.
Gouskova, M. 2004. Falling sonority onsets, loanwords, and syllable contact. Chicago Linguistic
Society 37: 175186.
Gouskova, M. 2001. Relational hierarchies in optimality theory: The case of syllable contact.
Phonology 21: 201250.
Hall, T.A. 1992. Syllable Structure and Syllable Related Processes in German. Tubingen: Niemeyer.
Harris, J. 1983. Syllable Structure and Stress in Spanish. A Non-linear Analysis. Cambridge MA:
The MIT Press.
Hogg, R & McCully, C. 1987. Metrical Phonology: A Coursebook. Cambridge: CUP.
Hooper, J. 1976. An Introduction to Natural Generative Phonology. New York NY: Academic
Press.
Karimi, S. 1987. Farsi speakers and the initial consonant clusters in English. In Interlanguage
Phonology: The Acquisition of a Second Language Sound System, G. Ioup & S. Weinberger
(eds), 305318. Cambridge MA; Newbury House.
Kenstowicz, M. 1994. Phonology in Generative Grammar. Oxford: Blackwell.
Kristoffersen, K. & Simonsen, H. 2006. The acquisition of #sC clusters in Norwegian. Journal of
Multilingual Communication Disorders 4: 231241.
Lovins, J. 1974. Why loan phonology is natural phonology? Papers from the Annual Regional
Meeting, Chicago Linguistic Society: Parasession on natural phonology, 240250.
Mildner, V. & Tomic, D. 2010. Acquisition of /s/-clusters in Croatian-speaking children with
phonological disorders. Clinical Linguistics and Phonetics 24(3): 224238.
Morelli, F. 1999. The Phonotactics and Phonology of Obstruent Clusters in Optimality Theory.
PhD dissertation, University of Maryland.
Ohala, D.K. 1999. The influence of sonority on childrens cluster reductions. Journal of Commu-
Parker, S. 2002. Quantifying the Sonority Hierarchy. PhD dissertation, University of Massa-
chusetts.
Pater, J. 2004. Bridging the gap between receptive and productive development with minimally
violable constraints. In Constraints in Phonological Acquisition, R. Kager, J. Pater & W.
Zonneveld (eds), 219244. Cambridge: CUP.
Pater, J. & Barlow, J.A. 2003. Constraint conflict in cluster reduction. Journal of Child Language
30: 487526.
Pinker, S. & Birdsong, D. 1979. Speakers sensitivity to rules of frozen word order. Journal of
Verbal Learning and Verbal Behavior 18: 497508.
mar. Oxford: Blackwell.
Romani, C. & Calabrese, A. 1998. Syllabic constraints on the phonological errors of an aphasic
patient. Brain and Language 64: 83121.
Selkirk, E.O. 1984. On the major class features and syllable theory. In Language Sound Structure:
Studies in Phonology Presented to Morris Halle by his Teacher and Students, M. Aranoff &
R.T. Oehrle (eds), 107136. Cambridge MA; The MIT Press.
Smolensky, P. 2006. Optimality theory in phonology, II: Markedness, feature domains, and local
constraint conjunction. In The Harmonic Mind: From Neural Computation to Optimality-
theoretic Grammar, Vol. 2, P. Smolensky & G. Legendre (eds), 27160. Cambridge MA; The
MIT Press.
Stemberger, J.P. & Treiman, R. 1986. The internal structure of word-initial consonant clusters.
Journal of Memory and Language 25: 163180.
Stenneken, P., Bastiaanse, R., Huber, W & Jacobs, A. 2005. Syllable structure and sonority in
language inventory and aphasic neologisms. Brain and Language 95: 280292.
Steriade, D. 1990. Greek Prosodies and the Nature of Syllabification. New York NY: Garland.
Trommelen, M. 1984. The Syllable in Dutch: With Special Reference to Diminutive Formation.
Dordrecht: Foris.
Venneman, T. 1972. On the theory of syllabic phonology. Linguistische Berichte 18: 118.
Yava, M. 2013. What explains the reductions in /s/-clusters: Sonority or [continuant]? Clinical
Linguistics and Phonetics 27(6): 394403.
Yava, M. 2011. The role of sonority in the acquisition of interlanguage coda clusters. In Achieve-
ments and Perspectives in SLA Speech, M. Wrembel, M. Kul & K. Dziubalska-Kolacyk (eds),
297307. Frankfurt: Peter Lang.
Yava, M. 2010. Sonority and the acquisition of /s/ clusters in children with phonological disor-
ders. Clinical Linguistics and Phonetics 24(3): 167176.
Yava, M. 2003. Role of sonority in developing phonologies. Journal of Multilingual Commmu-
Mehmet Yava
Yava, M. & Marecka, M. In press. Acquisition of /s/ clusters in Polish: Patterns in typical devel-
opments and in children with phonological disorders. International Journal of Speech and
Language Pathology.
Yava, M. & McLeod, S. 2010. Acquisition of #sC onsets in English-speaking children with pho-
nological disorders. Clinical Linguistics and Phonetics 24(3): 177187.
Yava, M., Ben-David, A., Gerrits, E., Kristoffersen, K. & Simonsen, H. 2008. Sonority and
Cross-linguistic acquisition of initial s-clusters. Clinical Linguistics and Phonetics 22(6):
421441.
Yava, M. & Core, C. 2006. Acquisition of #sC clusters in English-speaking children. Journal of
Multilingual Communication Disorders 4(3): 169181.
Yava, M. & Someillan, M. 2005. Patterns of acquisition of /s/ clusters. Journal of Multilingual
Communication Disorders 3: 5055.
Zamuner, T.S., Gerken, L. & Hammond, M. 2004. Phonotactic probabilities in young childrens
speech production. Journal of Child Language 31: 515536.
Appendix A
Cluster inventories of Germanic languages

Bold italic (negative sonority)
Bold underlined (level sonority)
ENGLISH
CC clusters pl, pr, pj, bl, br, bj, tr, tw, (tj), dr, dw, (dj), kl, kr, kj, gl, gr, gw
fl, fr, fj, vj, r, w, hj
mj, (nj), (lj)
sp, st, sk, sm, sn, sl, sw, (sj), r
() found in some speakers of American English.
DUTCH
CC clusters pl, pr, bl, br, tr, tw, dr, kl, hn, kw, (ps), (ts), (tj), (dw)
fl, fr, xl, xr, (ft), (fn), (fj)
sp, st, sx, sm, sn, sl, zw, (sk), (sf), (sj)
() uncommonly found clusters.
NORWEGIAN
CC clusters p, pr, pj, b, br, bj, tr, (tj), t, dr, (dj), d, kn, k, kr, k, gn, g,
gr, fn, f, fr, fj
sp, st, sk, sm, sn, , s, s
Appendix B
Cluster inventories of non-Germanic languages

Bold underlined (level sonority)
Bold italics (negative sonority)
HEBREW
CC clusters bx, px, b, p, bs, ps, bz, pz, bts, pts, dv, df, tv, tf, ts, tz, d, t,
dx, tx, gv, gf, kv, kf, gz, gs, kz, ks, g, k, gx, kx, tsf, tsx, ts, sj,
j, ml, mn
bn, pn, bl, pl, b, p, dm, tm, tn, dl, tl, d, gm, km, gn, kn, gl,
kl, g, k, dj, tj, vl, v, f, fj, zm, sm, m, tsm, zn, sn, n, tsn,
zl, sl, l, tsl, z, s, , tsf, tsv, zx
bg, bk, pg, pk, bd, bt, pd, pt, dg, dk, tg, tk, gd, kd, kt, kts, zv,
sv, sf, f, sx, v, x
sp, zb, b, zd, sd, st, d, t, zg, sk, sg, sk, g, k, tsd
CROATIAN
CC clusters k, p, bz, ks, ps, p, dm, dn, gm, gn, g, km, kn, k, pn, tm,
m, , sm, sn, n, n, , m, n, br, dr, gr, kr, pr, tr, bl, dl, gl,
kl, pt, tl, l, p, k, g, sr, sl, s, , , fr, f;, fj, xr, xl, xv, x,
mr, mj, ml, m, rj, r, l, r
bd, gd, kt, pt, tk, sf, sx, mn
zb, zd, zg, b, d, g, ft, sk, sp, st, s, k, p, t,
POLISH
CC clusters p, pr, pj, pw, pl, ps, p, px, br, bj, bl, bw, b, bz, tr, tl, t, tf, tw,
tj, tx, t, dr, dm, dl, dv, dw, d, dj, d, dn, kj, kl, kf, k, k, kw,
ks, kn, kr, gr, gw, gl, gd, gv, g, g, gn, gz, gm, gj, fr, fj, fl, vj,
vw, vr, v, vn, vl, vm, sm, sn, sr, sj, sl, sw, zn, zr, zj, zm, zw, z,
zl, l, w, m, l, m, w, r, l, , m, l, w, f, tsm, tsn, dzv,
tf, dzv, xw, xr, xj, xl, xm, mj, mw, ml, mr, nj, lj, wz
pt, db, tk, kt, gd, fs, fx, f, f, vz, v, vv, sf, sx, s, ss, v, v,
zv, zz, z, f, v, f, t, xf, x, mn, m
ft, fts, ft, f, fp, fk, vd, vb, st, sp, sk, s, sts, zg, zb, zd, zd, ,
k, p, t, b
t, p, t, xts, xt rt, rv, lv, rdz
The role of phonological context in childrens
overt marking of -s in two dialects
of American English

San Diego State University
We consider the role of preceding and following phonological context in the

variable marking of monosegmental English plural and 3rd-person singular
morphemes by African-American English speaking children (Study 1), and by
younger children speaking a more mainstream variety of American English
(Study 2). Results of both studies show that overt marking was favored in
utterance-final contexts and with plurals, consistent with prior research findings.
However, preceding consonant contexts favored the overt marking of the
morphemes, contrary to prior research findings. We discuss these findings in
terms of syllable organization and morphophonological complexity, and offer
suggestions for future research to further delineate the specific contributions of
phonological context to the overt marking of morphemes in child language.
Introduction
Much research on child language focuses on specific core areas of language

(e.g., phonology, morphology, or syntax) without considering the interaction that
occurs between these different areas. However, a surge in interest in the interfaces
between different language domains has recently developed (see, for example, an
entire issue dedicated to interfaces in a double issue of First Language in 2012),
along with recognition of the need to better account for them (Jackendoff 2002;
Rothman & Guijarro-Fuentes 2012). For instance, children often have difficulty
with phonological constructs (e.g., word-final consonant clusters in English, as in
[ts] in cats) that can impact morphological markers (such as plural -s; hereafter
plural-s). Similarly, the complexity of certain types of morphological markers
(such as the English morpheme that indicates past tense, -ed) likely impacts pho-
nological accuracy. Morphophonology is the interface that we consider herein.
Morphophonology
Morphophonology is a phenomenon that occurs naturally in all languages of the

world. For instance, in most varieties of English, plural-s undergoes a morphopho-
nological alternation depending on the surrounding phonological context. Specifi-
cally, when -s follows a voiceless consonant, such as [p t k], it is pronounced as [-s],
as with cats [kts] and caps [kps]. When -s follows a sibilant consonant, such
as [s z], it is pronounced as [-z], as with busses [bsz] and buzzes [bzz]. Fi-
nally, when -s follows a non-sibilant voiced consonant, such as [b d m n l ] or a
vowel, it is pronounced as [-z], as with cabs [kbz], cans [knz], and bees [biz].
Thus, plural-s has different phonetic realizations, depending on the surrounding
phonological context. Third-person singular -s (hereafter, 3s) shows an identical
distribution by phonetic context: eats [its], kisses [ksz], and runs [nz].
Morphophonology is also a phenomenon of child language, and represents
patterns that may not be present in the language of the surrounding community,
but are possible linguistic constructs nonetheless (e.g., Barlow & Keare 2008;
Camarata & Gandour 1985; Munson & Ingram 1985; Weismer, Dinnsen & Elbert
1981). Children may exhibit variation in their productions of morphological forms
due to phonological context, and this morphophonological variation can be differ-
ent from that of the adult language. For instance, Weismer and colleagues (1981)
observed that children with phonological disorders may produce an alternation
between lexical morphemes produced with and without a diminutive suffix. Spe-
cifically, they observed some children producing the word dog as [d], thereby
omitting the final []; however, when these same children produced the word
doggie (dog + diminutive suffix), they correctly produced the [], as in [di].
Thus, the lexical morpheme dog exhibited a morphophonemic alternation, at-
tributed to phonological context and a phonotactic constraint against word-final
(or syllable-final) consonants.
Generally speaking, syllables with final consonants or final consonant clusters
(CVC or CVCC(C)) are viewed as more complex relative to simple, open (CV)
syllables, given their relatively limited typological distribution and their later ac-
quisition (Blevins 1995; Demuth, Culbertson & Alter 2006; Kirk & Demuth 2005).
Because of these effects of complexity, the addition of any monosegmental bound
morpheme, such as plural-s for example, to any stem will create, at a minimum, a
CVC form. When the stem already ends with a consonant, the addition of the
morpheme creates a more complex syllable shape.
Many researchers assume that monosegmental word-level (or level-2) English
morphemes such as plural-s or 3s /-s, -z/, past tense -ed /-t, -d/, and ordinal -th
/-/ belong outside the syllable constituent as an appendix to the syllable or pro-
sodic word, at least when attaching to a stem with an already heavy rhyme
Phonological context and marking of -s
(Clements & Keyser 1983; Fudge 1969; Fujimura & Lovins 1978; Giegerich 1992;
Goldsmith 1990; Goldsmith 2011; Halle & Vergnaud 1980; Kiparsky 1985; Selkirk
1982). Cross-linguistically, such extrasyllabic segments occur only at word edges
word-initially and word-finally in those languages that allow them. Appendices
allow for the occurrence of additional consonants, often comprising morphologi-
cal content that would not be allowed in the onset or coda constituent, and are
furthermore immune to language-universal and language-specific phonotactic re-
strictions that apply to the syllable proper (Borowsky 1989). When produced in
the middle of an utterance, these forms may resyllabify with the syllable of the fol-
lowing word, as permitted by English phonotactics (e.g., if followed by a vowel).
In fact, a common trend cross-linguistically is for word-final consonants to be
syllabified into the onset of following vowel-initial words. In Spanish, for instance,
las alas the wings is syllabified as [la.salas] (where a period . denotes a syllable
boundary) (Colina 1997). In French, word-final consonants are regularly omitted,
yet are pronounced when followed by a vowel, as with les amis [le.zami] the
friends (De Jong 1990). The same phonological forces driving these patterns also
drive the morphophonological alternation between the indefinite articles a and an
in English, the latter of which occurs before words beginning with vowels. In each
case and in other sandhi phenomena, the language is showing a preference for syl-
lables with onsets, and, in some cases, an avoidance of syllables with (complex)
codas (Casali 1996; Jakobson 1968; Senturia 1998).
Given this cross-linguistic tendency, it stands to reason that the variable mark-
ing of morphemes in word-final position in sound systems may be influenced at
least in part by complexity of phonological context. Indeed, in African American
English (AAE), a dialect known for variable coda simplification and variable
marking of morphemes (particularly tense morphemes), phonological context
would be expected to influence the overt marking of these morphemes. Final con-
sonants are variably omitted, and final consonant clusters are variably reduced in
adult AAE (Green 2002; Thomas 2007), and both simplification patterns are de-
pendent on segmental factors. For instance, final consonant deletion typically af-
fects oral and nasal stop consonants more so than other manners of articulation,
and this is attributed to the fact that phonemic contrasts can be maintained on the
preceding vowel in the form of nasalization and vowel length differences (Moran
1993; Stockman 1996).
In AAE final consonant clusters involving a stop consonant as the second
member, the stop is more likely to be omitted following a nasal as opposed to a
liquid, a liquid as opposed to a sibilant, and a sibilant as opposed to another stop.
Moreover, final clusters with a large sonority distance are more likely to be reduced
than those with a small sonority distance (Thomas 2007). Additional research sug-
gests, however, that final clusters are less likely to be reduced when the second
(stop) consonant represents a morpheme (e.g., closed [kloz+d]; Thomas 2007;

Wolfram 1969). Interestingly, in stop + /s/ or /z/ clusters, the stop may be deleted,
especially if the fricative represents a separate morpheme (e.g., thats [ds];
Thomas 2007).
Cluster simplification of this sort in AAE may occur even when the following
word begins with a vowel; however, it is much more likely to occur when the fol-
lowing word begins with a consonant, particularly an obstruent (Green 2002;
Stockman 1996; Thomas 2007; Wolfram 2005; Wolfram & Thomas 2002). This is
similar to what has been reported for other dialects of English as well. In fact,
across dialects of English, word-final /t/ and /d/ may be omitted, and the likeli-
hood of omission varies inversely with the sonority of the following segment. The
segments are most likely to be omitted when followed by an obstruent, and least
likely when followed by a vowel (Guy 1991; Labov 1997). One phonological ac-
count of this phenomenon is that any segment that cannot be syllabified as part of
a preceding coda, for instance because the maximum number of segments has
been exceeded, and cannot be syllabified as the onset to the following word be-
cause it does not begin with a vowel, is subject to stray erasure (e.g., Guy 1991).
Variable morpheme marking in child language
The role of phonology in the variable marking of morphemes has also been evalu-
ated in developing sound systems. Researchers have evaluated English-speaking
childrens productions of grammatical morphemes, such as past tense and past
(or passive) participle -ed (as in he cooked), 3s (as in he eats), and plural-s
(discussed above), and phonological context has been determined to affect the ac-
curacy of these particular morphemes (for review, see Demuth this volume). The
past tense and past participle forms are more prone to omission in complex pho-
nological contexts by children who are developing language in a typical or atypical
fashion, and for children who speak different English dialects, including AAE
(Johnson & Morris 2007; Marshall & van der Lely 2007; Oetting & Horohov 1997;
Pruitt & Oetting 2009; Stemberger 2007). Specifically, the past tense morpheme is
more prone to omission when it follows a consonant (as in cooked [kkt]) than
when it occurs after a vowel (as in played [pled]) (Marshall & van der Lely 2007;
Oetting & Horohov 1997; Pruitt & Oetting 2009; Pruitt, Oetting & Hegarty 2011;
Redmond 2003). Prior research on childrens overt marking of past tense in AAE
also shows the effect of phonology. Pruitt and Oetting (2009) observed that AAE-
speaking children are more likely to mark the regular past tense when the verb
stem ends with a vowel as opposed to a consonant, consistent with prior descrip-
tions of the adult dialect (Green 2002; Rickford 1999).
Several studies have also addressed the role of phonology in the variable mark-
ing of plural-s and 3s in the acquisition of more mainstream varieties of English.
For instance, Song, Sundara and Demuth (2009) found that young children are
more likely to omit 3s when produced in the middle of an utterance, as opposed to
utterance-finally (see also Sundara, Demuth & Kuhl 2011). Moreover, they found
that the morpheme was marked less often when preceded by a consonant (as with
he eats) as opposed to a vowel (as with he sees). Thus, the morpheme was more
likely to be omitted when it occurred in a complex phonological context.
Theodore, Demuth and Shattuck-Hufnagel (2011) determined that 2-year-
olds are more likely to omit plural-s in connected speech when it occurs in the
middle of an utterance as opposed to the end of the utterance, consistent with the
findings of Song et al. (2009). However, these researchers did not observe an effect
of the preceding phonological context. That is, plural-s was no less accurate when
it followed a consonant as compared to a vowel. As suggested by Theodore et al.,
perhaps the different findings of the two studies are attributable to the relative
complexity of the morphological markers evaluated. That is, 3s is typically consid-
ered to be a more complex morpheme in that it marks person, tense, and number,
and is typically acquired later by children, as compared to plural-s, which only
marks number (Brown 1973; de Villiers & de Villiers 1973). This added complex-
ity may make the morpheme more vulnerable to the effects of phonological com-
plexity, thus rendering more errors in complex phonological contexts.
Perceptual salience could also explain the different behavior of the two mor-
phemes in the input. Hsieh, Leonard and Swanson (1999) found that plural-s has
a longer duration than 3s in child-directed adult speech. This was due in part to
the fact that the former were more frequent in utterance-final position where
phrase-final lengthening commonly occurs (Beckman & Edwards 1990; Dalal &
Loeb 2005; Price, Ostendorf, Shattuck-Hufnagel & Fong 1991; Song et al. 2009).
In contrast to those results reported by Theodore et al. (2011), Polite (2011)
did find an effect of word-final complexity on overt marking of plurals, such that
children were more likely to overtly mark plural-s on vowel-final as opposed to
consonant-final noun stems. Ettlinger and Zapf (2011) also found an effect of pho-
nological context on plural-s marking, once again with vowel-final noun stems
favoring overt marking. Moreover, they found that 2-year-olds were more likely to
mark plural-s following sonorant as opposed to obstruent consonants. The re-
searchers also found similar effects in these childrens comprehension of plural
forms. This is consistent with sonority sequencing principle accounts of syllable
structure, which assume that a falling sonority profile is preferred syllable finally
(Borowsky 1989; Clements 1990). Similar findings were reported by Oetting and
Horohov (1997) in their study of past tense marking in typically developing chil-
dren and those with language impairment. Recall also that cluster simplification
patterns in AAE and in developing systems are dependent on the segmental make-
up of the individual clusters, which can be interpreted in terms of sonority (Green
2002; Kirk & Demuth 2005; Thomas 2007).
Taking the prior research findings together, it appears that children are more
likely to overtly mark some morphemes after a vowel than a consonant. Moreover,
position within an utterance impacts the likelihood that a morpheme will be
overtly marked, such that it is more likely to be marked utterance-finally than ut-
terance-medially. Is there, however, an added effect of context for overt marking of
utterance-medial morphemes in terms of the segments that follow? That is, is a
morpheme more likely to be overtly marked when followed by a consonant or a
vowel? In the abovementioned studies, the researchers either intentionally includ-
ed forms that were followed by vowels in order to minimize the effect that the
following context might have on overt marking (Marshall & van der Lely 2007;
Pruitt & Oetting, 2009), or they intentionally excluded such forms in order to
avoid the likelihood that the morpheme would be syllabified with the following
word (Song et al. 2009, Study 1). In other studies involving experimental tasks, the
following contexts were limited to a specific kind of consonant, such as an ob-
struent stop or a nasal (Song et al. 2009, Study 2; Sundara et al. 2011; Theodore
et al. 2011). Yet, as discussed above, we know from research on different dialects of
adult English that following context impacts the production of word-final conso-
nant clusters.
Thus far, only one known study has manipulated the following phonological
context in evaluation of childrens morpheme production. In addition to evaluat-
ing the phonological complexity of the noun stem in overt plural-s marking, as
described above, Polite (2011) also evaluated whether the word following the plu-
ral form began with vowel or a consonant. Though preceding context influenced
overt plural-s marking, following context did not. Whether the following word
began with a vowel or consonant did not impact plural-s marking.
To summarize, there are inconsistent findings regarding the effects of preced-
ing phonological context on childrens overt marking of plural-s and 3s, and there
is limited information available regarding the effects of following phonological
context on morpheme marking. In effort to remedy this, in this paper we consider
the role of preceding and following phonological context in the variable marking
of monosegmental English plural and 3s morphemes (hereafter, -s) by AAE-
speaking children (Study 1), and by younger children speaking a more mainstream
variety of American English (MAE; Study 2). For both studies, we predicted that
overt marking of -s would be disfavored most in adjacent consonant contexts
(preceding and/or following), due to the complex syllable profile that would oth-
erwise result. We also predicted that differences would obtain between the two
different morphemes given that 3s is linguistically more complex than plural-s.
Study 1
Introduction
In Study 1, we consider the overt marking of -s in the speech of AAE-speaking

children. As stated above, variable marking of -s, particularly 3s, is a hallmark
feature of AAE (Green 2002), and our goal is to determine if AAE-speaking chil-
dren illustrate effects of phonological context on their overt production of -s.
Methods
Participants and Data. The data from 8 typically-developing children (6 girls,

2 boys), aged 5364 months (mean 58.6 months), were drawn from a larger study
of language acquisition by AAE-speaking children from Louisiana (Pruitt 2006;
Pruitt & Oetting 2009; Pruitt et al. 2011). These 8 children all scored within typical
range on measures of nonverbal intelligence, vocabulary, and morphosyntax. Fur-
ther, based on the results of a subjective listener rating scale (Oetting & McDonald
2002), all 8 children were judged to be occasional-to-heavy users of AAE features,
meaning that at least 25% of their utterances were perceived as featuring AAE dia-
lect features. Details about each of the children and their language skills are pro-
vided in Table 1.
The data were derived from 20-minute play-based language samples, each of
which featured at least 100 intelligible utterances. These samples were orthograph-
ically transcribed and coded for morphosyntax using Systematic Analysis of
Language Transcripts (SALT; Miller & Iglesias 2004) and phonetically transcribed
using broad notation of the International Phonetic Alphabet (IPA). Inter-rater
Table 1. Study 1 Child Characteristics (adapted from Keare 2009; Pruitt 2006)
Child Age (months) Sex MLU* Listener Judgment**
1 54 Female 6.7 6.3

2 64 Male 7.0 5.3
3 56 Female 6.4 5.0
4 53 Female 3.8 5.0
5 57 Female 7.2 5.0
6 58 Female 5.4 4.7
7 66 Female 6.6 4.7
8 61 Male 5.3 4.7
*Mean length of utterance (in morphemes)
**Listener judgment of dialect features on scale from 1 (no AAE features) to 7 (40% or greater use of AAE
features) (Oetting & McDonald 2002)
coding for SALT and for transcription reliability for consonant agreement both
were above 90%.
Analyses. Each language sample was evaluated for presence or absence of -s
(as a plural-s or 3s morpheme) and the corresponding phonological context. Be-
cause we were interested only in the [-s] and [-z] allomorphs and their effect on
syllable complexity, we excluded all forms that ended in a sibilant (and would thus
be marked with the [-z] allomorph). In some cases, over-regularization of plurals
occurred, whereby an already plural-s form had an additional plural marker
(e.g., pants as [pntsz]); these forms were likewise excluded. Each remaining
relevant form (n=158) was then coded for the specific morpheme (plural-s, 3s),
preceding sonority context (obstruent, sonorant, or vowel), and following sonority
context (obstruent, sonorant, vowel, or utterance-final). Sonority was limited to
these broad categories because the number of forms representing each individual
manner class (e.g. stop, fricative, nasal) across the childrens productions was in-
sufficient for statistical analysis.
AAE is characterized as a non-rhotic dialect; r-lessness is noted to occur to
some degree in virtually all varieties of AAE (Thomas 2007), and is especially like-
ly to occur in preconsonantal contexts. This feature was not prevalent in the speech
of the 8 children in the present study, who produced // regularly across contexts.
Because of this, the forms in which -s occurred after // (e.g., plural-s in cars
[kz]) were coded as post-sonorant, rather than postvocalic. /s/ forms occurring
after schwar // (e.g., flowers [flaz]), however, were treated as postvocalic.
AAE is likewise deemed an l-less dialect, in that postvocalic /l/ may be vocal-
ized as [o] (feel as [fio] or [fi]) or may be absent altogether (pull as [p]), par-
ticularly in preconsonantal contexts (Thomas 2007). Once again, this feature was
not prevalent in the utterances of the 8 children in this study. Thus, -s forms occur-
ring after /l/ (e.g., plural-s in balls [blz]) were also coded as post-sonorant, rather
than postvocalic. Not included in the 158 forms are those forms which were fol-
lowed by words beginning with /s/ or /z/ (e.g., eats spaghetti), due to the difficulty
in reliably determining whether or not the -s morpheme was actually marked.
Due to the heterogeneous nature of our sample, we used a mixed statistical
model to analyze the effects of the target morpheme, the preceding sonority con-
text, and the following sonority context on the overt marking of -s, which allowed
us to account for the effects of individual differences in number of productions by
child. Specifically, the model evaluated differences in overt marking by morpheme
(plural-s, 3s), preceding sonority context (obstruent, sonorant, vowel), and follow-
ing sonority context (obstruent, sonorant, vowel, utterance-final). Subject was
added as a random effect in order to control for the very unequal numbers of pro-
ductions per child. By doing so, any influence of one childs production patterns
would not be the cause for any significant effects of morpheme, preceding context,
or following context (or lack thereof).
Results
Of 158 -s forms analyzed (109 plural-s, 49 3s), 120 (76%) were overtly marked,
with the majority of zero-marked forms occurring for 3s. Specifically, plural-s was
overtly marked in 106 forms (97%), while 3s was overtly marked in 14 forms
(29%). All children made multiple attempts at plural-s, and overtly marked it more
often than they did not. The numbers for 3s require further mention given their
small numbers. Some children zero-marked all 3s forms: Specifically, Child 1
marked 0/10 forms, Child 2 marked 0/1 form, and Child 3 marked 0/5 forms.
In a preliminary analysis, the statistical model revealed main effects for mor-
pheme, F(1, 154) = 54.51, p < .05, but not for utterance position, F(1, 154) = 2.40,
p > .05. Moreover, there was no interaction between morpheme and utterance po-
sition, F(1, 154) = 1.89, p > .05, which is likely due in part to the relatively few oc-
currences of 3s forms in utterance-final position. As expected, plural-s forms (M =
97.3, SD = 16.4) were overtly marked more than 3s forms (M = 28.6, SD = 45.6).
Although the effect of utterance position was not significant, -s was overtly marked
more often utterance-finally (M = 94.4, SD = 23.1) than utterance-medially (M =
66.4, SD = 47.5). There were 104 utterance-medial (59 plural, 45 third -person
singular) and 54 utterance-final (50 plural, 4 third-person singular) forms.
All forms were then entered into a model that evaluated morpheme, preceding
context, and the interaction between the two. Results showed a significant main
effect of morpheme, F(1, 152) = 186.21, p < .05, and preceding context, F(1, 152)
= 7.61, p < .05. These main effects were qualified by a significant morpheme-by-
context interaction, F(2, 152) = 3.94, p < .05. Specifically, 3s was overtly marked
more following an obstruent than following a vowel. Table 2 shows the means and
standard deviations for each morpheme by preceding context.
Table 2. Means (and Standard Deviations) of Overt Marking by Preceding Context in AAE
Preceding Context
Morpheme Obstruent Sonorant Vowel Total
plural 100.0 (0) 100.0 (0) 94.2 (23.5) 97.3 (16.4)

n = 30* n = 27 n = 52 n = 109
3s 42.3 (50.4) 28.6 (48.8) 6.3 (25.0) 28.6 (45.6)
n = 26 n=7 n = 16 n = 49
Total 73.2 (44.7) 85.3 (36.0) 73.5 (44.4) 76.0 (42.8)
n = 56 n = 34 n = 68 n = 158
*Refers to total number of forms attempted in that context
Table 3. Means (and Standard Deviations) of Overt Marking of Utterance-Medial -s

by Following Context in AAE
Following Context
plural 94.7 (22.9) 90.0 (31.6) 100.0 (0) 96.6 (18.3)

n = 19* n = 10 n = 30 n = 59
3s 27.8 (46.1) 18.8 (40.3) 36.4 (50.5) 26.7 (44.7)
n = 18 n = 16 n = 11 n = 45
Total 62.2 (49.2) 46.2 (50.8) 82.9 (38.1) 66.4 (47.5)
n = 37 n = 26 n = 41 n = 104
* Refers to total number of forms attempted in that context
All utterance-medial forms were also analyzed for following context. Results
showed, once again, a main effect of morpheme, F(1, 98) = 96.82, p < .05, with
plural-s forms overtly marked more than 3s forms. The main effect for following
context was not significant, F(1, 98) = 1.26, p > .05, nor was the interaction be-
tween morpheme and following context, F(2, 98) = 0.95, p > .05. Table 3 shows the
means and standard deviations for each utterance-medial morpheme by following
context.
Discussion
Given the findings reported in prior research, we predicted that AAE-speaking

children would (1) show higher overt marking of plural-s as compared to 3s, and
(2) that overt marking would be disfavored in adjacent consonant contexts (either
preceding or following the morpheme), following the assumption that adjacent
consonants, particularly obstruents, create complex phonological contexts.
The results partially supported our predictions. Specifically, the children did
show significantly higher overt marking rates for plural-s versus 3s, and in fact the
plural-s forms were at ceiling levels. Given the greater complexity that is assumed for
3s and its relatively later age of acquisition reported (Brown 1973; de Villiers & de
Villiers 1973), these findings are unsurprising. Prior research has shown a clear ad-
vantage of utterance-final position over utterance-medial position (i.e., before a con-
sonant or vowel). Although there was no significant effect of utterance position on
-s production, the trend was in the direction that favored utterance-final position.
What was surprising, however, was the effect of context on the childrens pro-
ductions of the two morphemes. Though we predicted that phonological context
would impact the childrens overt marking, the effect was not in the direction
we predicted. Specifically, -s was overtly marked more in contexts following a
consonant, specifically an obstruent, as opposed to a vowel. Interestingly, there

was no effect of following sonority context in the overt marking of -s in the utter-
ance-medial forms, though the trend was in the direction that favored marking in
prevocalic contexts (sees a), which is consistent with our original predictions. In
addition, the lack of any effect of phonological context for plural-s specifically may
be due to the ceiling levels of marking for that morpheme.
Because some of the results of Study 1 are contrary to our predictions and
what has been reported in prior research, we set out to determine if this pattern
was unique to children acquiring AAE, or might also be observed in younger chil-
dren acquiring a more mainstream variety of American English. This was the aim
of Study 2.
Study 2
Introduction
In Study 2, we consider the overt marking of -s in the speech of MAE-speaking

children who are younger than those in Study 1. Variable marking of -s is not
characteristic of adult MAE, yet is observed in the speech of younger children who
have not yet fully mastered adult-like morpheme use. Our goal is once again to
determine if these MAE-speaking children illustrate effects of phonological con-
text on their overt production of -s.
Methods
Participants and Data. The longitudinal data from 6 typically-developing children

(3 girls, 3 boys), aged 1524 months (mean 21.2 months), were drawn from the
Providence Corpus (Demuth et al. 2006), the same corpus analyzed by Song and
colleagues (2009). Details about this corpus are available on the Child Language
Data Exchange System Project (http://childes.psy.cmu.edu/). These 6 children
were from the southern New England area of the United States. Two of the chil-
dren were characterized as speaking the dialect characteristic of the area, while the
other four were characterized as speaking a dialect more similar to Standard
American English.
The data were derived from 60-minute language samples of parent-child inter-
actions that occurred every two weeks from the onset of first word production.
These samples were orthographically transcribed and coded using the Codes for
the Human Analysis of Transcripts (MacWhinney 2000) and phonetically tran-
scribed using broad notation of the IPA. Reported inter-rater reliability for broad
Table 4. Study 2 Child Characteristics (adapted from Demuth et al. 2006; Song et al. 2009)
Child Age (months) Sex MLU*
Alex 29 Male 2.18

Ethan 16 Male 1.52
Lily 21 Female 1.84
Naima 17 Female 2.34
Violet 22 Female 1.59
William 22 Male 1.56
*Mean length of utterance in morphemes
phonemic transcription of all utterances ranged from 80% to 98% (Demuth et al.
2006). For the purposes of the current study, we evaluated productions at only the
youngest age of the range analyzed by Song and colleagues (2009) in order to avoid
greater variation in morpheme production associated with change over time. The
characteristics of each child as described by Demuth et al. (2006) and Song et al.
(2009) are provided in Table 4.
Analyses. The data were coded and statistically analyzed following the same
procedures as in Study 1, yielding 197 -s forms. Although two of the children
(Alex and Violet) were characterized as showing features of the southern New
English dialect (Song et al. 2009), which is characterized as non-rhotic, inspection
of their productions across the longitudinal database revealed numerous produc-
tions of postvocalic [] and rhotic vowels (e.g., Alex: [o ts ov d], oh, its over
there; Violet: [bn sneks fn] airborne snakes are fun). Therefore, to be
consistent with analysis procedures used in Study 1, the forms in which -s oc-
curred after // were coded as post-sonorant, rather than postvocalic. As before,
/s/ forms following schwar // were treated as postvocalic.
Results
Of 197 -s forms analyzed (174 plural-s, 23 3s), 169 (86%) were overtly marked.
Plural-s was overtly marked in 162 forms (93%), while 3s was overtly marked in 7
forms (30%). All children made multiple attempts at plural-s, and marked it more
often than they did not. As with Study 1, the numbers for 3s require further men-
tion given their small numbers. Some children did not attempt any 3s forms that
met our criteria (Alex and Naima). Additionally, Lily marked 1/3 forms, and
William marked 0/10 forms.
In a preliminary analysis, the statistical model revealed an interaction between
morpheme and utterance position, F(1, 193) = 66.12, p < .05. Main effects for mor-
pheme, F(1, 193) = 54.51, p < .05, and utterance position, F(1, 193) = 31.13, p < .05,
were also significant. As expected, plural-s forms (M = 93.10, SD = 25.41) were

overtly marked more than 3s forms (M = 30.43, SD = 47.05). Also as expected, -s
was overtly marked more often utterance-finally (M = 92.57, SD = 26.32) than ut-
terance-medially (M = 65.31, SD = 48.09), but this effect was limited to 3s forms.
There were 49 utterance-medial (34 plural, 15 third-person singular) and 148 ut-
terance-final (140 plural, 8 third-person singular) forms.
All forms were then entered into a model that evaluated morpheme, preceding
context, and the interaction between the two. Results showed a significant main
effect of morpheme as before, F(1, 191) = 12.99, p < .05, where plural-s was overt-
ly marked more than 3s. There was also a main effect of preceding context, F(1,
191) = 3.62, p < .05, such that -s was overtly marked more after sonorants than
vowels. The morpheme-by-context interaction was not significant, F(2, 191) =
2.78, p > .05. Table 5 shows the means and standard deviations for each morpheme
by preceding context.
All utterance-medial forms were also analyzed for the effects of following
context. Results showed, once again, a main effect of morpheme, F(1, 43) = 37.23,
p < .05, with plural-s overtly marked more than 3s. The main effect for following
context was not significant, F(1, 98) = 1.26, p > .05; however, the interaction be-
tween morpheme and following context was significant, F(2, 43) = 5.03, p < .05. 3s
was overtly marked more when followed by a vowel than a sonorant. Table 6 shows
the means and standard deviations for each morpheme by following context.
Discussion
The MAE-speaking children showed significantly higher overt marking rates for
plural-s versus 3s, which was unsurprising. Also, consistent with other reports in
the literature, including those describing these very children (Song et al. 2009;
Theodore et al. 2011), the utterance-final context favored overt marking of 3s.
Table 5. Means (and Standard Deviations) of Overt Marking by Preceding Context in MAE
Preceding Context
plural 97.9 (14.6) 94.4 (23.2) 90.1 (30.0) 93.1 (25.4)

n = 14* n = 36 n = 91 n = 174
3s 33.3 (57.7) 100.0 (0) 26.3 (45.2) 30.4 (47.0)
n=3 n=1 n = 19 n = 23
Total 94.0 (24.0) 94.6 (22.9) 79.1 (40.9) 85.8 (35.0)
n = 50 n = 37 n = 110 n = 197
Table 6. Means (and Standard Deviations) of Overt Marking of Utterance-Medial

-s by Following Context in MAE
Following Context
plural 94.4 (23.6) 100.0 (0) 75.0 (46.3) 91.2 (28.8)

n = 18* n=8 n=8 n = 34
3s 0 (0) 0 (0) 50 (70.7) 6.7 (25.8)
n=1 n = 12 n=2 n = 15
Total 89.5 (31.5) 40.0 (50.3) 70.0 (48.3) 65.3 (48.1)
n = 19 n = 20 n = 10 n = 49
Once again, we were surprised by the nature of the effect of phonological context
on overt marking of -s. MAE-speaking children overtly marked -s more after a
consonant, specifically a sonorant, than after a vowel. For following context, vow-
els were favored over sonorants among the 3s forms. In addition, as with the AAE-
speakers, the lack of any effect of phonological context for plural-s specifically may
be due to the near-ceiling levels of marking for that morpheme.
General discussion
To summarize findings from both studies, we observed that AAE- and MAE-
speaking children overtly marked plural-s more than 3s. We also observed that
utterance-final position was favored for both groups of children, though this effect
was significant only for MAE-speaking children in their productions of 3s, though
the lack of an effect of utterance position for plural-s may be due to ceiling effects.
Additionally, both groups of children marked -s more often after a consonant (e.g.,
eats or runs) than after a vowel. For AAE-speaking children, obstruents were fa-
vored over vowels, while for MAE-speaking children, sonorants were favored, con-
sistent with findings reported for younger MAE speakers (Ettlinger & Zapf 2011).
Effects of following context, for utterance-medial forms, showed a preference for
vowels over sonorants, though once again this was significant only for the MAE-
speaking childrens productions of 3s forms. This contrasted with Polites (2011)
findings, which showed no effect of following context for plural-s forms.
Why do these results differ from those of prior studies? We propose that the
results of prior studies may have been skewed because either there was no consid-
eration of following context (Ettlinger & Zapf 2011; Oetting & Horohov 1997),
or there was only a basic comparison of utterance-final versus utterance-medial
forms that also restricted following contexts, as discussed previously (Pruitt &
Oetting 2009; Song et al. 2009; Sundara et al. 2011; Theodore et al. 2011). In con-
trast, in the present research, we included all possible following contexts other
than those followed by /s/ or /z/-initial words due to difficulty in reliably deter-
mining overt marking. It should also be noted that we analyzed only a single point
in time from the MAE-speaking childrens longitudinal data (Demuth et al. 2006),
whereas Song and colleagues evaluated these same childrens productions drawing
from several different points in time, which may have obscured patterns that oc-
curred at individual points in time. Perhaps by evaluating data from only a single
point in time, and by including more following contexts, we uncovered an effect of
context that would not have otherwise been detected. Finally, our results may dif-
fer from those of Polite (2011) because of methodological differences. Polite con-
servatively included data from only those children who made three or more target
plural-s attempts for each following context (vowel or consonant), while in the
current study all childrens attempts (for both plural-s and 3s) were included.
The differences observed between the AAE- and MAE-speaking children
could be attributed to a number of factors. Most obviously, the children spoke dif-
ferent dialects that are known to differ in terms of phonology and morphology.
The two groups also were of different ages, and the AAE-speaking children
were producing longer, more complex utterances, as evidenced by their MLUs
(see Table 1 and Table 4). In addition, the data were elicited differently in the two
studies. The AAE data were based on interactions between the child and an exam-
iner who could guide the interactions to elicit certain types of grammatical
structure, such as tense morphemes. In contrast, the MAE data were based on in-
teractions between the child and caregiver who likely did not attempt to elicit
specific linguistic structures. Moreover, there were many more occurrences of
plural-s forms, particularly utterance-finally, in comparison to 3s forms, which is
due in part to English subject-verb-object syntax (Song et al. 2009). Because of
these differences, the data differed within and across the two studies in terms of
the number of -s forms evaluated, the number of plural-s and 3s forms, and the
number of utterance-medial versus utterance-final forms. These factors may have
contributed to the different profiles observed for the two groups.
Given that both preceding and following contexts have been observed to im-
pact overt marking of -s in the present study and in prior studies with different
dialects and ages of speakers, it is likely that there is a combinatorial effect of sur-
rounding context on marking of -s. That is, a context in which -s occurs between
consonants (e.g., eats the) is likely to impact the overt marking of -s differently
from an intervocalic context (e.g., sees a). As stated previously, some researchers
assume that -s and other word-final monosegmental suffixes are not part of the
syllable coda when they follow another consonant, but are rather considered
extrasyllabic (Borowsky 1989). Following this assumption, then, when -s follows

a vowel, but does not occur before a vowel-initial word (e.g., sees the), it syllabi-
fies in the coda of the word to which it attaches. When it occurs following an ob-
struent or a sonorant, but is not followed by a vowel-initial word (e.g., eats the), it
may be extrasyllabic, per assumptions about sonority, structure preservation, and
core syllable structure (Borowsky 1989). If this is so, then we might assume that
codas are more vulnerable to omission than extrasyllabic segments. Similar obser-
vations have been made for the acquisition of extrasyllabic /s/ as an adjunct in
word-initial clusters as compared to branching onsets (Barlow 2001; Gierut 1999).
But when -s occurs before another vowel (e.g., sees a), it may syllabify into the
onset of the following word. Because syllables with onsets are preferred over on-
setless syllables cross-linguistically (Blevins 1995), it is no surprise that -s is more
likely to be overtly marked when followed by a vowel.
What is the explanation for the overall higher overt marking of -s utterance-
finally? Why would a coda or an extrasyllabic segment be preferred utterance-
finally but not utterance-medially? In explanations of very young childrens patterns
of -s production, the greater phonetic salience associated with utterance-final
forms is often cited as a reason for childrens higher rates of marking of -s in that
context, suggesting limits to their mastery of these morphemes (Demuth this vol-
ume; Gerken & McGregor 1998; Sundara et al. 2011). Yet, it is difficult to accept this
explanation for the older AAE-speaking children (though see Johnson 2005), or
adult speakers of the dialect for that matter, who are assumed to have mastered the
-s morphemes, but are producing them variably in accordance with the degrees of
freedom allowed by the dialect. An alternative explanation is that phrase-final
lengthening allows for -s to be marked utterance-finally (Demuth this volume;
Gerken 1996; Song et al. 2009), presumably because more time is afforded for tim-
ing and sequencing of the articulatory gestures. Yet another explanation is that the
pattern is due to phonological constraints that refer to phrase edges. Cross-linguistic
evidence from fully-developed languages shows that not only the ends of words,
but also the ends of phrases, permit structure that is phonotactically impermissible
word- or phrase-internally (Wiltshire 2003).
Taken together, these findings, preliminary though they are, suggest that a
hierarchy of constraints may be interacting to yield the pattern of overt marking of
-s observed herein. It appears that utterance-position and coda complexity alone
cannot predict the patterns observed. Rather, the patterns may be best accounted
for under a constraint-based framework (McCarthy & Prince 1995; Prince &
Smolensky 2004) that allows for interaction between faithfulness to the input
(e.g., MaxMorpheme) and conspiring effects of context: an avoidance of codas
(e.g., *Coda) and extrasyllabic segments (e.g., Exhaustivity), a preference for
onsets (e.g., Onset), adherence to sonority sequencing (e.g., *M/) and the syllable
contact law (e.g., SyllCon), and an avoidance of complex structure utterance-

medially (e.g., Align-R(Phrase,C)) (Blevins 1995; Davis 1998; Green 2003; It
1986; Kisseberth 1970; McCarthy & Prince 1995; Prince & Smolensky 2004;
Wiltshire 2003). Following assumptions of factorial typology and language varia-
tion, it is moreover possible that the findings in Studies 1 and 2 reflect different
constraint rankings due to both age and dialect (Barlow & Gierut 1999; McCarthy
& Prince 1995; Prince & Smolensky 2004).
As discussed above, the data evaluated in each study were limited in that they
did not allow for a statistical evaluation of the full range of possible surrounding
contexts. This presents a logistical challenge, because an evaluation of surrounding
context requires a large enough language sample to allow for multiple productions
of -s in each possible combination of preceding and following contexts. An ex-
perimental task could be used, as with Song et al. (2009), Theodore et al. (2011),
and Ettlinger and Zapf (2011), but the task would need to be extensive enough to
include all such possible combinations, and this can be taxing for very young chil-
dren, such as were included in those prior studies. On the other hand, it would be
possible to administer such a task to older children who speak a dialect that has
variable -s marking, as with AAE.
Space limitations prevented the inclusion of a number of other types of rele-
vant analyses. For instance, we did not evaluate segmental patterns (e.g., substitu-
tions, omissions) associated with the target morphemes, or the phonemes that
preceded or followed those morphemes. A more detailed understanding of how
heteromorphemic structures may be phonologically simplified is also necessary,
and should be compared and contrasted with tautomorphemic structures, which
may pattern differently in both fully-developed and developing phonological sys-
tems (Kirk & Demuth 2005; Stemberger & Bernhardt 1997; Thomas 2007). We
also did not acoustically analyze the childrens productions; if we had, we may have
identified more cases of overt marking by the children than a perceptual analysis
alone (Theodore et al. 2011; Theodore, Demuth & Shattuck-Hufnagel 2012). Still,
a perceptual analysis has value, in that it reflects what is acceptable or noticeable
on the part of adult native speakers of the language. Additionally, future research
should also further evaluate the effect that lexical and sub-lexical factors, such as
word frequency, neighborhood density, age of acquisition, inflectional frequency,
and phonotactic probability, have on the overt marking of morphemes in chil-
drens speech, as preliminary research indicates that these factors may differen-
tially affect morphophonological interactions (Hoover, Storkel & Rice 2012; Polite
2008; cf. Theodore, Demuth & Shattuck-Hufnagel 2010).
The fact that so many different factors may impact childrens morphophono-
logical patterns does of course present a logistical challenge; nevertheless, research-
ers can aim to identify those that have the greatest influence. Future research that
involves much larger language samples, and/or experimental tasks that specifically
manipulate or control for these variables would shed further light on the variable
marking of -s and other morphemes such as past tense -ed. This ultimately will
aid our understanding of the interfaces between the different domains of language,
in both developing and fully-developed systems, and has the added benefit of in-
forming assessment and treatment strategies with individuals with speech/lan-
guage impairment who have difficulty with phonology and/or morphology.
In conclusion, phonological context influenced the rate of overt marking of -s
in both AAE and MAE and in the same general manner. Further study of syllabi-
fication patterns across word boundaries in (various dialects of) English is
necessary. This combined with further inquiry into morphonological patterns in
dialects like AAE and in the speech of young children will enhance our under-
standing of the complex relationship between phonology and morpheme use.
Acknowledgments
As first author, I am honored to have the fortune to have worked with Dan Dinnsen,
as his student and now his colleague. He is always a great source of knowledge,
encouragement, inspiration, and laughter. I am indebted to him, and I am so very
grateful that he is my friend. Thanks also go to Dr. Soonja Choi, members of SaD-
PhIG, and Roxanne Belauskas, Alanna Call, Skott Freedman, Louise Lam, Mary
Orton, Marissa Saridakis, Peter Torre III, and Elizabeth Wylie for valuable feed-
back and assistance on this study. Aspects of Study 1 also appear in Keares (2009)
masters thesis.
References
Barlow, J.A. 2001. The structure of /s/-sequences: Evidence from a disordered system. Journal of
Barlow, J.A. & Gierut, J.A. 1999. Optimality theory in phonological acquisition. Journal of
Speech, Language, and Hearing Research 42:14821498.
Barlow, J.A. & Keare, A.R. 2008. Acquisition of final voicing: An acoustic and theoretical ac-
count. In Indiana University Working Papers in Linguistics, Vol. 6: Phonological Opacity Ef-
fects in Optimality Theory, A.W. Farris-Trimble & D.A. Dinnsen (eds), 8197. Bloomington
IN: IULC Publications.
Beckman, M. & Edwards, J. 1990. Lengthening and shortening and the nature of prosodic con-
stituency. In Papers in Laboratory Phonology, I: Between the Grammar and the Physics of
Speech, J. Kingston & M. Beckman (eds), 152178. Cambridge: CUP.
Blevins, J. 1995. The syllable in phonological theory. In The Handbook of Phonological Theory,
J.A. Goldsmith (ed.), 206244. Oxford: Blackwell.
Borowsky, T. 1989. Structure preservation and the syllable coda in English. Natural Language
and Linguistic Theory 7:145166.
Brown, R. 1973. A First Language: The Early Stages. Cambridge MA: Harvard University Press.
Camarata, S.M. & Gandour, J. 1985. Rule invention in the acquisition of morphology by a lan-
guage-impaired child. Journal of Speech and Hearing Disorders 50: 4045.
Casali, R.F. 1996. Resolving Hiatus, PhD dissertation, UCLA.
Clements, G.N. 1990. The role of the sonority cycle in core syllabification. In Papers in Labora-
tory Phonology, I: Between the Grammar and Physics of Speech, J. Kingston & M.E. Beckman
Clements, G.N. & Keyser, S.J. 1983. CV Phonology: A Generative Theory of the Syllable. Cambridge
MA: The MIT Press.
Colina, S. 1997. Identity constraints and Spanish resyllabification. Lingua 103: 123.
Dalal, R.H. & Loeb, D.F. 2005. Imitative production of regular past tense -ed by English-speak-
ing children with specific language impairment. International Journal of Language & Com-
munication Disorders 40: 6782.
Davis, S. 1998. Syllable contact in Optimality Theory. Korean Journal of Linguistics 23: 181211.
De Jong, D. 1990. The syntax-phonology interface and French liaison. Linguistics 28: 5788.
de Villiers, J. & de Villiers, P. 1973. A cross-sectional study of the acquisition of grammatical
morphemes in child speech. Journal of Psycholinguistic Research 2:267278.
Demuth, K., Culbertson, J. & Alter, J. 2006. Word-minimality, epenthesis, and coda licensing in
the acquisition of English. Language & Speech 49:137174.
Ettlinger, M. & Zapf, J. 2011. The role of phonology in childrens acquisition of the plural. Lan-
guage Acquisition 18: 294313.
Fudge, E.C. 1969. Syllables. Journal of Linguistics 5: 253286.
Fujimura, O. & Lovins, J. 1978. Syllables as concatenative phonetic units. In Syllables and Seg-
ments, A. Bell & J.B. Hooper (eds), 107120. Amsterdam: North-Holland.
Gerken, L. 1996. Prosodic structure in young childrens language production. Language 72:
683712.
Gerken, L. & McGregor, K.K. 1998. An overview of prosody and its role in normal and disor-
dered child language. American Journal of Speech-Language Pathology 7: 3848.
Giegerich, H.J. 1992. English Phonology: An Introduction. Cambridge: CUP.
Gierut, J.A. 1999. Syllable onsets: Clusters and adjuncts in acquisition. Journal of Speech, Lan-
guage, and Hearing Research 42:708726.
Goldsmith, J.A. 1990. Autosegmental and Metrical Phonology. Cambridge MA: Blackwell.
Goldsmith, J.A. 2011. The syllable. In The Handbook of Phonological Theory, J.A. Goldsmith, J.
Riggle & A.C.L. Yu (eds), 164196. Malden MA: Blackwell.
Green, A.D. 2003. Extrasyllabic consonants and onset well-formedness. In The Syllable in Opti-
mality Theory, C. Fry & R.v.d. Vijver (eds), 238253. Cambridge: CUP.
Green, L.J. 2002. African American English: A Linguistic Introduction. Cambridge: CUP.
Guy, G. 1991. Explanation in variable phonology: An exponential model of morphological con-
straints. Language Variation and Change 3: 122.
Halle, M. & Vergnaud, J.-R. 1980. Three dimensional phonology. Journal of Linguistic Research
1: 83105.
Hoover, J.R., Storkel, H.L. & Rice, M.L. 2012. The interface between neighborhood density and
optional infinitives: normal development and Specific Language Impairment. Journal of
Hsieh, L.I., Leonard, L.B. & Swanson, L.A. 1999. Some differences between English plural noun
inflections and third singular verb inflections in the input: The contributions of frequency,
sentence position, and duration. Journal of Child Language 26: 531543.
It, J. 1986. Syllable Theory in Prosodic Phonology, PhD dissertation, University of Massachu-
setts, Amherst.
Jackendoff, R. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: OUP.
Jakobson, R. 1968. Child Language, Aphasia and Phonological Universals. The Hague: Mouton.
Johnson, B.W. & Morris, S.R. 2007. Clinical implications of the effects of lexical aspect and pho-
nology on childrens production of the regular past tense. Child Language Teaching and
Therapy 23:287306.
Johnson, V.E. 2005. Comprehension of third person singular /s/ in AAE-speaking children. Lan-
guage, Speech, and Hearing Services in Schools 36: 116124.
Keare, A.R. 2009. The Role of Phonological Context in the Omission of Plural, Possessive, and
3rd Person Singular Morphemes in AAE-Speaking Children. MA thesis, San Diego State
University.
Kiparsky, P. 1985. Some consequences of lexical phonology. Phonology 6: 285303.
Kirk, C. & Demuth, K. 2005. Asymmetries in the acquisition of word-initial and word-final
consonant clusters. Journal of Child Language 32: 709734.
Kisseberth, C. 1970. On the functional unity of phonological rules. Linguistic Inquiry 1: 291306.
Labov, W. 1997. Resyllabification. In Variation, Change, and Phonological Theory [Current Issues
in Linguistic Theory 146], F. Hinskens, R. van Hout & W.L. Wetzels (eds), 145179.
Amsterdam: John Benjamins.
MacWhinney, B. 2000. The CHILDES Project: Tools for Analyzing Talk. Mahwah NJ: Lawrence
Erlbaum Associates.
Marshall, C.R. & van der Lely, H.K.J. 2007. The impact of phonological complexity on past tense
inflection in children with Grammatical-SLI. International Journal of Speech-Language Pa-
thology 9: 191203.
McCarthy, J.J. & Prince, A.S. 1995. Faithfulness and reduplicative identity. In University of Mas-
sachusetts Occasional Papers 18, J.N. Beckman, L.W. Dickey & S. Urbanczyk (eds), 249384.
Amherst MA: GLSA.
Miller, J. & Iglesias, A. 2004. Systematic Analysis of Language Transcripts (SALT), English &
Spanish. Madison WI: Language Analysis Lab, University of Wisconsin.
Moran, M.J. 1993. Final consonant deletion in African American children speaking Black
English: A closer look. Language, Speech, and Hearing Services in Schools 24:161166.
Munson, J. & Ingram, D. 1985. Morphology before syntax: A case study from language acquisi-
tion. Journal of Child Language 12: 681684.
Oetting, J.B. & Horohov, J.E. 1997. Past-tense marking by children with and without specific
language impairment. Journal of Speech, Language, and Hearing Research 40: 6274.
Oetting, J.B. & McDonald, J.L. 2002. Methods for characterizing participants nonmainstream
dialect use in child language research. Journal of Speech, Language, and Hearing Research
45: 505518.
Polite, E.J. 2008. The Influence of Frequency- and Contextually-Related Factors on the Use of
Regular Noun Plural -s by Children with Specific Language Impairment. PhD dissertation,
Purdue University.
Polite, E.J. 2011. The contribution of part-word phonological factors to the production of regu-
lar noun plural -s by children with and without specific language impairment. First Lan-
guage 31: 425441.
Price, P.J., Ostendorf, M., Shattuck-Hufnagel, S. & Fong, C. 1991. The use of prosody in syntactic
disambiguation. Journal of the Acoustical Society of American 90: 29562970.
mar. Malden MA: Blackwell.
Pruitt, S.L. 2006. Grammatical Morphology of Children Reared in Poverty: Implications for
Specific Language Impairment. PhD dissertation, Louisiana State University.
Pruitt, S.L. & Oetting, J.B. 2009. Past tense marking by African American English-speaking
children reared in poverty. Journal of Speech, Language, and Hearing Research 52: 215.
Pruitt, S.L., Oetting, J.B. & Hegarty, M. 2011. Passive participle marking by African American
English-speaking children reared in poverty. Journal of Speech, Language, and Hearing Re-
search 54: 598607.
Redmond, S.M. 2003. Childrens productions of the affix -ed in past tense and past participle
contexts. Journal of Speech, Language, and Hearing Research 46:10951109.
Rickford, J.R. 1999. African American English: Features, Evolution, Educational Implications.
Malden MA: Blackwell.
Rothman, J. & Guijarro-Fuentes, P. 2012. Linguistic interfaces and language acquisition in child-
hood: Introduction to the special issue. First Language 32: 316.
Selkirk, E.O. 1982. The syllable. In The Structure of Phonological Representations, H. van der
Hulst & N. Smith (eds), 337383. Dordrecht: Foris.
Senturia, M.B. 1998. A Prosodic Theory of Hiatus Resolution. PhD dissertation, University of
California, Berkeley.
Song, J.Y., Sundara, M. & Demuth, K. 2009. Phonological constraints on childrens production
of English third person singular -s. Journal of Speech, Language, and Hearing Research 52:
623642.
Stemberger, J.P. 2007. Childrens overtensing errors: Phonological and lexical effects on syntax.
Journal of Memory and Language 57: 4964.
Stemberger, J.P. & Bernhardt, B.H. 1997. Phonological constraints and morphological development.
In Proceedings of the 21st Annual Boston University Conference on Language Development, E.
Hughes, M. Hughes & A. Greenhill (eds), 603614. Somerville MA: Cascadilla Press.
Stockman, I.J. 1996. Phonological development and disorders in African American children. In
Communication Development and Disorders in African American Children: Research,
Assessment, and Intervention, A.G. Kamhi, K.E. Pollock & J.L. Harris (eds), 117153.
Baltimore MD: Paul H. Brookes.
Sundara, M., Demuth, K. & Kuhl, P.K. 2011. Sentence-position effects on childrens perception
and production of English third person singular -s. Journal of Speech, Language, and Hear-
ing Research 54: 5571.
Theodore, R.M., Demuth, K. & Shattuck-Hufnagel, S. 2010. An acoustic examination of the ef-
fects of word frequency and utterance position on 2-year-olds production of 3rd person
singular -s. In Proceedings of the 13th Australasian International Conference on Speech
Science and Technology, M. Tabain, J. Fletcher, D. Grayden, J. Hajek & A. Butcher (eds),
158161. Melbourne, Australia: Australasian Speech Science and Technology Association
Theodore, R.M., Demuth, K. & Shattuck-Hufnagel, S. 2011. Acoustic evidence for positional and
complexity effects on childrens production of plural -s. Journal of Speech, Language, and
Hearing Research 54: 539548.
Theodore, R.M., Demuth, K. & Shattuck-Hufnagel, S. 2012. Segmental and positional effects on
childrens coda production: Comparing evidence from perceptual judgments and acoustic
analysis. Clinical Linguistics & Phonetics 26: 755773.
Thomas, E.R. 2007. Phonological and phonetic characteristics of African American Vernacular
English. Language and Linguistics Compass 1:450475.
Weismer, G., Dinnsen, D.A. & Elbert, M. 1981. A study of the voicing distinction associated with
omitted, word-final stops. Journal of Speech and Hearing Disorders 46: 320328.
Wiltshire, C.R. 2003. Beyond codas: Word and phrase-final alignment. In The Syllable in Opti-
mality Theory, C. Fry & R.v.d. Vijver (eds), 254270. Cambridge: CUP.
Wolfram, W. 1969. A Sociolinguistic Description of Detroit Negro Speech. Washington DC: Center
for Applied Linguistics.
Wolfram, W. 2005. African American English. In Clinical Sociolinguistics, M.J. Ball (ed), 87100.
Malden MA: Blackwell.
Wolfram, W. & Thomas, E.R. 2002. The Development of African American English. Malden MA:
Blackwell.
German settlement varieties in Kansas
Some unusual phonological and morphological
developments with the approach of language death
William D. Keel
University of Kansas
After discussing the unique opportunity for linguistic investigation afforded

by so-called speech-islands, speech communities embedded in a linguistic/
cultural context far removed from the parent linguistic/cultural situation,
we examine the appropriateness of the theoretical concept speech-island,
determining that a vaguer concept of speech settlement allows for a more
accurate depiction of the linguistic interaction with the surrounding dominant
culture. After a sketch of German settlements in Kansas, we conclude that the
current status of these settlements is rapidly approaching language death. In this
context, a number of unusual developments in the phonology (e.g., velarization
of [f]), morphology (emergence of a prepositional case) and lexicon (semantic
transfer of English to like) in Kansas German varieties present linguists with
explanatory challenges.
Introduction
Since the mid-nineteenth century, linguists have been intrigued by the notion of
the so-called Sprachinsel, or as Leonard Bloomfield termed it the speech-island
(1933: 53). How would a speech community separated by distance from its parent
speech community develop over time? Would its linguistic structures phonolo-
gy, morphology, syntax, lexicon, etc. mirror developments in the parent speech
community? Would these structures, in a sense, take on a life of their own? Would
they generalize or elaborate more limited developments in the parent community?
Would they adopt aspects of the surrounding, foreign speech community? Would
they succumb to interference or influence from that contact situation with a for-
eign, perhaps dominant, speech community in which they were situated? Would
that language contact over time lead to total assimilation of the speech-island into
William D. Keel
the foreign language/culture and ultimately the language death of the isolated
speech community?
The distinctive character of linguistic developments in such speech-islands or
isolated foreign settlements was recognized by researchers of German dialectol-
ogy during the first half of the twentieth century such as Viktor Schirmunski who
stated that these settlements represented a linguistic laboratory of extraordinary
significance (1930: 113). This has been reiterated numerous times by scholars in
German dialectology, including as recently as Alfred Wildfeuer (2012: 18). Clear-
ly, the investigation of such speech communities, often in distant isolation from
their parent communities, may shed much light on issues of language variation
and change.
Before examining aspects of the linguistic development in speech-islands of
German varieties in Kansas, we need to first consider the terminology of speech-
island itself. At first blush, the impression may be given of an intact, isolated
speech community of a parent language that has through group migration found
a new home within another speech community or in its original sense, left be-
hind as the larger community receded and was replaced by a foreign language
except in the island. This image of an island is already problematic. Particu-
larly in the American Midwest, German immigrant communities developed over
time, typically via chain migration from the European homeland. There was inter-
action with the surrounding English-speaking community from the very begin-
ning of settlement. As the group of German-speaking settlers grew larger, they
typically established a church which became the focal point of the community. At
that stage, one might begin to view the German variety as the language of a
speech-island. But this theoretical linguistic isolation was not without penetra-
tion from the surrounding dominant culture.
The situation in Kansas was as Carroll Reed noted for Pennsylvania German
(1971: 7): Most of these German speech islands have been surrounded or pene-
trated by English, both early and late, and often enough the inroads of English
have been introduced by force. Perhaps the self-identification of the speech com-
munity as something separate from the surrounding, dominant, culture is the
most salient feature for characterizing the group as a speech-island as is argued
by Mattheier (2003: 17; all translations from German by the author):
In addition, however, especially in speech-islands, we must consider the identity
component. Speech islands arise, not the least from the discourse of the people
who are participants in these enclaves. And this discourse determines what is
considered to be part of the speech-island and what belongs to another neighbor-
ing speech-island. This approach leads, in general, to small rural communities [of
isolated farmsteads] or groups of two or three villages, which are viewed by its
speakers and also by the surrounding society as allochthonous enclaves.
German settlement varieties
We are thus left with a rather vague notion of speech-island that defies easy char-
acterization. Wildfeuer (2012: 13ff.) argues convincingly that the German varieties
found in the American Midwest would be better described as speech settlements
allowing for the complex interaction of those communities with the existing and
dominant American English society in which the German communities are lo-
cated. After rejecting a number of other theoretical positions and notions such as
enclave regarding these German varieties outside of the Central European home-
land (Russia, Hungary, and the United States), Wildfeuer (2012: 17) comes to the
conclusion, emphasizing his interest in German Bohemian (Bavarian) dialects in
the United States, South America and New Zealand, that:
the current German Bohemian settlements date back to a concentrated, typically
short phase of settlement and were often from the beginning and for several gen-
erations in constant contact with other varieties of German and other languages.
Hence they do not represent a hermetically sealed community. The contact situ-
ation resulted beginning with the first generation of settlers in an at least partial
bilingualism or multilingualism, which in some cases resulted in alterations in the
German Bohemian first language.
Based on some thirty-five years of researching the German settlement varieties in

Kansas and Missouri, I concur with Wildfeuers assessment and will refer to the
German varieties discussed in what follows as settlement dialects or varieties of
German knowing full well that the terminology allows for a broad range of cir-
cumstances and linguistic interactions with the surrounding culture.
An additional aspect affecting the study of these German settlement dialects
in the Midwest is that, for the most part, we are witnessing the final stages of com-
plete assimilation of these varieties to American English. With the exception of
communities of Old Order Amish in which what is commonly called Pennsylva-
nia German or Pennsylvania Dutch is still acquired by children, frequently used in
the home environment and mandated for use in community worship services
(Meindl 2009) and the influx of Low German-speaking Mennonite families from
northern Mexico to such places as southwestern Kansas, the German settlement
varieties of Kansas and Missouri are no longer being acquired by younger genera-
tions of speakers. As documented in the comprehensive study by J. Neale Carman
(1962), the most viable of the German settlement varieties of nineteenth-century
Kansas were no longer being actively acquired by younger generations after the
mid-1940s. The fieldwork undertaken since 1980 to record and document these
varieties has thus captured the usage of what many would call semi-speakers,
heritage speakers, or rememberers whose use of the German variety is typi-
cally occasional. For everyday functions, English is now the predominant variety
in these communities. Given the fragile nature of the speech community and the
William D. Keel
last generation of relatively competent speakers, it is understandable that we might

encounter a high degree of variation in the speech of the remaining heritage
speakers, perhaps even more so as, for instance, Jean Aitchison notes for everyday
speech (2001: 159):
One of the major discoveries of the twentieth century was the tremendous amount
of variation that exists in speech sounds: the same sound is measurably different
when spoken by different speakers, in different words, at different speeds, and at
different levels of loudness.
This type of variation might likely be exaggerated as a speech community ap-

proaches the end of its existence and not only be restricted to its phonology, but
also morphological, syntactic and other aspects of its linguistic systems.
German settlement varieties in Kansas
From the 1850s until the present day, Kansas has been the destination for numer-
ous groups speaking varieties of non-standard (non-literary) German. In addi-
tion to immigrants from the area known as the German Reich (1871), groups
from Switzerland, Austria-Hungary and particularly, the Russian Empire found
new homes in Kansas in the late nineteenth and early twentieth centuries. Even
today, Kansas continues to receive new immigrants from Mennonite colonies in
northern Mexico whose primary language is known as Plautdietsch (Mennonite
Low German). And, as a destination for secondary settlement from the eastern
United States, Kansas has a number of groups representing established German
settlement varieties such as the Pennsylvania German of the Old Order Amish.
Nearly every county in Kansas has or had a settlement of Germans, in some cases
several distinct settlements. Most of them assimilated to the dominant English
culture within one or two generations. Today, some 30 percent of Kansans claim
German ancestry based on the most recent data provided by the U.S. Census Bu-
reau (20072011).
In what follows, we will sketch the German settlement history of Kansas, re-
stricting ourselves to only those rural settlements of Germans designated by
Carman as linguistically significant. Carman utilized such factors as (1) size, (2)
concentration, (3) historical importance, and (4) persistence of identity to deter-
mine each settlements significance (1962: 2). For our purposes, linguistic signifi-
cance will be based on Carmans assessment of the settlements German variety as
Super or Ultra High (1962: 2, 322) on his scale of significance.
This yields the following settlements in Kansas with a German variety, using
Carmans designations with the year of initial settlement:
1. Primary Catholic Russian German west central Kansas 1875 Super.

2. Concentrated Mennonite District south central Kansas 1874 Super.
3. Great Wabaunsee Wabaunsee County 1868 Ultra High.
4. Bremen-Horseshoe Marshall and Washington counties 1858 Ultra
High.
5. Ellis Bukovinians Ellis County 1884 Ultra High.
6. Andale-Colwich Sedgwick County 1872 Ultra High.
These six groups can be further subdivided into two major categories: those with
immigrants primarily from what became the German Reich in 1871, groups 3, 4,
and 6; those with immigrants primarily from the Russian Empire, groups 1 and 2,
or from the Austro-Hungarian Empire, group 5.
The Bremen-Horseshoe area extending over two counties and actually into
neighboring southern Nebraska represents a large area of settlement by Germans
from northern Germany from the vicinity of Bremen and Hannover in modern
Germany. Settled largely in the 1860s and 1870s, the area today has at most a cou-
ple of hundred speakers of a Low German variety. It does boast a Low German
Heritage Society. Historically, this was an area of Low German dialect speech, with
at least two distinct varieties of Low German, Eastphalian and East Frisian (Seeger
2006: 92109; cf. also Bender 1970).
The other two settlement areas with immigration directly from Germany, the
Great Wabaunsee and Andale-Colwich, also established in the decades of and after
the Civil War, found themselves in close proximity to two urban areas, Topeka and
Wichita, respectively, and rapidly lost their foreign character. The Wabaunsee
settlement along Mill Creek was a dialectally mixed but largely all German settle-
ment. Catholics from the western Rhineland settled in the Andale-Colwich dis-
trict. Carman designates the period 19181922 (1962: 289) as the critical year
for German usage in the Great Wabaunsee District and 1935 for that in Andale-
Colwich (1962: 269).
Three groups, however, that could be considered double transplants have
persisted the longest, or at least equally as long as the Low Germans of the Bremen-
Horseshoe group. Two of these groups originated in the Russian Empire and emi-
grated to the U.S. at the same time following the loss of colonial privileges in Russia
in the 1870s. Recruited by the transcontinental railroads to settle in Kansas, thou-
sands of German Mennonites from colonies near the Black Sea began settling
along the Santa Fe Railroad in the 1870s, while to the north along the Kansas
Pacific Railroad, similar numbers of German Catholics and Protestants from colo-
nies on the Southern Volga River were recruited to establish new homes on the
Great Plains of western Kansas. Their ancestors having migrated to Russia begin-
ning in the 1760s, these Russian Germans represent a unique group of German
William D. Keel
settlers in Kansas having language contact with Russian and English in the course
of the past two centuries plus dialect mixture occurring in both the original settle-
ment in Russia and again in the second settlement in Kansas.
The third non-German group of immigrants also comes from Eastern Europe
to Kansas. However, their route took them from the border areas of Bohemia with
Bavaria and the Rhenish Palatinate to the Austrian Crownland of Bukovina in the
late eighteenth century before a second migration to Kansas beginning in the late
1880s. Again, contact with Czech, Ukrainian, Rumanian among other languages
prior to the encounter with English in Kansas makes this group of Austrians
similarly unique for our research (Lunte 2007).
All three of these Eastern European German groups persisted in using their
respective variety of German until the period after World War II. The breakup of
the agricultural society with movement to the urban areas and the general mobil-
ity of the population following the introduction of the automobile and paved roads
in the middle of the twentieth century created pressures that did not permit these
rural communities to sustain themselves. Thus all three of these groups are now in
a phase of heritage societies with little or no attempt to resurrect the dying im-
migrant German variety. The Middle Bavarian variety of the Bukovina Germans
(Lunte 2007), the South Hessian and Western Palatine varieties of the Volga
Germans (Johnson 1994, Keel 2004, Khramova 2011), the Plautdietsch of the
German Mennonites from Russia (Baerg 1960), all of these immigrant varieties
are gradually fading from the Kansas landscape. Only the new immigration from
Mexico gives Plautdietsch a temporary reprieve. Only time will tell whether this
German variety can survive in Kansas.
Puzzling developments in the German heritage varieties of Kansas
With the exception of the Old Order Amish and the recent Mennonite groups
from Mexico, the transition to English in these groups has been largely completed,
with only elderly heritage speakers and semi-speakers able to communicate in the
variety of their ancestors to any extent. The process of acquisition and ultimately
attrition in these dying varieties has yielded a number of interesting and perhaps
unique phonological and morphological innovations over the course of several
generations in Kansas.
Velarization of f
For instance, representing the large numbers of Volga Germans who settled in West
Central Kansas in the 1870s, some of the current group of semi-fluent speakers
exhibit phonological traits not found in any other variety of German, whether in
Europe or the United States. For instance, in his study of the Volga German variety
spoken in the community of Schoenchen (Ellis County), Johnson documented for
some speakers in some circumstances that word final [f] has developed into an [x]
as in the common preposition uff > ux/ox on (literary German auf on). This is
clearly audible in the online sound files of the Linguistic Atlas of Kansas German
Dialects <http://www2.ku.edu/ ~germanic/LAKGD/ Atlas_Intro.shtml>.
Data from Johnsons subjects are presented in an orthographic rather than a
phonetic transcription (based on Posts [1990] system for transcribing Palatine
dialects and using x and ch for the velar and palatal fricatives, respectively) and are
based on a translation task of the 40 Wenker Sentences used in German dialectol-
ogy since the 1870s for data collection as well as supplementary texts. To elicit data
from subjects, we translated the original German version of the Wenker Sentences
into English. This led to a number of issues regarding the data including interfer-
ence from English, difficulty in the task itself, etc. But by and large, the data elicited
using this method are comparable to those collected around the world by research-
ers of German dialectology using this set of sentences. The numbers in (1) refer to
the number of the Wenker Sentence; the letters refer to Johnsons subjects:
(1) Wenker 32. Didnt you (plural) find a piece of soap for me on my table?
d. ir hot mol al e schtickelje saaf ux en disch kfunne for mich?
you (pl.) have once already a little piece soap on a table found for me?
e. du host mol schtick saaf ux meine disch sen laie?
you (sing.) have once piece soap on my table seen lay?
The other seven informants either avoided the use of the preposition auf or used a
form with the expected labial fricative [f], in one instance lenited to [v]. Similarly,
three informants used [x] in this preposition in (2).
(2) Wenker 36. What kind of little birds are sitting up there on the little wall?
a. vas fir glaane fegelje sits da ox der glaan vant?
what for little birdies sit there on the little wall?
d. vas fir glaane fegel sein des vas sitse dran ux der glaane vand?
what for little birds are those that sit up there on the little wall?
e. ves fir glaane fegelje sitse dort ux der glaan vand?
what for little birdies sit there on the little wall?
The other six informants used the expected form of the preposition with either [f]
or [v] or avoided using this preposition entirely. A survey in 1999 of all recordings
made in the Volga German settlements in Ellis and Rush counties by Michael
Putnam and Bradley Weis (2004) found only one additional subject in the town of
Liebenthal, Rush County, producing a velarized [f] in Wenker Sentence 36.
William D. Keel
Interestingly, in other Wenker Sentences recorded by Johnson, this preposi-

tion is articulated with an [f] in final position by all informants in which this form
is attested. Johnson does include two narrated texts by informants in which the
preposition or verbal suffix (literary German auf) is produced with the velar frica-
tive, as in (3) (1994: 180, 181):
(3) des geld vill ich uxheve
the money want I (to) save (literary German aufheben)
un hun ax uxs land gevont (literary German aufs)
and (they) have also in the country lived
Based on our studies of the Volga German varieties in Kansas (Johnson 1994; Keel
2004; Khramova 2011) as well as Kaull (1996), these examples cited in Johnson
appear to be isolated shifts of [f] to [x], restricted to the form which regularly ap-
pears in this dialect as uff or off and corresponding to the form auf in literary
German. Perhaps this shift is idiolectal or familial. A cursory comparison of data
in Berend (1997) reveals no similar forms in the massive corpus of data collected
in the former Volga German Colonies in the Soviet Union prior to 1930, admit-
tedly the form auf only occurs in prefixed samples and not in isolation. The forms
collected by Johnson appear to be unique in Russian German varieties.
In Germany itself, the velarization of the voiceless labial fricative f is not un-
common, if not in the particular context of the Kansas examples. For instance, in
some varieties of colloquial German, we find a similar phenomenon in the envi-
ronment of a following dental stop, as in colloquial Bavarian fuchzg (literary
German fnfzig, fifty). In Schirmunskis survey of the phonology and morpholo-
gy of all modern German dialects, the process of shift of [f] to [x] in the context of
a following [t] is described in other contexts as well. This backing of the labial
fricative when followed by a voiceless dental stop is characteristic of a broad swath
of German(ic) dialects from the Netherlands across the Low German dialects
of northern Germany and extending into the Middle Franconian dialects of
Ripuarian (Cologne), Moselle Franconian (Trier, Luxembourg) and Lorraine
(Metz) (Schirmunski 1962: 402403). Some examples are in (4).
(4) after > achter behind (cf. English after)
luft > lucht air
saft > sacht soft
In addition to being part and parcel of Modern Dutch, this phenomenon has also
entered literary German with a few isolated forms: Nichte niece (cf. medieval
German niftel), Schacht shaft, Gercht rumor (cf. medieval German gerechte <
ruofen to call). An interesting side to this development is that one regular/weak
verb in Modern Dutch with a stem-final p shifts this p to ch in the past tense and
past participle, as in (5). This reflects the earlier pre-historical change of pt > ft in
Germanic:
(5) Dutch kopen kochte gekocht buy bought bought
Obviously, velarization or backing of the labial fricative f is not an isolated phe-
nomenon relegated to idiolectal errors in a Volga German settlement dialect in
western Kansas. Velar Fronting of [x] is attested for a number of languages, includ-
ing English where, for instance, one can point to the development from Old English
to Modern English in such forms as those in (6).
(6) hlahhan > laugh to laugh cf. German lachen
rh > rough rough cf. German rauh and the compound Rauchwaren
furs
where the historical [x] has become an [f] word-finally (Hickey 1984: 352). It may
also be that the Jakobson/Ladefoged distinctive feature [+ grave] shared by labials
and velars plays a role in this process and that labials and velars form a natural
class (Ladefoged 1975: 265; Campbell 1974: 54). It would be intriguing to explore
the entirety of this particular sound change in the context of Optimality Theory
perhaps a reversal of the phenomenon of Velar Fronting in Dinnsen (2011). But
that would be beyond the scope of this essay.
Nasal assimilation in unusual circumstances
Assimilation of following obstruents to nasals as well as nasalization of vowels is

quite common in the history of Germanic languages and in the modern German
dialects. A unique process of such an assimilation with subsequent nasalization of
the vowel and/or loss of the nasal is exemplified by the example in (7).
(7) kinder > kinner > kinr > kir children (singular kind)
In subject interviews in a dying speech community known as Milberger in south-
western Russell County in west central Kansas, Khramova (2011: 3738) encoun-
tered what she termed an unusual development. Expected assimilations of nasal
and obstruent occurred in forms such as those in (8).
(8) finne < finden to find
hinne < hinten behind
kfunne < gefunden found, past participle
verschtunne < verstanden understood, past participle
However, Khramova notes that complete loss of the nasal also occurs, at times
with slight nasalization of the vowel as in (9).
William D. Keel
(9) dorschtak < Donnerstag Thursday

kenr/kinr/ker/kir < Kinder children
schtenr/schter < Stnder barrel
ur < unter under
urdas < Untertasse saucer
In an additional example, the prepositional phrase hichem boum behind the tree
derives from an underlying form that could be reconstructed as *hintig, a suffixed
form of the stem *hint- ultimately yielding via assimilation of the obstruent [t] to
the nasal and realization of the final consonant as a fricative the anticipated dia-
lect form hinnich.
Khramova could not find any comparable assimilation in the modern German
dialects after a thorough review of the relevant data discussed by Schirmunski
(1962: 394ff.). Schirmunski reports no examples of intervocalic n+d assimilation
with concomitant loss of the nasal, especially preceding the suffix -er. A compa-
rable example noted by Schirmunski to those in Khramovas data would be, for
instance, Lorrainese wunner (literary German Wunder miracle). Nasalization of
the vowel and loss of n does occur, but without concomitant assimilation of the d,
as in Swabian kheend (literary German Kind child).
She also had her data examined by two experts (Allard Jongman, University of
Kansas, and Olga Bolotova, University of St. Petersburg) using Praat to determine
whether any nasalization of vowels could be confirmed in the formants of the vow-
els in the words in question. Both linguists reported independently that only slight
evidence of nasalization of the vowel could be found in a few realizations of the
words. The majority of the forms exhibited not only total loss of the nasal but also
no trace of nasalization. This total loss of nasality in these forms remains a phe-
nomenon apparently unique to a dying German variety in western Kansas.
Morphological reconfiguration of noun case in the definite article
Commonplace in the history of the Germanic languages is the reduction of noun

case morphology. Assuming eight grammatical cases for the pre-historic Indo-
European parent language, most experts reconstruct six cases for the noun phrase
in the Proto-Germanic period. Historically attested forms of the Germanic lan-
guages, such as Gothic in the 4th century or Old High German in the 8th century,
exhibit at most five cases, at this stage still partially realized in suffixation to the
noun stem, but already, especially in Old High German reducing case marking to
the accompanying demonstrative pronoun aka definite article. The noun in Mod-
ern literary German is typically classified as having four cases: nominative, accusa-
tive, dative, genitive. In addition, all nouns are assigned one of three grammatical
genders in Modern German masculine, neuter or feminine and inflect for num-
ber. Other branches of the Germanic family, in particular Afrikaans and English,
have largely abandoned any case morphology in the noun and its definite article
with the exception of the English possessive/genitive (cf. mans, womans etc.).
The modern dialects of German have largely replaced the genitive case with
other constructions and for the most part have reduced the remaining three cases
(nominative, accusative, dative) to either a nominative vs. objective system, in
which the accusative or dative case is maintained in opposition to the nominative,
or a common case vs. dative system, with the nominative and accusative cases hav-
ing merged as they largely have except for masculine nouns in literary German
(cf. Shrier 1965).
For the Volga German dialects, which derive primarily from Hessian and
Palatine dialects in the West Middle German area of Germany, we can assume
historically a three-case starting point: nominative, accusative, dative and a par-
adigm for the definite article as in (10) (using forms based on literary German).
(10) masculine neuter feminine plural
nominative der das die die
accusative den das die die
dative dem dem der den
In studying the Volga German dialect of Victoria, Kansas, in Ellis County, Keel
(1994, 2004) believes that a unique reconfiguration of this case system for the def-
inite article has occurred. Rather than reducing the number of forms or case op-
positions entirely, it appears that a new prepositional case has emerged and has
resulted in the paradigm in (11).
(11) masculine neuter feminine plural
nominative der des die die
accusative den des die die
dative den den der die/denne
prepositional den den die/der die/denne
First of all, the feminine and plural forms have remained largely intact, with the
exception of the alternate analogous die for the plural dative and the extended
form denne in the plural dative case (this extended form is found in a number of
German dialects today). The pattern of usage remains essentially unchanged from
the assumed parent variety. The forms representing the prepositional case reflect
the historical distinction of case usage following certain prepositions the accusa-
tive form is used following those prepositions expected to have an accusative ob-
ject, while dative forms follow prepositions expected to have a dative object. The
alternation in the plural forms does complicate this pattern and may well signal an
William D. Keel
impending total collapse of case distinctions in the plural definite article. Unfortu-
nately, with the imminent demise of this variety, that future development cannot
be tracked. It should also be noted that other aspects of case development such as
the absence of case distinctions following an indefinite article as seen in example
(1) below e gleener abbel are not treated here (see Keel 2004 for a discussion of
the complete noun phrase morphology in this variety). Examples are in (12).
(12) a. nominative feminine and plural
die jung fraa hot e gleener abbel gess
the young woman ate a little apple
die frohe kinner sin aus der schul gschprung
the happy kids ran out of the school
b. accusative feminine and plural
ich gleich die weiss seef
I like the white soap
die soll die Gleider fertich nehe for dei Modder
she should finish sewing the clothes for your mother
c. dative feminine and plural
mir han der gut fraa e bissje kees geb
we gave the good woman a little cheese
mir han die/denne menner gholf (alternate forms die/denne)
we helped those men
d. prepositional feminine (accusative and dative usage maintained dis-
tinct)
ich sinn . . . iewer die viese unn in den veetsfeld gfahr (accusative an-
ticipated)
I drove . . . across the pasture and into the wheat field
vas fore gleene fegeljer sitse dort owe uff der gleen vand (dative antici-
pated)
what kind of little birdies are sitting there on the little wall
e. prepositional plural (mixed usage)
die duhe mol was for die arme leit (accusative anticipated and realized)
are they going to do something for the poor people
die buwe han mit die menner gschaaft (dative anticipated and not real-
ized)
the boys worked with the men
was soll ich duhe mit denne fette kieh (dative anticipated and realized)
what should I do with those fat cattle
For the masculine definite article, we see a reduction in the dative article to den,
analogous to the accusative form, with otherwise no other changes in the rest of
the paradigm for the masculine forms, as in (13).
(13) a. nominative masculine
der veets vaar gut des johr
the wheat was good this year
b. accusative masculine, including accusative prepositions
die schloose han den gantse veets ferschlah
the hail destroyed the entire wheat crop
schier kohle in den owe (following accusative preposition)
put coals into the stove
c. dative masculine, including dative prepositions
der hot den alde mann e neies buch gebrung
he brought the old man a new book
do is mol was los mit dort den braune gaul (following dative preposi-
tion)
something is wrong with that brown horse
Neuter nouns on the other hand also exhibit the dative form den as their definite
article but extend its usage to all instances following a preposition, rather than
maintaining the alternation of the parent variety and maintained by feminine
forms between accusative and dative. Examples are given in (14).
(14) a. nominative neuter
des gleene beemje hot fill ebbel
the little tree has a lot of apples
b. accusative neuter
ven hot der des neie schtickelje fertsehlt
who did he tell the new story to
c. dative neuter
ich han den scheene meedje blume geb
I gave the pretty girl flowers
d. prepositional neuter (no distinction between former accusative and
dative)
mir sin in den kalde vasser gfall (accusative anticipated)
we fell into the cold water
der gude, alde mann is dorch den eis gebroch (accusative anticipated)
the good, old man fell through the ice
der hot in den kalde vasser gsots (dative anticipated)
he sat in the cold water
William D. Keel
Following the rather common reduction of the final -m in the historical dative
forms for the masculine and neuter definite articles (dem > den) as well as the re-
duction of the vowel in the neuter nominative/accusative definite article (das >
des) and the introduction of the plural dative alternate denne (a suffixed demon-
strative pronoun), a new paradigm for the definite article/noun case morphology
has emerged. As in (15), the new system follows a two-case distinction for each
noun gender in the singular and instability in the plural definite article.
(15) Masculine: nominative case = der vs. all other instances = den
Neuter: nominative and non-prepositional accusative = des
vs. dative and all-prepositional usage = den
Feminine: non-dative = die vs. dative = der
Plural: all instances = die
Alternate dative/prepositional = denne
Neither Shrier (1965), nor Schirmunski (1962) hint at the possibility of the devel-
opment of this type of prepositional case for masculine and neuter nouns in the
modern German dialects in their comprehensive surveys. Research on varieties of
German in Russia as reported in several essays in a survey of those dialects (Berend
and Jedig 1991) also yields no comparable phenomena. Once again, the dying
German varieties on the High Plains of western Kansas offer another unusual lin-
guistic innovation for our consideration.
Semantic transfer: ich gleich die veiss seef I like white soap
Finally, I would like to comment on a lexical phenomenon that has puzzled me for
nearly four decades. It is the ubiquitous German-American verb gleiche(n) to like.
It is common throughout the German-American dialect landscape, whether in
Pennsylvania, Texas, Wisconsin or Kansas or anywhere else German settlements
have been established in the United States. This verb in recognizable morphologi-
cal shape is found in all types of these German-American varieties, whether from
northern Germany as in the Low German variety of Bremen, Kansas, from
Switzerland in the south as in the High Alemannic variety of New Glarus, Wiscon-
sin, or from the Middle German dialects with numerous representatives in the
United States from Pennsylvania to Kansas and from Texas to Wisconsin. This
verb appears to have been fully adopted from American English usage and fully
incorporated into the verb morphology of the given dialect. One of the earliest
citations of this usage was documented by Francis Lieber for Pennsylvania German
in the 1830s (Reichmann 1945: 27). But the question remains, how did this appar-
ent semantic transfer occur in its multifaceted realizations?
For the better part of a century, linguists beginning with Marcus Lambert
(1924: 66), Leonard Bloomfield (1933: 462) and Paul Schach (1951: 258259) have
puzzled over this instance of apparent semantic transfer. As Schach (258) notes:
Among the numerous semantic loans in Pennsylvania German there are several
which reveal an almost startling shift in meaning. Thus, for example, the German
verb gleichen means to be similar, to be like; but Pennsylvania German gleiche has
the completely different and unrelated meaning to like, to be fond of.
Both Lambert and Bloomfield explain it as the derivation of a verb in the German-
American dialects on the basis of the common adjective gleich similar, like, equal
which then parallels the usage of American English to like in German-American
dialects.
Schach cannot accept that argument, calling it puzzling and non-explanatory
since a verb derived from an adjective meaning similarity cannot then lead to a
verb meaning fondness (258). Schach would rather attribute this to bilingual
confusion due to the similar phonetic forms of English to like and to be like
which then led monolingual speakers of the given German-American dialect to
create the puzzling verb, including both regular and irregular morphology in the
dialects. For instance, some speakers use the regular (weak) past participle for the
verb gleiche > gegleicht to like > liked; other use the irregular (strong) past parti-
ciple gleiche > gegliche to like > liked. Schach also notes that the use of a simple
verb such as gleiche in the American English sense of to like brings with it syntac-
tic and morphological simplicity for the expression of liking something. However,
this appears to be speculation on Schachs part. Schach (258259) is forced to con-
clude that
interestingly enough, German gleichen has undergone the same striking shift in
meaning quite generally in this country [U.S.]. I have heard it used in this sense
in New York, Chicago, Philadelphia, Reading (Pa.), and Lincoln (Nebr.), as well as
in many rural communities; and Professor Arndt, of the Concordia Theological
Seminary in St. Louis, informs me that gliken to like is current among speakers of
Low German throughout the Midwest.
Needless to say, in our studies of the German settlement varieties in Kansas and
Missouri, forms of gleiche/glike/gliche to like occur with high frequency. No one
has adequately explained this apparent transfer from American English to these
various dialects across the entire American landscape. Some have sought in vain to
try to find an internal source for this verb in historical dictionaries of German and
its dialects. Isolated reports of similar usage in southeastern German speech-islands
in Hungary and Carinthia or perhaps the Austrian Burgenland have never been
substantiated. Kaull, however, in his study of the Volga German dialects in Kansas
claims that the explanation of semantic transfer from English may not be correct
William D. Keel
(1996: 69). He uses an example from a dialect in Baden die sau hat mir glei gegliche
I liked the pig immediately to claim that the Volga Germans still use gleiche with
that same meaning (70). He then argues that perhaps the influence of English to
like supported and helped preserve this meaning in the Volga German dialects. At
the same time this meaning has been lost in German and its modern dialects.
Unfortunately, Kaulls explanation has numerous weaknesses. What direct
connection is there between a modern dialect in Baden and those of the Volga
Germans who left Germany in the 1760s, lived along the Volga until the 1870s and
then migrated to new settlements in western Kansas? One could claim that other
High German dialects might have had a verb with a similar meaning historically.
Another problem is the morphology of the cited gleiche. In the Baden dialect the
verb requires the human liker to be in the dative case; the thing liked is in the
nominative case. This is the classic case of 7 Up and its slogan introduced in 1936:
You like it, it likes you. Kaull would now need to explain how the verb gleichen
changes its syntactic requirements from requiring a dative human object (it likes
you) to a nominative human subject (you like it) to achieve the structure we now
find in the German-American dialects. This remains a truly puzzling develop-
ment, not only in the Kansas German varieties, but throughout the German-
American dialects.
Final remark
With this excursus into some puzzling issues in the study of German settlement
varieties in Kansas, I hope to encourage others to examine these phenomena and
expand on the preliminary explanations presented here. I would also like to ex-
press my deepest appreciation for the inspiration, advice and mentoring that
Daniel Dinnsen provided me as a doctoral student in Bloomington, Indiana, near-
ly forty years ago. I congratulate him on his extraordinary career and wish him the
very best for his retirement.
References
Aitchison, J. 2001. Language Change: Progress or Decay?, 3d edn. Cambridge: CUP.

Baerg. M. 1960. Gnadenau Low German: A Dialect of Marion County, Kansas. PhD disserta-
tion, University of Chicago.
Bender, J. 1970. Die getrennte Entwicklung gleichen niederdeutschen Sprachgutes in Deutschland
und Nebraska. PhD dissertation, University of Nebraska.
Berend, N. 1997. Wolgadeutscher Sprachatlas aufgrund der von Georg Dinges 19251929 gesam-
melten Materials.Tbingen: A. Francke.
Berend, N. & H. Jedig (eds.). 1991. Deutsche Mundarten in der Sowjetunion: Geschichte der
Forschung und Bibliographie. Marburg: N. G. Elwert.
Bloomfield, L. 1933. Language. New York NY: Holt, Rinehart and Winston.
Campbell, L. 1974. Phonological features: Problems and proposals. Language 50(1): 5265.
Carman, J. N. 1962. Foreign Language Units of Kansas, I: Historical Atlas and Statistics. Lawrence
KS: University of Kansas Press.
Dinnsen, D., et al. 2011. On the interaction of velar fronting and labial harmony. Clinical Lin-
guistics & Phonetics 25(3): 231251.
Hickey, R. 1984. On the nature of labial verlar shift. Journal of Phonetics 12: 345354.
Johnson, D.C. 1994. The Volga German Dialect of Schoenchen, Kansas. PhD dissertation, Uni-
versity of Kansas.
Kaull, P. 1996. Untersuchungen zum Deutsch in Kansas: Migration und Entwicklung des
Wolgadeutschen in Ellis und Rush Counties. MA thesis, University of Mannheim.
Keel, W. 1994. Reduction and loss of case marking in the noun phrase in German-American speech
islands: Internal development or external influence? In Sprachinselforschung: Eine Gedenk-
schrift fr Hugo Jedig, N. Berend & K. Mattheier (eds), 93104. Frankfurt: Peter Lang.
Keel, W. 2004. A grammatical sketch of the Victoria (Herzog) dialect of Kansas Volga German.
In The Volga Germans of West Central Kansas: Aspects of Their History, Politics, Culture and
Language, W. Keel et al. (eds), 217239. Lawrence KS: Max Kade Center for German-
American Studies.
Khramova, M. 2011. The Volga German Dialect of Milberger, Kansas. PhD dissertation, Univer-
sity of Kansas.
Ladefoged, P. 1975. A Course in Phonetics. New York NY: Harcourt, Brace and World.
Lambert, M.B. 1924. A Dictionary of the Non-English Words of the Pennsylvania-German Dialect.
Lancaster PA: Pennsylvania German Society.
Lunte, G. 2007. The Catholic Bohemian German Dialect of Ellis County, Kansas: A Unique Bavar-
ian Dialect. Frankfurt: Peter Lang.
Mattheier, K. 2003. Sprachtod: berlegungen zur Entwicklungsdynamik von Sprachinseln. In
German Language Varieties Worldwide. Internal und External Perspectives, W. Keel & K.
Mattheier (eds.), 1331. Frankfurt: Peter Lang.
Meindl, J. 2009. Language Use in an Older Order Amish Community in Kansas. PhD disserta-
tion, University of Kansas.
Post, R. 1990. Pflzisch: Einfhrung in eine Sprachlandschaft. Landau: Pflzische Verlagsanstalt.
Putnam, M. & Weis, B. 2004. Recessive features or phonological nuances? Paper presented at the
Annual Symposium of the Society for German-American Studies, New Ulm MN, April
2225.
Reed, C. 1971. The dialectology of American colonial German. In The German Language in
America: A Symposium, G. Gilbert (ed.), 313. Austin TX: University of Texas Press.
Reichmann, F. Francis Lieber. 1945. Pennsylvania German Dialect. The American-German Re-
view 11: 2427.
Schach. P. 1951. Semantic borrowing in Pennsylvania German. American Speech 26(4): 257267.
Schirmunski, V. 1930. Sprachgeschichte und Siedelungsmundarten, I. Germanisch-romanische
Monatsschrift 18: 113122.
Schirmunski, V. 1962. Deutsche Mundartkunde: Vergleichende Laut- und Formenlehre der
deutschen Mundarten,Wolfgang Fleischer (transl). Berlin: Akademie Verlag.
William D. Keel
Seeger, G.S. 2006. Socio-Economic Influence on Low German in North-Central Kansas: From
Immigrant Language Lost to Heritage Language Revived. PhD dissertation, University of
Kansas.
Shrier, M. 1965. Case systems in German dialects. Language 41(3): 420438.
Wildfeuer, A. 2012. Sprachstand Sprachenkontakt und Mehrsprachigkeit Sprachverlust:
Deutschbhmische Siedlungen im englischsprachigen Raum. Habilitationsschrift. Univer-
sity of Regensburg.
U.S. Census Bureau. 20072011. American Community Survey 5-Year Estimates. <http://fact-
finder2.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=bkmk> (23 Feb-
ruary 2013).
section 4
Theoretical advances in the field

Constraint-based approaches
Dinnsen has long been known for his application of cutting-edge work in phono-
logical theory to the study of development. He has identified the implications of
different theories for the analysis of developing phonologies, and he has used ac-
quisition as a test case for many theories (Chin & Dinnsen, 1992; Dinnsen, 1984,
1996, 1998; Gierut, Cho & Dinnsen, 1993). Since the advent of Optimality Theory
in the 1990s, Dinnsen has been one of its strongest advocates, repeatedly illustrat-
ing important theoretical concepts in the domain of acquisition (Barlow & Dinns-
en, 1998; Dinnsen, 2002; Dinnsen & Barlow, 1998; Dinnsen & Farris-Trimble,
2009; Dinnsen & Gierut, 2008 and the papers therein; Dinnsen, Gierut & Farris-
Trimble, 2010; Dinnsen, Green, Gierut & Morrisette, 2011; Dinnsen, Green, Mor-
risette & Gierut, 2011; Dinnsen & McGarrity, 2004; Dinnsen, McGarrity, OConnor
& Swanson, 2000; Dinnsen & OConnor 2001a,b; Dinnsen, OConnor & Gierut,
2001; Morrisette, Dinnsen & Gierut, 2003). His work has always followed the prin-
ciple that any theoretical account of language must also be able to account for how
language is acquired.
This final section pays tribute to Dinnsens contributions to phonological the-
ory, and especially constraint-based theories.
McGarrity discusses syllabic factors that influence primary and secondary
stress patterns (The role of onsets in primary and secondary stress patterns). She
couches this analysis within an optimality theoretic framework and contrasts pat-
terns in fully-developed languages with those in acquisition.
Farris-Trimble re-analyzes Amahls acquisition of marked structures, arguing
that Amahls seemingly disparate acquisition patterns may all have a common goal
(A faithfulness conspiracy: The selection of unfaithful mappings in Amahls gram-
mar). She presents this as a conspiracy of faithfulness and discusses the implica-
tions for constraint-based theories.
Finally, Green and Davis demonstrate superlinear constraint ordering in their
examination of Bambara (Superadditivity and limitations on syllable complexity
in Bambara words). They provide an account of the phenomenon in a fully-
developed language and discuss how such a pattern may arise over the course of
acquisition.
These papers reflect the latest developments in phonological theory, and re-
mind the reader that studies of developing phonologies often serve to push the
theory forward.
References
Barlow, J.A. & Dinnsen, D.A. 1998. Asymmetrical cluster development in a disordered system.
Chin, S.B. & Dinnsen, D.A. 1992. Consonant clusters in disordered speech: Constraints and
correspondence patterns. Journal of Child Language 19: 259285.
M. Elbert, D.A. Dinnsen, & G. Weismer (Eds.), Phonological Theory and the Misarticulating
Child (pp. 517). Rockville, MD: ASHA.
Dinnsen, D.A. 1996. Context-sensitive underspecification and the acquisition of phonemic con-
trasts. Journal of Child Language 23: 5779.
Dinnsen, D.A. 1998. On the organization and specification of manner features. Journal of Lin-
guistics 34: 125.
Dinnsen, D.A. 2002. A reconsideration of childrens phonological representations. In B.
Skarabela, S. Fish & A. H.-J. Do (Eds.), Proceedings of the 26th Annual Boston University
Conference on Language Development (pp. 123). Somerville, MA: Cascadilla Press.
Dinnsen, D.A. & Farris-Trimble, A.W. 2009. Developmental shifts in phonological strength rela-
tions. In K. Nasukawa & P. Backley (Eds.), Phonological Strength (pp. 113149). New York:
Mouton de Gruyter.
Dinnsen, D.A., Gierut, J.A., & Farris-Trimble, A.W. 2010. Comparative markedness and induced
opacity. Language Research, 46: 138.
Dinnsen, D.A., & Gierut, J.A. 2008. Optimality Theory, Phonological Acquisition, and Disorders.
London: Equinox Publishing Ltd.
Dinnsen, D.A., Green, C., Gierut, J.A. & Morrisette, M.L. 2011. On the anatomy of a chain shift.
Journal of Linguistics 47: 275299.
Dinnsen, D.A., Green, C., Morrisette, M.L. & Gierut, J.A. 2011. On the interaction of velar front-
ing and labial harmony. Clinical Linguistics & Phonetics 25: 231251.
Dinnsen, D.A., & McGarrity, L.W. 2004. On the nature of alternations in phonological acquisi-
tion. Studies in Phonetics, Phonology and Morphology 10: 2341.
Dinnsen, D.A., McGarrity, L.W., OConnor, K.M., & Swanson, K.A. 2000. On the role of sympa-
thy in acquisition. Language Acquisition 8: 321361.
Dinnsen, D.A., & OConnor, K.M. 2001a. Typological predictions in developmental phonology.
Dinnsen, D.A., & OConnor, K.M. 2001b. Implicationally-related error patterns and the selec-
tion of treatment targets. Language, Speech and Hearing Services in Schools 32: 257270.
Dinnsen, D.A., OConnor, K.M., & Gierut, J.A. 2001. The puzzle-puddle-pickle problem and the
Gierut, J.A., Cho, M.-H., & Dinnsen, D.A. 1993. Geometric accounts of consonant-vowel inter-
actions in developing systems. Clinical Linguistics & Phonetics 7: 219236.
Morrisette, M.L., Dinnsen, D.A., & Gierut, J.A. 2003. Markedness and context effects in the ac-
quisition of place features. Canadian Journal of Linguistics 48: 329355.
The role of onsets in primary and secondary
stress patterns
Laura W. McGarrity
University of Washington
It is found that languages with onset-sensitive stress, in which the location of stress
is dependent upon the presence or quality of an onset, exhibit the same patterns
of symmetrical and asymmetrical stress assignment as languages with rime-
based weight (McGarrity 2003). An OT analysis of languages with asymmetrical
patterns of onset-sensitive stress (in which primary stress behaves independently
of secondary stress) is presented which appeals to primary-stress-specific versions
of several prominence-enhancing positional markedness constraints that require
stressed syllables to have onsets of low sonority. The relative paucity of fully
developed languages with onset-sensitive stress is contrasted with the fairly
common tendency in childrens early words for low-sonority onsets in unstressed
(deleted) syllables to relocate to the syllable bearing stress.
1. Introduction
This paper expands on the wide variety of phonological phenomena known to

interact with primary and secondary stress assignment in such a way as to produce
both symmetrical and asymmetrical patterns. In some languages, primary and
secondary stresses behave identically or symmetrically with respect to a particular
process. For example, in Chimalapa Zoque (Knudson 1975), both primary- and
secondary-stressed open syllables undergo vowel lengthening (e.g., [:.to.:.pith]
if he had spoken). In other languages, only primary stress interacts with a process
while secondary stress does not. In these languages, the stress pattern is asym-
metrical. For example, in Wargamay (Dixon 1981), only primary stressed syllables
undergo vowel lengthening while secondary stressed syllables do not (e.g.,
[g:awlu] freshwater jewfish).
McGarrity (2003) showed that asymmetrical stress patterns can be accounted
for within Optimality Theory (OT) by appealing to primary-stress-specific position-
al markedness or augmentation constraints (constraints referring to phonologically
prominent or strong positions; Beckman 1998; de Lacy 2001; Smith 2005). Primary
Laura W. McGarrity
stressed syllables, being strongest of the strong, make a logical focus for positional
constraints. A crucial assumption is that constraints may not refer exclusively to
secondary stress, as they are phonologically weak. It is this asymmetry in the formu-
lation of the stress constraints that allows for asymmetrical stress patterns.
Almost all of the phenomena discussed in McGarrity (2003) that were shown
to produce both symmetrical and asymmetrical patterns of stress (i.e., stressed
syllable lengthening effects, quantity-sensitivity, and sonority-sensitivity) were
rime-based. In this paper, the role of onsets in primary and secondary stress as-
signment will be examined. In 2, some of the ways in which languages show
sensitivity to onsets in stress assignment will be introduced. In 3, languages with
onset-sensitivity that display both symmetrical (3.1) and asymmetrical (3.2)
patterns of stress are presented. 4 concludes the paper by discussing how the
comparative rarity of onset-sensitive stress in adult languages contrasts with the
fairly common tendency for childrens early words to show sensitivity to onsets in
stressed syllables.
2. Onset-sensitivity in stress assignment: Background
Onsets have traditionally been assumed to be inert for prosodic phenomena that
depend upon syllable weight, such as the assignment of stress to heavy syllables in
quantity-sensitive languages (Goedemans 1998; Gordon 1999; Hayes 1989; Hyman
1985). In such analyses, syllable weight is determined by the elements in the rime:
the nucleus and (on a language-specific basis) the coda. Onsets are argued to be
weightless. However, an increasing number of languages have been analyzed as
having stress systems in which onsets play a determining role in the location of
stress (see Davis 1988; Gordon 2005; and Topintzi 2010 for surveys of onset-sensi-
tive languages). In this section, I present examples of languages with various kinds
of onset-sensitive stress. Different patterns of onset-sensitivity can be distinguished
based on whether it is the presence (vs. absence) of an onset or its quality that in-
teracts with stress assignment as well as whether it is the nature of the onset that
drives the stress pattern or vice versa.
2.1 Sensitivity to the presence of an onset
A fair number of languages have been found to have similar patterns whereby
stress is sensitive to the presence of an onset. For instance, in Banaw (Arawan
language of Brazil; Buller, Buller & Everett 1993), primary stress falls on the first
mora of the first syllable in consonant-initial words (1a). In vowel-initial words,
however, primary stress shifts rightward, falling on the first mora of the second
The role of onsets in primary and secondary stress
syllable (1b). That is, stress in Banaw is onset-sensitive, falling on the first mora in
the word that is preceded by an onset.1
(1) a. Consonant-initial words b. Vowel-initial words
tme foot ida to marry
mkar cloth owri one
ttikne hair ufbun I drink
While Banaw shifts primary stress away from its default position to fall on an
onsetful syllable elsewhere, other languages maintain the default location of stress
but alter the suboptimal onsetless syllable by creating an onset. For example, in
Ainu (Topintzi 2010, citing Shiraishi 1998 and Tamura 1996), stress must fall with-
in a two-syllable window of the beginning of the word. If the first syllable is heavy
((C)VC), it receives stress, e.g., [p.to] rain. Otherwise, stress falls on the second
syllable, e.g., [a.p] entrance.
When morpheme concatenation creates vowel sequences within the two-
syllable stress window such that default stress would fall on an onsetless high vow-
el, glide formation occurs, changing the high vowel into a glide to create an onset
for the stressed syllable. However, glide formation only occurs when it would pro-
vide an onset for an otherwise onsetless stressed syllable (2a). Outside of the two-
syllable stress window, glide formation does not apply (2b).
(2) a. Ainu glide formation applies in stressed syllables
/ko-i-o-map/[koymap] (*ko..o.map) to love anothers child
b. Ainu glide formation does not apply in unstressed syllables2
/i-ko-i-o-map/[ikiyomap] (*i.ko.yo.map) to love anothers children
Topintzi (2010) describes languages like Banaw as exhibiting onset-on-stress
effects; the presence of an onset influences the placement of stress, shifting it away
from its default position. Ainu, in contrast, is described as having stress-on-onset
effects; the presence of stress on an otherwise onsetless syllable drives the forma-
tion of an onset.
This preference for stressed syllables to have onsets relates to the notion of
prominence. Stressed syllables, which are themselves prominent acoustically and
1. Other languages with this pattern of initial stress in C-initial words but peninitial stress in
V-initial words include Alyawarra (Yallop 1977), Western Aranda (Strehlow 1942), Iowa-Oto
(Robinson 1975), as well as a number of areal Cape York and Arandic languages in Australia (see
Blevins 2001 for references).
2. Topintzi (2010: 68) (citing Shiraishi, p.c.) is careful to point out that the glide in the pho-
netic form in (2b) is not the result of the glide formation seen in (2a); rather, it results from an
independent rule of glide insertion that operates to resolve hiatus outside of the two-syllable
stress window.
Laura W. McGarrity
perceptually, frequently alter to become more prominent (e.g., by lengthening

their vocalic or consonantal elements, by altering their vowels to become more
sonorant, etc.). Further, stress is often attracted to the most prominent syllable(s)
in the word (i.e., those bearing long vowels, high-sonority vowels, etc.). Syllables
with onsets are preferred to onsetless syllables because they are perceptually more
prominent (see Gordon 2005; Smith 2005). An onset consonant provides a con-
trast to the following vowel, thereby enhancing perception. Languages in which
stressed syllables are sensitive to the presence of an onset, then, are augmenting
their inherent perceptual prominence.
2.2 Sensitivity to the quality of an onset
While some languages have stress systems that show sensitivity to the presence of
an onset, other languages have stress assignment that is sensitive to the type or
quality of an onset in a syllable. Karo (Tup language of Brazil; Gabas 1999) exhib-
its a stress attraction pattern of this type. In Karo, the quality of the onsets in the
penultimate and final syllables can determine stress placement. Stress falls on the
final syllable in Karo if the onset is voiceless (3a). (In such cases, the voiceless onset
predictably lengthens, creating a geminate.) However, if the onset of the final syl-
lable is a voiced obstruent and the penultimate syllable contains a voiceless onset,
then stress shifts leftward onto the penult (3b).3
(3) a. Primary stress on final
mop:k bird (sp.)
mat:k palm tree (sp.)
pak: fish (sp.)
b. Penultimate stress if final has voiced obs. onset and penult has voice-
less
pb foot
tgp bow
icg quati (sp.)
Stressed syllables in Karo prefer onsets that have no [voice] feature over those that
do. As with the onset-on-stress languages discussed above, this preference for
voiceless onsets over voiced onsets is related to issues of prominence. That voice-
less consonants are preferred as the onsets of stressed syllables is due to their lower
3. Sonorant consonants pattern with voiceless obstruents in being tolerated as onsets in final
syllables (e.g., [kj] crab) and being preferred as onsets in the penult over voiced obstruents
onsets in final syllables (e.g., [mga] mouse). Topintzi (2010) argues that voiced obstruents are
specified for [voice] while voiceless obstruents and sonorants lack [voice]. Therefore, Karo stress
preferentially falls on syllables with onsets lacking the feature [voice].
sonority (Gordon 2005; Smith 2005). A stressed syllable with a voiceless onset cre-
ates the largest sonority difference between the onset and a following nucleus,
thereby enhancing phonetic and perceptual prominence.4
While the onset-on-stress pattern of Karo and the stress-on-onset pattern of
Ainu are representative of the kinds of onset-stress interactions that have been
discussed in the literature, both of these languages only have one stress per word,
so the data only show how onsets influence the placement of primary stress. In the
following section, I examine data from several onset-sensitive languages that have
multiple stresses. I show that they are comparable to languages with rime-based
weight in exhibiting both symmetrical and asymmetrical patterns of stress-onset
interactions. OT analyses are presented to account for the patterns.
3. Patterns of primary and secondary stress in onset-sensitive languages
In this section, I look at five onset-sensitive languages with multiple stresses. In

3.1, I discuss two languages with symmetrical patterns of onset-stress interac-
tions: Jma (3.1.1) and Dutch (3.1.2). In 3.2, I discuss three languages whose
stress systems display asymmetrical patterns of onset-sensitivity: Arabela (3.2.1),
Tmpisa Shoshone (3.2.2), and Niuafoou (3.2.3). For each pattern, languages with
both onset-on-stress and stress-on-onset effects are discussed.
3.1 Symmetrical patterns of primary and secondary stress
3.1.1 Onset-on-Stress language: Jma

Jma is a Tup language spoken in Brazil. The data presented here come from
Abrahamson and Abrahamson (1984). Stress in Jma follows an alternating pat-
tern, counting from the end of the word. Primary stress typically falls on the final
syllable, with secondary stresses falling on every other syllable before the primary
stress (4). (The transcriptions shown here are phonemic and do not reflect precise
allophonic variants.)
(4) tupa thunder5
krer red rat
gja rat
4. See Topintzi (2010) who argues against a prominence-based analysis of Karo in favor of a
moraic-based analysis. The question of whether a prominence-based or moraic analysis better
accounts for onset-sensitive stress patterns is a complex one and is beyond the scope of this
discussion. I refer the reader to Topintzi (2010) for arguments therein.
5. I would like to thank Clarissa Surek-Clark for her help in providing English translations of
the original Portuguese.
Laura W. McGarrity
In certain words, however, primary stress falls on the penult. According to

Abrahamson and Abrahamson (1984:158), vowels in sequence belong to separate
syllables, yielding onsetless syllables. In words where the final syllable is onsetless,
primary stress shifts to the penult (5).6
(5) [pi.r.i] piranha
[m.a.k.i] type of food
[pe.j.ko.p.a] bird (sp.)
This suggests that primary stress assignment in Jma is sensitive to the presence of
an onset, as it is in Banaw. Smith (2005) argues that sensitivity to the presence of
an onset can captured by appealing to the prominence constraint Onset/, which
demands that stressed syllables have onsets. By ranking this above the stress place-
ment constraint that requires final primary stress (AlignHd-R: Align the head
of a PWd with the right edge of the PWd; Zoll 1997), the stress attraction pattern
can be accounted for, as shown in (6).
(6) Primary stress shifts onto onsetful penultimate syllable: [piri] piranha
/pirai/ Onset/ AlignHd-R

a. pi(r.i) *
b. pi(ra.) *!
That Onset/ is defined generally for all stressed syllables in Jma is evident when
the secondary stress pattern is examined. As seen in (7), secondary stress shifts
leftward away from an onsetless syllable in the regular alternating pattern to fall on
a syllable with an onset.
6. It should be noted that Topintzi (2010: 60) calls Abrahamson and Abrahamsons (1984)
analysis into question, saying that the Jma data may be reanalyzed as having diphthongs rather
than heterosyllabic vowel sequences. It would then demonstrate a basic weight-sensitive system
in which diphthongs are heavy, attracting stress. But there are reasons to accept Abrahamson
and Abrahamsons analysis. In the forms cited throughout their description, seven different
vowel sequences occur: [i, ia, ea, ie, a, , ui]. If these were diphthongs rather than vowel se-
quences, the size of the vowel inventory would more than double. Beyond that, however, there
are forms in which vowel sequences fail to receive stress, calling into question the assumption
that such sequences must be heavy. There is one vowel sequence that seems to be a true diph-
thong in Jma. It is transcribed as [ai] with the second vowel superscripted, distinguishing it
from heterosyllabic vowel sequences. Further, it behaves differently, acting as a tautosyllabic
sequence, as can be seen in a word like [gamip] he was very late (*[gmaip]; cf. [k..m]
corn drink). Here, the diphthong is skipped over in the alternating secondary stress pattern,
failing to attract stress as one would expect if it were heavy.
(7) [m.e.ti.wp], *[mi..ti.wp] salt container

The regular alternating pattern would assign stress to the second syllable, yielding
*[mi..ti.wp]. Instead, secondary stress shifts leftward, skipping a syllable, to fall
on the initial onsetful syllable. Unfortunately, this is the only form with the right
sequence of syllables to illustrate the shifting of secondary stresses word-medially.
However, there is additional evidence that suggests that secondary stresses avoid
onsetless syllables. In odd-parity words bearing final stress, an alternating second-
ary stress would typically fall on the initial syllable. However, this only holds if the
initial syllable has an onset (8a). If the initial syllable is onsetless, secondary stress
is typically not assigned, yielding a stress lapse (8b).7
(8) a. Secondary stress on initial CV
pku pacu fish
ttu small terrapin
kunt young lady
b. No secondary stress on initial V
ahaw not yet
omo he is teaching
nmuth rooster
Avoidance of secondary stress on an initial V is actually expected by the undomi-
nated ranking of Onset/ above the general stress constraints FtBin (which re-
quires that feet be disyllabic) and Parse- (which requires that syllables be parsed
into feet) as shown in (9).
(9) No secondary stress on initial V, e.g., [nmuth] rooster
/nmthu/ Onset/ FtBin Parse-

a. n(mut)(h) * **
b. (na)(mut)(h) **! *
c. (n)(mut)(h) *! *
7. This pattern is not exceptionless. Of the 165 words that are marked for stress in Abrahamson
and Abrahamsons data, 65 have initial syllables that would be expected to bear an alternating
secondary stress. Of the 26 of these that have an initial CV, 25 are marked for initial stress. Of
the remaining 39 words that have an initial onsetless syllable, 23 have no initial secondary stress
marked but 16 do. At the very least, secondary stress optionally patterns with primary stress in
avoiding onsetless syllables.
Laura W. McGarrity
Stressing the initial onsetless syllable, as in (c), incurs a fatal violation of undominated
Onset/. Avoiding the onsetless syllable by shifting secondary stress rightward by
forming a degenerate, single-syllable foot as in (b), incurs a fatal violation of FtBin.
This leaves the candidate with only a single secondary stress as the optimal form.
In Jma, primary and secondary stresses behave symmetrically, such that both
avoid onsetless syllables by shifting away from their default position to fall on an
onsetful syllable to its left. This pattern falls out from ranking Onset/, general-
ized to require that all stressed syllables begin with an onset, above the constraints
responsible for the default stress pattern.
3.1.2 Stress-on-Onset language: Dutch

Examples of languages with stress-on-onset patterns are not as robust as those
with onset-on-stress patterns. It is difficult to find many cases involving multiple
stresses to determine how the different stresses might interact with onsets. That
said, Dutch has a stress-on-onset pattern in which primary and secondary stresses
may behave symmetrically with respect to onset-sensitivity. Though discussion on
this pattern is sparse, I present this case here for completeness.
In Dutch (Booij 1995), onsetless syllables arising from vowels in hiatus (where
V1 is [a]) are generally tolerated, e.g., /xas/ [x:.s] chaos. However, the lack of
an onset is not tolerated if the syllable bears stress. If stress falls on an onsetless
syllable, a glottal stop is epenthesized to serve as onset, e.g., /arta/ [a.r.ta]
aorta. This preference for an onset at least optionally extends to syllables with
secondary stress, e.g., /farao/ [f:.ra.:] pharaoh (Topintzi 2010:64, citing per-
sonal communication with Carlos Gussenhoven).
The same Onset/ constraint presented in 3.1.1 for Jma can account for the
Dutch data as well. Defined generally, it requires that stressed syllables of any kind
must have onsets. Unlike with Jma, however, this constraint would be ranked
high with the general stress placement constraints for Dutch, with Dep, the con-
straint forbidding epenthesis, low-ranked.
(10) Stressed-syllable onset epenthesis
/xs/ Onset/ Dep

a. x:.s
b. x:.s *
/fra/ Onset/ Dep
c. f:.ra.: *
d. f:.ra.: *!
In a word with only primary stress, candidate (b) with an epenthesized onset fa-
tally violates Dep, allowing candidate (a) without epenthesis and with stress on an
onsetful syllable to win. In a word with multiple stresses, candidate (d) with stress
on an onsetless syllable would be ruled out by high ranking Onset/, leaving can-
didate (c), with an epenthetic onset in a stressed syllable, as optimal.
3.2 Asymmetrical patterns of primary and secondary stress
Of the languages with onset-sensitivity that allow multiple stresses per word, far
more have asymmetrical than symmetrical patterns of primary and secondary
stress. I will discuss three languages with asymmetrical onset-sensitive stress:
Tmpisa Shoshone and Arabela, which have onset-on-stress patterns, and
Niuafoou which exhibits a stress-on-onset pattern.
3.2.1 Onset-on-stress pattern: Arabela

Arabela is a Zaparoan language spoken in Peru. The data come from Topintzi (2010),
citing Rich (1963) and Payne and Rich (1988). Stress in Arabela follows a basic left-
to-right trochaic pattern, falling on the initial syllable and every other syllable there-
after. Primary stress is assigned to the rightmost foot in the word (11a). However, an
exceptional pattern of stress shift can be observed in some odd-parity words (11b).
According to Payne and Rich, if a word-final syllable that would have received stress
has a voiced onset, and the immediately preceding syllable has a voiceless onset, then
the syllable with the voiceless onset is stressed (1988:6).
(11) a. (tna)(kri) afternoon
(sma)(r) spirit
(hwa)(hni)(y) peaceful
b. (nwa)()(no) *(nwa)(a)(n) brightened
(spo)(h)(sno) *(spo)(hsa)(n) deceived
(mwra)(t)(tynu) *(mwra)(ttye)(n) cause to be seen
In the forms in (11b), the last stress in the word would be predicted to fall on the
fifth and final syllable. Instead, it falls on the fourth syllable, shifting leftward off of
a syllable with a voiced onset to appear on a syllable with a voiceless onset. As in
Karo (2.2), stress in Arabela is sensitive to the voicing of an onset.
Forms in which primary stress shifts off of a syllable with a voiced onset to fall
on a syllable with a voiceless onset can be dealt with by appealing to the positional
augmentation constraints of Smith (2005). In addition to Onset/ (introduced in
3.1.1 for Jma), Smith proposes a constraint banning high-sonority voiced seg-
ments (D) from the onsets of stressed syllables which is universally ranked over a
Laura W. McGarrity
complementary constraint banning low-sonority voiceless segments (T) from

stressed-syllable onsets: [*Onset/D]/ >> [*Onset/T]/.
According to Topintzi (2010: 90), it is only primary stress that is attracted to a
syllable with a voiceless onset in Arabela; secondary stress is not. Unfortunately,
she cites only one form that demonstrates the weight-insensitivity of secondary
stress, [skamnah]. Assuming it is representative of the Arabela pattern, this
form demonstrates that secondary stress does not behave like primary stress in
shifting leftward off of a syllable with a voiced onset to fall on a syllable with a
voiceless onset, e.g., skamnah, (ska)(mna)(h) *(s)(kma)na(h).
To account for the fact that primary and secondary stresses behave asymmet-
rically with respect to voiceless onsets, Arabela requires a primary-stress-specific
version of Smiths constraint, [*Onset/D]/1, which only bans primary stressed
syllables with voiced onsets. In an odd-parity word in which primary stress shifts
off of a final syllable with a voiced onset onto a penultimate syllable with a voice-
less onset, this constraint must be ranked above the general stress constraint that
requires optimally right-aligned stress feet in Arabela, AlignFt-R.
(12) Primary stress shifts onto with voiceless onset, e.g., [spohsno] de-
ceived
/sapohosano/ [*Onset/D]/1 AlignFt-R

a. (spo)(h)(sno) **,***
b. (spo)(hsa)(n) *! *,***
As shown in (12), candidate (b) with the default stress pattern places primary
stress on the final syllable with a voiced onset, incurring a fatal violation of
[*Onset/D]/1. This leaves candidate (a), with shifted stress and misaligned feet as
the optimal form.
That secondary stress does not shift from its default position onto a syllable
with a voiceless onset falls out from the fact that it is not subject to evaluation by
primary-stress-specific [*Onset/D]/1.
(13) No shift of secondary stress onto with voiceless onset, e.g., [skam-
nah]
/sakamanaha/ [*Onset/D]/1 AlignFt-R [*Onset/D]/

a. (ska)(mna)(h) *,*** *
b. (s)(kma)na(h) **,***!*
In (13), both candidates fare equally well with respect to [*Onset/D]/1, as they
both have primary stress falling on a syllable with a voiceless onset. AlignFt-R
decides the winner, choosing the form with the default pattern of stress in (a) over
the form with shifted secondary stress in (b) as it has optimal right alignment of
feet. This tableau shows that the general version of the [*Onset/D]/ constraint
must be ranked below the general stress placement constraints as the optimal form
does place a secondary stress on a syllable with a voiced onset.
3.2.2 Onset-on-stress pattern: Tmpisa Shoshone

Tmpisa Shoshone (Uto-Aztecan; Dayley 1989) also shows sensitivity to the qual-
ity of an onset. Stress in Tmpisa Shoshone is quantity-sensitive (to long vowels)
and follows a left-to-right moraic trochaic stress pattern. Primary stress falls on
the first syllable if it is heavy (CVV), or if both it and the following syllable are light
(CV or CVC) (14a). However, if the second syllable contains a long vowel, it re-
ceives primary stress (14b). Secondary stress is assigned to the first vowel follow-
ing a primary-stressed long vowel as well as on every other (non-final) mora
thereafter.8 (The transcriptions given here are phonemic and do not reflect precise
allophonic variants. The symbol represents the high central vowel [].)
(14) a. Initial primary stress in ... and LL... words
p:kntn having water
su:m:tn ten
tkumpnapin sky
tkummhannknna cook for
b. Peninitial primary stress in L... words
tup:mpi desert
pihn:wtn bee
wkku:npph fog
The default pattern for primary stress in Tmpisa Shoshone, then, is on the first
syllable unless the second syllable contains a long vowel, in which cases it shifts to
the peninitial syllable.9 However, there are words in the language in which pri-
mary stress falls on the second syllable even if it contains a short vowel.
8. Secondary stress on a final mora is apparently optional, dependent largely upon whether
the word is phrase-final (Dayley 1989: 436). The account presented here analyzes the pattern
without final secondary stress.
9. This would fall out from ranking the constraint Weight-to-Stress Principle (WSP:
Heavy syllables must be stressed) above the constraint requiring an initial head foot bearing
primary stress, AlignHdFt-L.
Laura W. McGarrity
(15) Peninitial primary stress following voiceless geminate onset

tupppitn black
kuttnna to hit
ktanna to bite
ikkah cut
tosspitn white
As shown in (15), if the second syllable begins with a voiceless geminate obstruent
(pp, tt, tts, kk, ss), it receives primary stress.10 The geminate obstruents are the only
phonemes in Tmpisa Shoshone that surface as voiceless word-medially. While
there are singleton voiceless phonemes /p, t, ts, k/, they only surface as voiceless
word-initially; medially between voiced sounds, they are always voiced due to a
number of lenition processes operating in the language. For this reason, Gordon
(2005) groups Tmpisa Shoshone with languages like Karo and Arabela as having
stress assignment that is sensitive to the voicing of an onset, preferring voiceless
over voiced. However, this does not explain why primary stress shifts to the second
syllable when the onset of the initial syllable is also voiceless, as in all of the forms
presented in (15). While voicing will be shown to be a factor, what seems to play a
bigger role in why stress shifts to the second syllable is the fact that the onset is also
a geminate.
Topintzi (2010) discusses several languages in which word-medial geminates
pattern as moraic onsets, influencing stress placement and other weight-sensitive
prosodic phenomena. For example, in Marshallese (Zewen 1977), stress is quanti-
ty-sensitive, preferably falling on a heavy syllable with a long vowel. In disyllabic
words in which both syllables are either light (LL) or heavy (HH), stress defaults to
the initial syllable, e.g., [nbwar] LL to praise, [m:ja:j] HH to be clear of under-
wood. However, disyllabic words with a medial geminate have final stress, suggest-
ing that the geminate is parsed as an onset, e.g., [jibwbw] LH morning,
[yemwmwn] LH good. Topintzi assumes (along with Hayes 1989; Hyman 1985),
that geminates are underlyingly moraic. When they are parsed as onsets, as in
Marshallese, they contribute their weight to the syllable, making it heavy.
As in Marshallese, syllables with geminate onsets in Tmpisa Shoshone seem
to be heavier than those with singleton onsets, attracting stress from its default
initial position. What makes Tmpisa Shoshone different from Marshallese, how-
ever, is that it is not any syllable with a geminate onset that attracts primary stress,
10. According to Dayley (1989:410) ... both ng and s are always phonetically long and fortis
(like geminate consonants) between vowels. But, since there is no distinction between simple
and geminate ng and s, they are always written single. Because of how it patterns, I transcribe
this sound as a geminate [ss] for consistency.
but only those with voiceless geminate onsets; syllables with voiced geminates do
not, as shown in (16).
(16) Voiced geminates fail to attract stress
tmmono *tommno year
nnmp *innump thief
kmmaknna *kimmkinna to come here
In this respect, Tmpisa Shoshone is also similar to Arabela (3.2.1) in that voice-
less onsets are preferred over voiced for stressed syllables; however, it also more
restrictive than Arabela in that only voiceless geminate onsets are preferred and
not voiceless onsets in general.11
A moraic analysis of Tmpisa Shoshone, along the lines of Topintzis analysis
for Marshallese, that assumes that geminates are underlyingly heavy will run
into trouble in explaining why only some bimoraic syllables (with voiceless gem-
inate onsets) attract stress while other bimoraic syllables (with voiced geminate
onsets) do not.
To account for the Tmpisa Shoshone facts I appeal to an analysis proposed by
Gordon (1999; 2005) which references the combined weight or prominence of a
syllable based on multiple dimensions, in this case, the preference for voiceless
onsets over voiced alongside the inherent prominence of long or geminate onsets
over singletons. Gordon represents such distinctions by using a skeletal X-slot
model of the syllable (in which all short segments have one X-slot and all long seg-
ments have two) coupled with featural information (e.g., [voice], [+vocalic], etc.).
For example, a hypothetical language with a four-way prominence distinction in
which syllables with a long vowel (CVV) are preferred over syllables with voiceless
onsets (TV) which are preferred over syllables with voiced onsets (DV) which in
turn are preferred over onsetless syllables (V) would have the representations in
(17) (where R indicates a rimal slot).
(17) CVV > TV > DV > V
[X[XX]R] [X[X]R] [X[X]R] [[X]R]
| |
[voi] [+voi]
Weight distinctions are made by comparing differences in the number of weight
units along with featural differences. The first prominence distinction CVV > TV
11. That voiceless geminates should play a role in stress where prominence is concerned is
not surprising. Gordon (2005:632) notes that lengthening of an onset consonant is a strategy
for providing a following stressed vowel with an auditory boost by increasing its perceptual
prominence. Recall that Karo (2.2) uses such a strategy by geminating the voiceless onsets of
stressed syllables.
Laura W. McGarrity
involves a difference in the number of rimal weight units (2 > 1). According to
Gordon, rimal weight always takes precedence over onset weight. The second dis-
tinction TV > DV involves a difference in featural specification. This represents the
inherent prominence of voiceless onsets over voiced onsets. The last distinction
DV > V involves a difference in the number of weight units in the syllable overall,
representing the preference for syllables with onsets over those without onsets.
Gordon uses these representations to formulate (both positively and negatively)
a series of constraints based on Prom (the constraint which prefers stressing more
prominent syllables over less prominent syllables). The constraint Prom[[XX]R]
requires that a syllable with a branching rime receive stress; *Prom[[X]R] prohibits
an onsetless syllable from receiving stress; Prom[X[voi][X]R] requires that a syllable
with a voiceless onset carry stress. At first glance, these constraints appear to simply
be alternatively named versions of constraints that have already been introduced
(e.g., WSP, Onset/, [*Onset/D]/ respectively). What makes Gordons Prom con-
straints different is that the distinctions made in different constraints, e.g., reference
to branching rimes in Prom[[XX]R] and voiceless onsets in Prom[X[voi][X]R], can
be conflated into one constraint, e.g., Prom[X[voi][XX]R]. This constraint is active
in Pirah (Everett & Everett 1984), in which TVV syllables (with a long vowel and a
voiceless onset) are the most prominent syllable type, always receiving stress over
lighter competitors.
While Gordon (2005) does not examine the prominence of geminate onsets
over singletons, referencing such a distinction in a Prom constraint for Tmpisa
Shoshone in combination with the preference for voiceless onsets is a fairly
straightforward extension of his analysis. As Topintzi (2010) shows for Marshallese,
geminate onsets can be more prominent than singleton onsets. This can be cap-
tured by the prominence hierarchy in (18a). The prominence of voiceless onsets
over voiced onsets is repeated from (17) in (18b). Reference to the two promi-
nence hierarchies can be conflated into one constraint requiring that syllables with
voiceless geminate onsets receive stress, as shown in (18c).
(18) a. Geminate prominence hierarchy for onsets: Prom[XX[X]R]12
GV > CV

[XX[X]R] [X[X]R]
12. I use a ligature to indicate that the two X-slots in the onset belong to a geminate as opposed
to an onset cluster. Gordon (2005) analyzes Nankina and Bislama as languages in which CCV
syllables are more prominent than CV syllables for stress. Though he does not specify it, pre-
sumably a Prom constraint along the lines of Prom[XX[X]R], which must be distinct from
Prom[XX[X]R], would capture this pattern (though see Topintzi (2010:224225) who chal-
lenges analyses of Nankina and Bislama that assume onset clusters are more prominent than
simple onsets).
b. [voice] prominence hierarchy for onsets: Prom[X[voi][X]R]

TV > DV
[X[X]R] [X[X]R]
| |
[voi] [+voi]
c. Prom[XX
[voi][X]R]: Syllables with voiceless geminate onsets must be
stressed.
One last adjustment must be made to this constraint to account for the Tmpisa
Shoshone facts. While primary stress shifts from its default position to fall on a
syllable with a voiceless geminate onset, secondary stress does not, as seen in the
forms in (19).
(19) kuptpptn (*kuptapputn) long
k: nmokkpphkntn (*k: nmokkuppuhkantn) not having money
toppttsittsi (*toppttsttsi) short
kassttsiknt (*kassttskant) bird
As primary and secondary stress behave asymmetrically with respect to syllables
with voiceless geminate onsets, the constraint requiring that syllables with a voice-
less geminate onset receive stress must be relativized to primary stress only:
PromMain[XX[voi][X]R].
(20) Only primary stress sensitive to [voi] geminate onsets, e.g. [kassttsiknt]
bird
/kassattsikant/ PromMain Align Prom

[XX[voi][X]R] HdFt-L [XX [voi][X]R]
a. ka(ssttsi)(knt) * *
b. (kssa)(ttskan)t *! *
As shown in (20), PromMain[XX[voi][X]R] must outrank AlignHdFt-L (Align

the head foot of a PWd with the left edge of the PWd) to compel primary stress to
shift from the intial syllable onto the penult (a). Candidate (b) with left-aligned
primary stress is eliminated since it fails to place primary stress on the more prom-
inent syllable with the voiceless geminate onset. Since the optimal form does have
a (secondary) stress on a syllable with a singleton onset when there is a more
prominent syllable with a voiceless geminate onset available, the general version of
the Prom[XX[voi][X]R] constraint must be ranked low.
Laura W. McGarrity
3.2.3 Stress-on-Onset language: Niuafoou

Just as onset-on-stress languages can show asymmetrical patterns of primary and
secondary stress assignment, so can stress-on-onset languages. In such languages,
the presence of primary stress on a syllable forces a change in the onset but a sec-
ondary stress does not. Niuafoou represents such an example.
In Niuafoou (de Lacy 2001; Smith 2005; Tsukamoto 1988), glide onsets are
dispreferred. While native Niuafoou words do not contain glides, loanwords from
English do. If a loanword contains a glide in the onset position of an unstressed
syllable, it surfaces faithfully (21a). In such syllables, glide onsets are tolerated.
However, if a glide would surface in the onset of a stressed syllable, it vocalizes,
becoming a high vowel (21b). The following data are from de Lacy (2001), citing
Tsukamoto (1988).
(21) a. Glide onsets in unstressed syllables
ju.n.ti unit
wa..a wire
b. Glide onsets disallowed in stressed syllables
i..te *[j.te] yard
u..pi *[w.pi] whip
ku..ta *[kw.ta] quarter
While glide onsets are preferred over having no onset in unstressed syllables, glides
are not allowed as onsets in stressed syllables. Smith (2005) accounts for the
Niuafoou pattern by appealing to the constraint *[Onset/Gli]/ which bans high-
sonority glides as onsets of stressed syllables. This constraint must be ranked high,
above the Onset/ constraint demanding onsets in stressed syllables, as well as
above Faith-, the constraint that is violated by the vocalization of a glide into a
high vowel, as shown in (22).
(22) Glide onsets banned in stressed syllables, e.g., [i..te] yard
/jte/ *[Onset/Gli]/ Onset/ Faith-

a. i..te * *
b. j.te *!
That primary and secondary stresses appear to behave asymmetrically in Niuafoou

with respect to glide onsets is evident in the form [nj.i..ka] New York. The first
syllable [nju] contains a glide in the onset and bears secondary stress, but the pri-
mary stressed syllable cannot contain a glide onset and so is preceded by syllabic
[i] instead. This demonstrates that *[Onset/Gli]/ must refer specifically to pri-
mary-stressed syllables.
(23) Primary-stress specific *[Onset/Gli]/1, e.g. [njika] New York
/njjka/ *[Onset/ Onset/ *[Onset/ Faith-

Gli]/1 Gli]/
a. nj.i..ka * * *
b. ni..i..ka **! **
c. nj.j.ka *! **
As seen in (23), ranking primary-stress-specific *[Onset/Gli]/1 high eliminates

candidate (c) which has primary stress falling on a syllable with a glide onset. Can-
didate (b), with no glide formation, incurs too many violations of Onset/. This
constraint must be ranked above the general version of *[Onset/Gli]/, so that
candidate (a), which demonstrates that sensitivity to the quality of the onset holds
only for primary but not secondary stressed syllables, can be selected as optimal.
4. Discussion: A comparison with child language acquisition
It has been shown that languages with onset-sensitive stress assignment are com-
parable to languages with rime-based weight-sensitive phenomena in demonstrat-
ing both symmetrical and asymmetrical patterns of primary and secondary stress,
as summarized in the table in (24).
(24) Summary of primary and secondary stress patterns in onset-sensitive
languages
Symmetrical Asymmetrical
Stress-on-onset Dutch (3.1.2) Niuafoou (3.2.3)
Onset-on-stress Jma (3.1.1) Arabela (3.2.1), Tmpisa Shoshone
(3.2.2)
In the case of asymmetrical stress systems, primary-stress-specific versions of

several of Smiths (2005) positional augmentation constraints, as well as an exten-
sion of Gordons (2005) Prom constraints, were proposed to account for the
Laura W. McGarrity
pattern: *[Onset/Gli]/1 for Niuafoou, *[Onset/D]/1 for Arabela, and

PromMain[XX[-voi][X]R] for Tmpisa Shoshone.
Despite the fact that the breadth of patterns discussed here is comparable to
what is found in rime-based stress systems, the number of languages that have been
analyzed with onset-sensitive stress remains relatively small. Gordon (2005) lists 13
languages in his survey (excluding the areally related languages of Australia men-
tioned in footnote 1). Topintzi (2010) mentions about eight more (some with very
limited data). The paucity of languages with onset-sensitive stress has led many to
overlook or even ignore such data, maintaining that onsets should be excluded as
possible players in prosodic phenomena. However, the fact that onsets should play
a role in optimizing stressed syllables in fully developed languages should not be
considered anomalous given that the tendency for stressed syllables to preferen-
tially select a low-sonority onset is quite common in first language acquisition.
Gnanadesikan (2004) discusses such an interaction involving onsets and stress
in her analysis of her daughter Gitanjalis (G) early words (between 2;3 and 2;9). At
this stage of her phonological acquisition, G disallowed complex onsets. In a pat-
tern that is quite well attested in both normally developing and disordered pho-
nologies (Barlow 1997; Barlow & Dinnsen 1998; Chin 1996; Chin & Dinnsen
1992; Ohala 1996, 1999), G would reduce an onset cluster to the least sonorous seg-
ment (e.g., [kin] clean, [fn] friend, [so] snow, [gai] sky). However, this pattern was
not restricted to word-initial position or monosyllabic words. The preference for
low-sonority onsets extended to medial stressed syllables. Word-initial unstressed
syllables in Gs speech were overwritten with a dummy syllable fi-, as shown in (25).
As above, when the tonic syllable began with a cluster, it would reduce to the least
sonorous element (25a). However, while the initial unstressed syllable of the target
word was replaced by fi-, there is evidence that G still retained the segments of the
overwritten syllable in her underlying representation. When the stressed syllable of
the target word had a high sonority glide or liquid as a singleton onset, the low-
sonority onset of the overwritten syllable would replace it (25b).
(25) G (2;32;9)
a. [fi-by] umbrella b. [fi-kala] koala
[fi-gio] mosquito [fi-pis] police
[fi-din] Christina [fi-bt] barrette
If the dummy syllable simply replaced the initial unstressed syllable, koala would
be expected to be pronounced as *[fi-ala] or *[fi-wala]. Instead, the stressed sylla-
ble selects a more optimal onset, in this case, the low sonority voiceless stop of the
target unstressed syllable. Recall that this is similar to the stress-on-onset language
Niuafoou (3.2.3), in which high sonority glide onsets are tolerated in all positions
in a word except in primary-stressed syllables. This suggests that the constraint
proposed for Niuafoou *[Onset/Gli]/ could also have been active in Gs gram-
mar at this stage in her development. The two patterns differ, however, in the
choice of repair. Niuafoou vocalizes the glide (violating Faith-) while G selects a
non-contiguous segment, violating the faithfulness constraint Contiguity.
Very similar patterns, in which a high sonority onset in a stressed syllable is re-
placed by a low-sonority onset from a (deleted) unstressed syllable, have been dis-
cussed in other descriptions of childrens early words. Smith (1973) cites several
words in his son Amahls speech that pattern like Gs, produced at nearly identical
ages. While Amahls speech differs from Gs at this stage in that the unstressed syl-
lable deletes fully and is not replaced by a dummy syllable,13 it is similar in that a
high sonority onset in a stressed syllable is replaced by a lower sonority onset from
deleted syllable, as in (26). A comparison with the forms in (26b) demonstrates that
when the initial segment of the target was equally or more sonorous than the stressed-
syllable onset, the stressed-syllable onset surfaces faithfully, suggesting that it is not
simply a desire to preserve the word-initial segment that is driving the pattern.
(26) Amahl (2;4 2;9)
a. [bu:n] balloon b. [pho:d] supposed
[b] belong [thv] disturb
[b:n] banana [bdi] chaptti
[phno:] piano [tn] return
This pattern is also evident in the early productions of children acquiring Dutch
(Fikkert 1994). Fikkert finds that in the earliest stages of their prosodic acquisi-
tion, children tend to select the stress foot of the adult target and map it onto a
trochaic template, deleting target unstressed syllables outside of the stress foot. She
notes that in many cases, as with Amahl, certain segments from these deleted syl-
lables may surface in the childs truncations, as shown in (27).14
(27) Tom papegaai /pa:pa:j/ [pa:j] (1;5.14)
Catootje ballon /bln/ [bo:mi:] (1;10.11)
Elke politie /po:li:tsi:/ [pi:si:] (2;4.29)
Robin macaroni /ma:ka:ro:ni:/ [ma:ku:ni:] (2;0.4)
Noortje konijn /ko:nin/ [kin] (2;3.21)
13. Interestingly, Amahl does use dummy syllables at a later stage of his development (age
3;53;7) in a way that is almost identical to G (except with ri- instead of fi-, e.g., [ri-tk] attack,
[ri-ta:] guitar), including by using the lower sonority onset of the overwritten syllable to replace
the high sonority stressed syllable onset, e.g., [ri-dzf] giraffe.
14. These forms were chosen to illustrate the prevalence of the same pattern across multiple
children. Each child, in fact, produced multiple forms with the same pattern.
Laura W. McGarrity
In each case, the onset of an unstressed syllable from the adult target is realized as
the onset of the stressed syllable in the childs truncation when it is less sonorous.
Fikkert argues that it is actually a dispreference for coronal segments that drives
whether or not a segment from a deleted syllable will be included in a childs
truncated productions. According to Fikkert, ...the coronal place of articulation
of the target foot-initial consonant is replaced by either a labial or a dorsal one.
...[I]f the target foot-initial consonant is labial or dorsal no substitution takes
place (1994:241). However, the data do not always bear this out. As seen in (28a)
many of the same children from (27) exhibit forms in which a coronal onset in a
stressed syllable is not replaced by labial or dorsal if the coronal is equally or less
sonorous. Likewise, the forms in (28b) demonstrate that a coronal may substitute
for a labial or dorsal onset in a stressed syllable, contra Fikkerts prediction, when
the coronal is less sonorous than the target onset. These forms give reason to be-
lieve that onset sonority may indeed play a role in the acquisition of stress in
Dutch as it does in English.
(28) a. Catootje Cato /ka:to:/ [to:] (1;11.25)
Elke pantoffel /pntfl/ [tfu:] (2;4.29)
Robin muziek /my:si:k/ [si:k] (2;0.4)
Robin krokodil /kro:ko:dl/ [di:w] (1;10.7)
Noortje locomotief /lo:ko:mo:ti:f/ [ti:f] (2;6.5)
b. Tom telefoon /te:lfo:n/ [t] (1;7.9)
Robin televisie /te:lvi:si:/ [ti:si:] (1;5.25)
As these patterns from childrens acquisition are similar to the stress-on-onset pat-
terns of Dutch (3.1.2) and Niuafoou (3.2.3), in which the presence of stress in a
syllable triggers the creation or alteration of an onset, we might expect to see on-
set-on-stress patterns in child speech as well, in which the presence or quality of
an onset attracts stress away from its default position. However, childrens early
words tend to exhibit faithfulness to the stressed syllables of the adult forms due to
their perceptual salience (Echols & Newport 1992), frequently retaining them and
producing them with greater accuracy than unstressed syllables which are subject
to deletion. In their investigation of childrens early productions of multisyllabic
(2- to 5-syllable) words, Kehoe and Stoel-Gammon (1997) found that children
overwhelmingly retain the stressed syllables (as well as word-final unstressed syl-
lables) of the target in their truncated forms. Further, stress errors, in which the
childs placement of stress differs from the adult target, are very rare. When they
do occur, errors involving transposition of target primary and secondary stresses
(e.g., WS SW, where S represents a strong or stressed syllable and W is a
weak or unstressed syllable) are most frequent, with errors involving the placement
of stress on an unstressed target syllable (e.g., WS SW) occurring very rarely. In

sum, childrens early productions are highly faithful to the position of stress in
target forms, rarely shifting it to another syllable, regardless of the onset it may
have. As such, onset-on-stress patterns in which stress shifts to a syllable with a
lower sonority onset (as in Arabela, 3.2.1) do not seem to be a strategy used in
childrens early productions.15
The last finding of this paper that can be compared with childrens speech is
the tendency for some languages to treat primary and secondary stressed syllables
asymmetrically. Determining whether primary and secondary stresses behave
symmetrically or asymmetrically in childrens prosodic acquisition can be a chal-
lenge to investigate, as childrens early words are often quite short, lacking second-
ary stresses altogether. However, Kehoe and Stoel-Gammon (1997) included a
number of two- to five-syllable target words with both primary and secondary
stress in their analysis. In some respects, children seem to treat both stress types
symmetrically. For example, both primary and secondary stresses are commonly
preserved in childrens truncated forms. In SS, SW, and WS words, childrens
most frequent truncations included both stressed syllables (e.g., [khwu]
kngaro). When children truncated these forms to a single stressed syllable, it
was always the last stressed syllable that was retained, regardless of whether it bore
primary or secondary stress in the target word (e.g., [w] kngaro, [b:] tckeb16).
However, children were far more likely to reduce SS, SW, and WS words to a
single stressed syllable (as opposed to both stressed syllables) if the target word
had primary stress on the final syllable (WS) than if it had secondary stress on the
final syllable (SW). Kehoe and Stoel-Gammon conclude from this that children
are aware of the distinction between primary and secondary stress and do treat
them differently. This pattern indicates that primary-stress-specific constraints (in
this case, a version of Faith-Stress relativized to primary stress) may be operative
in childrens early prosodic acquisition.
In conclusion, onsets have been found to play a role in stress assignment in
both fully developed and developing languages. In particular, a preference for low
sonority onsets in stressed syllables is in evidence in adult languages (e.g., Arabela
and Tmpisa Shoshone) as well as in childrens early words. However, differences
15. While Kehoe and Stoel-Gammons analysis largely focuses on the prosodic structure of
childrens productions rather than on their segmental content, they do note that segmental-
prosodic interactions occur in their data, observing that childrens truncation patterns may
consist of segmental features from other syllables.... In particular, onsets are often taken from
any syllable (1997:123). They include two examples that are consistent with sonority-based
onset selection, e.g., [bwun] balloon and [i] chimpanzee.
16. This was one of several nonsense words included in Kehoe and Stoel-Gammons analysis,
used as a control for potential word familiarity effects.
Laura W. McGarrity
exist in how fully developed vs. developing phonologies achieve this preference.
Fully developed languages may either shift stress or alter an onset to achieve en-
hanced prominence in stressed syllables, while childrens productions tend to em-
ploy only the latter option due to high-ranked faithfulness to the position of stress.
Finally, that children are able to distinguish stressed vs. unstressed syllables is evi-
dent in their tendency to retain stressed over unstressed syllables, but studies of
longer word productions have also shown that children can distinguish between
degrees of stress, differentiating primary vs. secondary stressed syllables. This
leaves open the possibility that future studies will find further parallels between
the stress systems of fully developed languages and childrens prosodic acquisition
including the interaction of various phonological phenomena with stress to form
both symmetrical and asymmetrical patterns.
References
Abrahamson, A. & Abrahamson, J. 1984. Os fonemas da lngua Jma. In Estudos sobre lnguas
Tup do Brasil [Serie Lingstica 11], R. Dooley (ed.), 157174. Brasilia: Summer Institute
of Linguistics.
Barlow, J.A. 1997. A Constraint-Based Account of Syllable Onsets: Evidence from Developing
Systems. PhD dissertation, Indiana University.
Barlow, J.A. & Dinnsen, D.A. 1998. Asymmetrical cluster development in a disordered system.
Beckman, J.N. 1998. Positional Faithfulness. Ph.D. dissertation, University of Massachusetts,
Amherst.
Blevins, J. 2001. Where have all the onsets gone? Initial consonant loss in Australian Aboriginal
languages. In Forty Years On: Ken Hale and Australian Languages, J. Simpson, D. Nash, M.
Laughren, P. Austin & B. Alpher (eds.), 481492. Canberra: Pacific Linguistics.
Booij, G. 1995. The Phonology of Dutch. Oxford: Clarendon Press.
Buller, B., Buller, E. & Everett, D. 1993. Stress placement, syllable structure, and minimality in
Banaw. International Journal of American Linguistics 59: 280293.
Chin, S.B. 1996. The role of the sonority hierarchy in delayed phonological systems. In Pathologies
of Speech and Language: Contributions of Clinical Phonetics and Linguistics, T.W. Powell (ed.),
109117. New Orleans LA: International Clinical Phonetics and Linguistics Association.
Chin, S.B. & Dinnsen, D.A. 1992. Consonant clusters in disordered speech: Constraints and
correspondence patterns. Journal of Child Language 19: 259286.
Davis, S. 1988. Syllable onsets as a factor in stress rules. Phonology 5: 119.
Dayley, J. 1989. Tmpisa (Panamint) Shoshone Grammar. Berkeley CA: University of California
Press.
Dixon, R.M.W. 1981. Wargamay. In Handbook of Australian languages, Vol. II, R.M.W. Dixon &
B.J. Blake (eds.), 1144. Amsterdam: John Benjamins.
Echols, C.H. & Newport, E.L. 1992. The role of stress and positioning in determining first words.
Everett, D.L. & Everett, K. 1984. On the relevance of syllable onsets to stress placement. Linguis-
tic Inquiry 15: 705711.
Fikkert, P. 1994. On the Acquisition of Prosodic Structure [HIL Dissertations 6]. The Hague: HAG.
Gabas Jr., N. 1999. A Grammar of Karo (Tup). PhD dissertation, University of California at
Santa Barbara.
Gnanadesikan, A. 2004. Markedness and faithfulness constraints in child phonology. In
Constraints in Phonological Acquisition, R. Kager, J. Pater & W. Zonneveld (eds.), 73108.
Cambridge: CUP.
Goedemans, R. 1998. Weightless Segments. The Hague: HAG.
Gordon, M. 1999. Syllable Weight: Phonetics, Phonology, Typology. PhD dissertation, Univer-
sity of California at Los Angeles.
Gordon, M. 2005. A perceptually-driven account of onset-sensitive stress. Natural Language and
Linguistic Theory 23: 595653.
Hayes, B. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20: 253306.
Hyman, L. 1985. A Theory of Phonological Weight. Dordrecht: Foris.
Kehoe, M. & Stoel-Gammon, C. 1997. The acquisition of prosodic structure: An investigation of
current accounts of childrens prosodic development. Language 73: 113144.
Knudson, L.M. 1975. A natural phonology and morphophonemics of Chimalapa Zoque. Papers
in Linguistics 8: 283346.
de Lacy, P. 2001. Markedness in prominent positions. In HUMIT 2000: MIT Working Papers in
Linguistics 40, O. Matushansky, A. Costa, J. Martin-Gonzalez, L. Nathan & A. Szczegielniak
(eds), 5366. Cambridge, MA: MIT Working Papers in Linguistics.
McGarrity, L.W. 2003. Constraints on Patterns of Primary and Secondary Stress. PhD disserta-
tion, Indiana University at Bloomington.
Ohala, D.K. 1996. Cluster Reduction and Constraints in Acquisition. PhD dissertation, Univer-
sity of Arizona at Tucson.
Ohala, D.K. 1999. The influence of sonority on children's cluster reductions. Journal of Com-
munication Disorders 32: 397422.
Payne, D. & Rich, F. 1988. Sensitivity to onset in Arabela stress. Ms, Informacin de Campo 743.
Yarinacocha, Pucallpa, Per: Instituto Lingstico de Verano.
Rich, F. 1963. Arabela phonemes and high-level phonology. In Studies in Peruvian Indian Lan-
guages I [Summer Institute of Linguistics Publications in Linguistics and Related Fields 9],
Benjamin Elson (ed.), 193206. Norman OK: Summer Institute of Linguistics of the Uni-
versity of Oklahoma.
Robinson, L. 1975. Some phonological rules of Iowa-Oto (Siouan). Proceedings of the Mid-
America Linguistics Conference, 439448. Lawrence: University of Kansas.
Shiraishi, H. 1998. Evidence for prominence-based accent shifting in Ainu (the Saru dialect). In
Proceedings of the 117th Meeting of the Japanese Society of Linguists, 266271.
Smith, J. 2005. Phonological Augmentation in Prominent Positions. New York NY: Routledge.
Strehlow, T. 1942. Aranda phonetics. Oceania 12: 255302.
Tamura, S. 1996. The Ainu-Japanese Dictionary: Saru Dialect. Tokyo: Sofukan.
Topintzi, N. 2010. Onsets: Suprasegmental and Prosodic Behaviour. Cambridge: CUP.
Tsukamoto, A. 1988. A Grammar of Niuafoou. PhD dissertation, Australian National University.
Yallop, C. 1977. Alyawarra: an Aboriginal language of Central Australia. Research and Regional
Studies 10. Canberra: Australian Institute of Aboriginal Studies.
Zewen, F.-X.N. 1977. The Marshallese Language: A Study of its Phonology, Morphology, and
Syntax. Berlin: Dietrich Reimer.
Zoll, C. 1997. Conflicting directionality. Phonology 14: 263286.
A faithfulness conspiracy
The selection of unfaithful mappings
in Amahls grammar
Simon Fraser University
Children frequently reduce marked target structures to unmarked outputs.

However, multiple reduction strategies are often available, and pinpointing
a principle that unifies them can be difficult. This paper examines several
markedness-reducing processes in Amahls developing phonology (Smith 1973),
showing that seemingly unrelated repairs actually had a coherent objective: to
avoid the accumulation of multiple repairs. This finding is significant on two
levels: first, the pattern challenges analyses that rely on ranked constraints, in
which violations cannot accumulate across constraints; second, it appears that
multiple phonological processes (unfaithful by definition) conspire to preserve
faithfulness. This pattern is defined as a faithfulness conspiracy, and the concept is
fleshed out with other examples from Amahls development as well as cases from
fully-developed languages.
1. Introduction
Conspiracies (Kisseberth 1970) have long been a fundamental phenomenon for

which any phonological grammar must account. They occur when multiple repair
processes within a grammar conspire to achieve a single function, for instance,
elimination of some marked construct. In Yawelmani Yokuts, for instance, several
different deletion and epenthesis processes work together to avoid sequences of
three consonants, either by eliminating such sequences when they arise, or by fail-
ing to apply when they would be produced. It was this observation that led
Kisseberth (1970) to propose that phonological processes could be described by
their function rather than simply their structure. Kisseberths conceptualization of
structural conspiracies made it possible for phonologists to posit a link between a
grammars teleological goals and the processes that achieve those goals.
Constraint-based approaches to phonological theory, like Optimality Theory

(OT; Prince & Smolensky 1993/2004), were conceived in part to address the fact
that phonological processes often seem to serve larger functions within a gram-
mar. Such theories propose that marked structures and repair processes are inde-
pendent components of the grammar, and the ranking of constraints on each
determines the repair(s) for any given marked structure. In particular, conspira-
cies arise when a constraint against a marked structure (a markedness constraint)
is ranked above more than one constraint requiring that the output form be faith-
ful to its underlying representation (faithfulness constraints). By separating the
avoidance of a marked structure from its repairs, OT captures the functional and
not simply the descriptive aspects of the grammar.
Childrens grammars are ripe for conspiracies, because children tend to reduce
many marked structures in early stages of development, resulting in productions
that are often unfaithful in multiple ways (e.g., Dinnsen 2011; Pater & Barlow 2003).
Smolensky (1996) accounted for these early unmarked productions by positing that
in the initial state, markedness constraints are ranked above faithfulness constraints.
This ranking opens the door for conspiracies, and indeed they are frequently seen
in developing grammars. For instance, Gitanjali (Gnanadesikan 2004) did not pro-
duce onset clusters at age 2. She exhibited deletion (of the more sonorous segment)
and coalescence (in the case of labial clusters). These two very different repairs were
functionally unified in their goal of eliminating clusters from the grammar.
Conspiracies of the type described by Kisseberth function to eliminate par-
ticular marked structures, but he did not consider cases in which multiple pro-
cesses may conspire to achieve other types of goals. The current paper describes a
series of repair processes in the grammar of a single child, Amahl (Smith 1973).
Amahl exhibits several traditional conspiracies, in which two or more processes
each serve to eliminate some structure. Though Amahls conspiracies at first glance
appear independent from one another, they are in fact united by a single function.
That is, when the processes are considered together, they unite to create a larger
emergent conspiracy whose goal cannot be defined by the avoidance of a particu-
lar marked structure. This epiphenomenal pattern will be defined as a new type of
conspiracy. Moreover, each of the patterns presented is a cumulative faithfulness
effect (CFE; Farris-Trimble 2008, 2009, 2010): a case in which multiple otherwise-
acceptable individual repairs are not permitted to co-occur. These cumulative in-
teractions are a key part of Amahls emergent conspiracy. In spite of its facility in
handling conspiracies, classic Optimality Theory is not equipped to explain CFEs.
Instead, analyses are presented that appeal to local constraint conjunction (LC;
ubowicz 2002; Smolensky 1995, 2006) and Harmonic Grammar (HG; Legendre,
Miyata & Smolensky 1990a,b), a theory that relies on weighted rather than ranked
constraints.
2 describes three cases in Amahls phonology that unite together in a single

functional goal. 3 introduces a new type of conspiracy based on the data in 2. 4
turns to a series of cases in which the conspiracy is violated. A discussion and
conclusion are presented in 5.
2. Multiple conspiracies in Amahls grammar
Amahl grew up in Great Britain in the late 1960s. His linguistic development was
chronicled by his phonologist father. In the resulting book, Smith (1973) laid out
a full set of rules describing Amahls early grammar as well as his longitudinal
development. The book also contains a rich dataset of hundreds of tokens pro-
duced at different stages of development.1 It is an ideal place to look for the
conspiratorial interaction of multiple error patterns. The following sections will
present three cumulative faithfulness effects in Amahls grammar. Each CFE con-
tains a conspiracy, but the key fact is that the three CFEs, though consisting of very
different error patterns, all work together toward one larger goal.
2.1 Word-final voiced fricatives
The earliest stages of Amahls grammar consisted almost entirely of unmarked seg-
ments and syllable structures. For instance, adult fricatives were realized as stops
in all word positions, as in the examples in (1a). Amahl also did not produce voiced
obstruent codas at these early stages all obstruent codas were realized as voice-
less, as in (1b). Given the stopping and devoicing patterns, words that end in a
voiced fricative are a perfect test case to determine how the two repairs interact.
The examples in (1c) reveal that Amahl deleted final voiced fricatives.
(1) Amahl: Stopping, word-final devoicing, and deletion
a. Fricatives realized as stops (Stopping)
bus bt (1) side dait (1)
brush bt (1) shirt dt (1)
bath bat (1) thirsty ddi (2)
there d (1) zoo du (4)
1. This paper focuses on Amahls earliest stages of development. His grammar was quickly
changing, and so narrowing the scope down to a single stage is impractical. In general, though,
the data span the time from Stage 1 to Stage 14, which encompasses ages 2;60 (years;days) to
2;256, or a total of about seven months. Two examples in (14) are from later stages; these data
are taken from Dinnsen, OConnor and Gierut (2001). Stages are always shown in parentheses
in the data below.
b. Word-final voiced obstruents are realized as voiceless (Word-final de-

voicing)
bead bi:t/bi:d (1) red dt (1)
peg bk (1) leg k (1)
crib u:p (4) rub wp (4)
c. Word-final voiced fricatives are deleted
noise ni (4) cheese di (4)
nose nu (1) glasses ai (1)
pliers bai (2) news nu (14)
please bi (1) scissors did (1)
Two conspiracies occur in these data, one for each of the marked structures. First,
both devoicing and deletion eliminate word-final voiced obstruents, and second,
both stopping and deletion eliminate fricatives. The two conspiracies overlap in
the case of word-final voiced fricatives, which is where the deletion occurs. It is
tempting to argue that these are simply three different processes with three differ-
ent structural descriptions: one process targets final voiced stops, another targets
voiceless or non-final fricatives, and the third simply has a more specific target,
final voiced fricatives. However, this misses the larger generalization that there are
only two marked constructs involved in the equation: final voiced obstruents and
fricatives. The fact that a different repair is invoked when those two marked struc-
tures coincide is unexpected.
If Amahl employed both stopping and voicing independently, why is a word like
noise not realized as [nit]? This output would eliminate the conspiracies and limit
the number of overall processes used. However, this repair would require the accu-
mulation of multiple repair processes for a single marked structure. In fact, the two
repairs did not interact at all; instead, they were superseded by a third repair, dele-
tion. This pattern is typical of a CFE (Farris-Trimble 2008): two otherwise indepen-
dently-allowable repairs are prohibited from co-occurring in some domain (e.g., a
segment or a syllable). Instead, a third repair, the fell-swoop repair, eliminates both
marked structures in some other way. In this case, deletion of the final voiced frica-
tive is the fell-swoop repair: it eliminates the doubly-marked structure in one blow.
CFEs are interesting theoretically in that even though they contain traditional
conspiracies, they are not predicted by standard constraint-based grammars. In
OT, for example, CFEs would require that a candidate that violates multiple low-
ranked constraints be worse than one that violates only a single higher-ranked
constraint. This contradicts a key tenet of OT: strict domination, which states that
the satisfaction of multiple low-ranked constraints can never make up for the vio-
lation of a higher-ranked constraint. That is, the ranking of constraints is absolute,
and two constraints cannot work together to gang up on a third constraint.
The constraints relevant to an account of Amahls CFE are defined in (2), and
the tableaux in (3)(5) illustrate the ranking paradox that arises. The two marked-
ness constraints, *VoicedCodaObstruent and *Fricative, must both be high-
ranked, since neither of those structures occurs at this stage. This is another
hallmark of a CFE (Farris-Trimble 2008): two high-ranked markedness constraints
are both violated by the fully-faithful candidate. The tableau in (3) provides evi-
dence that Max must be ranked above Ident[voice] to rule out deletion as a repair
for final voiced stops. The tableau in (4) shows that Max must be ranked above
Ident[continuant] to rule out deletion as a repair for final voiceless fricatives.
However, as in (5), ranking Max above the two Ident constraints incorrectly
eliminates candidate e., the attested candidate.
(2) Constraints
Markedness:
*VoicedCodaObstruent: Voiced coda obstruents are prohibited
*Fricative: Fricatives are prohibited
Faithfulness:
Max: Input segments have output correspondents
Ident[voice]: Input and output correspondents have the same value
for the feature [voice]
Ident[continuant]: Input and output correspondents have the same
value for the feature [continuant]
(3) Max >> Ident[voice]
/bd/ bed *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. bd *!
b. bt *
c. b *!
(4) Max >> Ident[continuant]
/bs/ bus *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. bs *!
b. bt *
c. b *!
(5) Ranking paradox
/niz/ noise *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. niz *! *
b. ni:s *! *
c. nid *! *
d. ni:t * *
e. ni *!
Even though candidate e. in (5) violates Max, it is still preferred over candidate
d., which violates both Ident[voice] and Ident[continuant]. This constraint
ranking is indicative of a conspiracy: one or more markedness constraints ranked
above two or more faithfulness constraints, which are themselves ranked
(McCarthy 2002). The lowest-ranked faithfulness constraint provides the typical
repair for the marked structure, but the higher-ranked faithfulness constraint
(in this case, Max) is sometimes violated, when its violation is compelled by a
higher-ranked constraint. The problem is that there is no high-ranked constraint
here to compel the violation of Max. Instead, what compels the Max violation is
the violation of both of the lower-ranked faithfulness constraints, but standard
OT has no way of allowing this.
What is necessary to deal with this CFE is an account that will allow the mul-
tiple violation of low-ranked constraints to overcome the satisfaction of a higher-
ranked constraint. There are at least two ways to do this. One is local constraint
conjunction (LC; ubowicz, 2002; Smolensky, 1995; Smolensky, 2006): two con-
straints are conjoined to create a new constraint that is violated only when both its
conjuncts are violated. The LC constraint is always ranked higher than its indi-
vidual conjuncts. In this case, it is the LC constraint that compels violation of
Max, even though Max is ranked above both of the individual components of the
LC constraint. The relevant LC constraint is defined in (6).
(6) LC(Ident[voice]&Ident[continuant])seg: Input and output correspon-
dents must have the same value for both the features [voice] and [continu-
ant] (violated only if both conjuncts are violated within the domain of a
segment)
The tableaux in (7)(9) illustrate the outcome of the LC constraint. It plays no role
in (7) or (8), because neither of the individually-marked structures violates both
Ident constraints. However, in (9), candidate d., which has undergone both stopping
and devoicing, does violate both, and thus it violates the high-ranked LC constraint
as well, which eliminates it.
(7) Max >> Ident[voice]
/bd/ bed LC *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. bd *!
b. bt *
c. b *!
(8) Max >> Ident[continuant]
/bs/ bus LC *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. bs *!
b. bt *
c. b *!
(9) LC >> Max
/niz/ noise LC *VoiCodaObs *Fricative Max ID[voice] ID[cont]
a. niz *! *
b. ni:s *! *
c. nid *! *
d. ni:t *! * *
e. ni *
There are many advantages to local constraint conjunction for instance, it can
account for some opacity effects, which are difficult to account for otherwise
(Kirchner 1996; Moreton & Smolensky 2002). However, it raises a number of the-
oretical questions. Does Con contain all possible conjunctions of faithfulness con-
straints, or is there an operation to create them? Can more than two faithfulness
constraints be combined in cases in which a grammar allows two-step repairs but
avoids three-step ones (e.g., Hawaiian loanwords in Farris-Trimble 2008). And
what is the proper domain over which to conjoin constraints?
Harmonic Grammar (HG; Legendre, Miyata & Smolensky 1990a,b), a prede-

cessor of OT, provides an alternative account that avoids these questions. In HG,
constraints are weighted rather than ranked, so that the cumulative violation of
multiple low-weight constraints can overcome a single violation of a higher-weight
constraint. This is illustrated in the tableaux in (10)(12). Numerical weights are
shown below each constraint, and each candidates total weight (number of viola-
tions of a given constraint multiplied by the weight of that constraint, summed
across all constraints) are presented in the final column. Because violations penal-
ize a candidate, they have negative values, and the candidate with the highest value
(lowest absolute value) wins.2 The actual values of the constraint weights are less
important than the relationships between them; that is, though the values 2, 1.5,
and 1 are used here, 200, 150, and 100 would have worked equally well.
In the tableaux in (10) and (11), the weightings produce the same outcome
that the rankings did in (3) and (4). That is, Max has a greater weight than either
Ident[voice] or Ident[continuant], and so devoicing and stopping are preferred
repairs. However, the tableau in (12) illustrates the cumulative power of HG: the
sum of the weights of Ident[voice] and Ident[continuant] is enough to outweigh
Max. Candidate d., which violates both of the Ident constraints, is eliminated in
favor of one that violates only Max.
(10) w(Max) > w(ID[voice])
/bd/ *VoiCodaObs *Fric Max ID[voice] ID[cont] H

bed w=2 w=2 w=1.5 w=1 w=1
a. bd 1 2
b. bt 1 1
c. b 1 1.5
(11) w(Max) > w(ID[continuant])
/bs/ *VoiCodaObs *Fric Max ID[voice] ID[cont] H

bus w=2 w=2 w=1.5 w=1 w=1
a. bs 1 2
b. b t 1 1
c. b 1 1.5
2. Legendre, Sorace and Smolensky (2006) provide for the possibility (not explored in this
paper) of constraints that reward candidates by giving them positive violation marks, rather
than penalizing them with negative violations.
(12) w(Max) < w(ID[voice]) + w(ID[continuant])
/niz/ *VoiCodaObs *Fric Max ID[voice] ID[cont] H

noise w=2 w=2 w=1.5 w=1 w=1
a. niz 1 1 4
b. nis 1 1 3
c. nid 1 1 3
d. nit 1 1 2
e. ni 1 1.5
OT was originally favored over HG in part because of such gang effects, which
were thought to be unattested. However, this example and the ones below show
that gang effects among faithfulness constraints are in fact not uncommon (see
also Farris-Trimble 2008). In this case, it appears that Amahls grammar chose the
repair that was the least unfaithful by some metric, and HG provides a way to
quantify that metric. Moreover, weighted constraints dovetail with other psycho-
linguistic theories that involve weighting of linguistic structures (e.g., McMurray,
Cole & Munson 2011; McMurray & Jongman 2011; Toscano & McMurray 2010)
in a way that local conjunction does not, and HG has been shown to accurately
predict many acquisition patterns (Farris-Trimble 2008; Jesney 2011; Jesney &
Tessier 2011). In either case, it is clear that Amahls grammar avoided the multi-
ply-unfaithful output in favor of a candidate in which the offending marked
structure was simply deleted. The question remains whether this avoidance of
multiply-unfaithful forms was an anomaly or if it was a guiding principle else-
where in Amahls development. In the next section we turn to another series of
Amahls repairs.
2.2 Coda nasal clusters
This section presents a conspiracy by which Amahl avoided accumulating repairs

within a consonant cluster. Smith begins his characterization of Amahls grammar
by describing his treatment of nasal+consonant sequences, which were prohibited
in all word positions. This is not an uncommon pattern, as NC clusters are prob-
lematic in many adult languages as well (e.g., Pater 1999, 2001). When a nasal ap-
peared before a voiceless obstruent in the adult form, Amahl deleted the nasal, as
in (13a). When the nasal preceded a voiced obstruent, however, Amahl retained
the nasal and deleted the obstruent instead, as in (13b).
(13) Amahl: Nasal deletion, obstruent deletion

a. Nasals are deleted in word-final nasal+voiceless obstruent clusters
(Nasal deletion)
ant t (1) stamp dp (1)
bump bp (1) tent dt (1)
drink ik (1) trunk k (1)
b. Obstruents are deleted in word-final nasal+voiced obstruent clusters
(Obstruent Deletion)
hand n (1) rubber-band bbbn (1)
mend mn (1) friend wn (2)3
round daun (1) orange in (2)
These two patterns present a striking asymmetry: why would Amahl choose dif-
ferent repairs for two such similar constructs? Again, the answer may lie in the
combination of processes necessary to achieve the final result. Recall from (1b)
above that Amahl repaired word-final voiced obstruents with a devoicing process.
If Amahl had deleted the nasal preceding a final voiced stop, that would in turn
have necessitated devoicing. However, rather than combining nasal deletion and
devoicing, Amahl chose obstruent deletion as a fell-swoop repair. Obstruent dele-
tion eliminates both the cluster and the final voiced obstruent in one step an-
other CFE. Similarly, this pattern is a conspiracy: both nasal deletion and obstruent
deletion are used to repair marked NC# clusters.
As in the account of final voiced fricatives, a ranking paradox arises in an OT
account of the data. The additional constraints are in (14). Here we have split the
Max constraint defined above into a series of constraints prohibiting deletion of
certain types of segments. Max[obstruent] (which replaces the simple Max con-
straint defined in (2)) and Max[nasal] are the relevant constraints for this set of
data; below we will see that Max[sonorant] comes into play as well. It is important
to note that these Max constraints are not Max-feature constraints, militating
against the deletion of a feature. They are constraints on the deletion of segments,
relativized to certain types of segments. Whether such constraints must be ranked
in a fixed hierarchy is an open question not explored in this paper.
(14) Constraints
Markedness:
*NC: The sequence of a nasal consonant followed by an
obstruent is prohibited
3. [wn] was in free variation with [wnd] at this stage.

Faithfulness:
Max[obstruent]: Input obstruent segments have output correspon-
dents
Max[nasal]: Input nasal segments have output correspondents
As before, Max[obs] must be ranked above each of the other two faithfulness con-
straints, but this creates a ranking paradox in the case of a final nasal+voiced
fricative, as in (15). For considerations of space and repetition, only the crucial
ranking arguments will be shown in this and the following sections. Recall from
the tableau in (3) that Max[obstruent] (called simply Max in (3)) must outrank
Ident[voice] at this stage of Amahls grammar to avoid deletion as a repair for il-
licit voicing.
(15) Constraint rankings lead to a paradox for word-final nasal+voiced ob-
struent clusters
Input Winner-loser pair Ranking

a. /bd/ bed [bt] [b] Max[obs] >> ID[voice]
b. /nt/ ant [t] [n] Max[obs] >> Max[nas]
c. /hnd/ hand [n] [t] ID[voice] or Max[nas] >> Max[obs]
As in the previous example, both LC and HG are capable of accounting for this
CFE. Reduction to a nasal coda is only allowed when the alternative would require
both nasal deletion and devoicing. In an LC account, a constraint conjoining
Ident[voice] and Max[nasal] within the domain of a syllable would outrank
Max[obstruent], as in the ranking arguments in (16).
(16) Constraint rankings for an LC account of final nasal clusters

a. /bd/ bed [bt] [b] Max[obs] >> ID[voice]
b. /nt/ ant [t] [n] Max[obs] >> Max[nas]
c. /hnd/ hand [n] [t] LC(ID[voice]&Max[nas]) >> Max[obs]
In an HG account, the constraints against nasal deletion and devoicing would

be individually low-weight, but their combined weight must be enough to out-
weigh the constraint against obstruent deletion. This weighting is illustrated in
(17). Even though the weight of Max[obstruent] is greater than the weight of
either Max[nasal] or Ident[voice], the cumulative weight of Max[nasal] and

Ident[voice] is enough to overcome a violation of Max[obstruent], and so the
obstruent is deleted in this case alone.
(17) Constraint weightings for an HG account of final nasal clusters
Input Winner-loser pair Weighting

a. /bd/ bed [bt] [b] w(Max[obs]) > w(ID[voice])
b. /nt/ ant [t] [n] w(Max[obs]) > w(Max[nas])
c. /hnd/ hand [n] [t] w(Max[nas]) + w(ID[voice])
> w(Max[obs])
Thus far two CFE conspiracies in Amahls grammar have been presented. In both
cases, Amahl used both a fell-swoop repair that eliminated a doubly-marked
structure in a single change. In the next section, we turn to a third CFE, this one
concerning onset clusters.
2.3 Onset clusters
Amahl reduced most onset clusters to the least sonorous segment, as in (18a). This
pattern would be suggested by the Sonority Sequencing Principle (Clements 1990).
Onset /s+sonorant/ clusters, however, were reduced to the sonorant, as in (18b).
(18) Amahl: Cluster reduction to least sonorous segment
a. Most onset clusters reduced to least sonorous segment
blanket bki: (6) brake be:k (8)
clock k (5) guava a:v/ a:w (2)
cross t (10) quite kaip (9)
drop dp (8) tripped dipt (11)
grass a:t (11) twice daif (9)
music mu:i: (2) cute u:t (9)
b. s+sonorant clusters reduced to sonorant segment4
sweet wi:t (6) swing wi/wi (11)
sleep li:p (11) slicer lait (14)
smile maiu (8) sniff nif (11)
Recall that the data in (1) illustrated that Amahl employed a stopping process to
eliminate the fricative /s/. A possible avenue for repairing s+sonorant clusters,
then, would have been to delete the sonorant and reduce the /s/ to a stop, such that
4. Repairs for /fl/ and /fr/ are somewhat more complicated and are addressed in 4.
a word like swing would be produced as [di]. Instead, Amahl chose the fell-swoop
repair and deleted the /s/. In fact, Amahl deleted the /s/ in s+obstruent clusters as
well, as the additional examples in (19) show. In this case, it is unclear whether the
repair is s-deletion as in (18b) or deletion of the more sonorous segment as in
(18a), but it is possible that s-deletion was chosen for all s-clusters because the al-
ternative required multiple repairs. In sum, sonorant deletion and s-deletion con-
spired to repair onset clusters, and s-deletion served as the fell-swoop repair in this
CFE, avoiding the accumulation of sonorant-deletion and stopping.
(19) /s/ is deleted in s+obstruent clusters
spider baid (2) stamp dp (1)
skin in (1)
This CFE produces another ranking paradox in OT. New constraints are defined in
(20). As in the previous two examples, the two markedness constraints are high-
ranked. Because we have now seen evidence that Amahl did not allow either word-
initial clusters or word-final NC clusters, we will collapse these two prohibitions
into a single constraint against complex syllable margins, *Complex. Here we also
introduce Max[sonorant], a constraint that militates against the deletion of sono-
rant consonants and another member of the deconstructed Max family of con-
straints originally introduced in (14). The ranking argument in (21a) shows that
Max[obstruent] must be ranked above Max[sonorant], because in a word like
guava, Amahl preserved the [g] and not the [w]. In (21b) we see that Max[obstruent]
must be ranked above Ident[continuant] so that fricatives undergo stopping rath-
er than deletion. This is the ranking that was shown previously for word-final
fricatives in (11), but at that point the Max family of constraints had not been in-
troduced. Finally, in (21c), a ranking paradox arises: with both Max[sonorant]
and Ident[continuant] ranked below Max[obstruent], this ranking predicts that
an onset [sw] cluster would undergo both glide deletion and stopping to be real-
ized as [d].
(20) Additional constraints
Markedness
*Complex: Complex onsets and codas are prohibited
Faithfulness
Max[sonorant]: Input sonorant consonants have output correspon-
dents
(21) Ranking paradox in the constraint rankings for onset cluster reduction

a. /wav/ guava [a:v] [wa:v] Max[obs] >> Max[son]
b. /sun/ soon [du:n] [u:n] Max[obs] >> ID[cont]
c. /swi/ swing [wi] [di] Max[son] or ID[cont] >> Max[obs]
By now, the solutions are familiar: allow two low-ranked faithfulness constraints
to gang up on a higher-ranked constraint with either LC or HG. A high-ranking
LC constraint like the one in (22), which conjoins Max[sonorant] and
Ident[continuant] within the domain of a syllable, would eliminate the multiply-
unfaithful candidate.
(22) Constraint rankings for an LC account of onset cluster reduction

/swi/ swing [wi] [di] LC(Max[son]&ID[cont]) >> Max[obs]
Likewise, the weighting in (23), in which the combined weight of the two lowest-
weight faithfulness constraints is enough to overcome Max[obstruent], ensuring
its role in eliminating doubly-marked structures.
(23) Constraint weightings for an HG account of onset cluster reduction
Input Winner-loser pair Weighting

a. /wav/ guava [a:v] [wa:v] w(Max[obs]) > w(Max[son])
b. /sun/ soon [du:n] [u:n] w(Max[obs]) > w(ID[cont])
c. /swi/ swing [wi] [di] w(Max[son]) + w(ID[cont])
> w(Max[obs])
In this and the preceding two sections, we have seen three examples of CFEs in
Amahls grammar. By invoking a fell-swoop repair, obstruent deletion, Amahls
grammar created conspiracies to avoid fricatives, voiced coda obstruents, and
both onset and coda clusters. In the next section, we propose a new type of con-
spiracy to account for this broad pattern. In each case, he avoided a doubly-marked
structure without accumulating unfaithful mappings.
3. Faithfulness conspiracies
It was stated earlier that childrens early phonologies are ripe for conspiracies in
which multiple processes work together to eliminate or avoid a particular marked
structure. In conspiracies of this sort, it is the target marked structure that is the
focus, rather than the processes that eliminate it. Amahl exhibited at least four
such conspiracies, each of which can be defined by the marked structure that is the
target of multiple processes: word-final voiced obstruents, fricatives, final NC
clusters, and onset clusters. These targets appear to be independent there is no
necessary relationship between Amahls avoidance of word-final voiced codas and
his avoidance of onset clusters, for instance.
However, when the overall pattern of repairs is examined, one common thread
leaps out: in every case, Amahl avoids repairs that produce multiply-unfaithful out-
puts. Though he could combine individual repairs, he does not. (Note that had he
done this, the afore-mentioned conspiracies would not have occurred.) Instead,
Amahl used a single fell-swoop repair, obstruent deletion, to circumvent the applica-
tion of multiple repairs. Viewed from this wider perspective, it becomes clear that not
only did Amahls phonology avoid a number of marked-structure targets, but there
was also a broader functional target: the avoidance of multiply-unfaithful outputs.
Of course, it so happens that the same repair recurs in each of the cases above
(i.e., obstruent deletion supersedes multiply-unfaithful mappings). In a typical
conspiracy, the grammar contains multiple different strategies to avoid the target.
To really claim that Amahls phonology contains a conspiracy whose target is mul-
tiply-unfaithful outputs, another example in which those targets are avoided in a
different way is necessary. Such a case does exist. Amahls puzzle-puddle-pickle
chain shift was one of his most intriguing error patterns (Dinnsen & Barlow 1998;
Dinnsen, OConnor & Gierut 2001; Smith 1973). In this shift, Amahl produced
words like puzzle as puddle, replacing the voiced fricative with a stop. However, a
word like puddle was produced as puggle, with velarization of the stop before the
[l]. The illustrative data are in (24), taken in part from Dinnsen et al. (2001). What
is interesting about this pattern is that puzzle was not produced as puggle, as one
would expect given the velarization process. This creates a surface opacity in which
it appears that the process by which coronal stops are velarized before a syllabic [l]
has underapplied.
(24) Amahls puzzle/puddle/pickle chain shift (Stopping and velarization)
a. Puzzle words realized as puddle words (Stopping)
puzzle pdl (11)
pencil bntil/pntl (12)
whistle witl (13)
b. Puddle words realized as pickle words (Velarization)

puddle pgl (14)
bottle bkl (14)
handle gl/hgl (15)
c. Pickle words realized target appropriately
pickle pikl (23)
circle t:kl (14)
buckle bkl (14)
The crucial fact is that the doubly-unfaithful mapping, in which puzzle would be
produced as puggle, is avoided. Moreover, it is not simply the combination of stop-
ping and velarization that cannot co-occur. In fact, any derived alveolar stop is
immune to velarization, as in (25). These additional examples suggest that the pat-
tern was broad: any combination of velarization with another error pattern was
avoided.
(25) Velarization blocked in other derived alveolar stops
a. [t] derived from medial cluster simplification, as in master [ma:t]
(11)
pistol pitl (14)
nostril ntil (11)
b. [t] derived from deaffrication, as in matches [mtit] (12)
satchel tl (15)
c. [t] derived from metathesis, as in elephant [flnt] (18)
icicle aikitl (16)
testicle tktl (19)
In this case, Amahl allowed a marked structure to surface in order to avoid a mul-
tiply-unfaithful output. However, this strategy satisfies the same general goal as
the CFEs in 2. This is the final piece in Amahls conspiracy he employed sev-
eral independent repair processes, but all ensured the same outcome: that multi-
ply-unfaithful outputs were avoided.
That this chain shift is an essential piece of the conspiracy is at least partially
supported by its theoretical analysis. Chain shifts are notoriously problematic for
classic OT, and unfortunately HG fares no better. Local constraint conjunction,
however, is effective in accounting for chain shifts. In this case, several LC con-
straints conjoining a constraint militating against velarization (like Ident[coronal])
with constraints militating against stopping (Ident[continuant]), cluster simplifi-
cation (Max), or metathesis (Linearity) would be high ranked. The only differ-
ence between the analysis of the chain shifts and the analysis of the CFEs is in the
ultimate choice of repair in the chain shifts, no fell-swoop repair occurs. Instead,
a marked structure that would otherwise be repaired is allowed to surface.
There is a clear difference between the original sense of conspiracy in Kisseberth
(1970) and the type of conspiracy exhibited by Amahl. Kisseberths original sense
of conspiracy could be thought of as a markedness conspiracy, as its target can be
stated in terms of the output, without reference to how that output is achieved. In
contrast, we might term Amahls consistent avoidance of repair accumulation a
faithfulness conspiracy. That is, its target refers to how the output is achieved rather
than what the output is. In Amahls case, the target is the accumulation of multiple
repair processes. It is important to emphasize that a single CFE does not constitute
a faithfulness conspiracy. Rather, the conspiracy arises when a language uses a
variety of processes to avoid one particular class of unfaithful mappings.
In defining the concept of a faithfulness conspiracy, a number of other phe-
nomena come to mind that might fall into this category. For instance, in her argu-
ment for positional faithfulness constraints, Beckman (1998) discusses several
phonological facts and processes in Ibibio, a Niger-Congo language in which the
initial syllable of the root is privileged. Contrasts are allowed in this position that
are not allowed elsewhere, the root-initial syllable triggers but does not undergo
assimilation, and syllabification is based on root maximization. That is, a number
of different unrelated processes conspire together to preserve faithfulness in the
root-initial syllable. The target of this conspiracy is not defined in terms of a
marked output structure, but rather in the degree of faithfulness required in a
certain portion of the output.
Another set of processes that make up a faithfulness conspiracy occur in
Dakota reduplication (Marantz 1982). A palatalization process results in palatal-
ized initial velars when they are preceded by a prefix ending in /i/. This process
overapplies in reduplicative forms: when the syllable containing the palatal is re-
duplicated, the palatal is also copied, even if it no longer follows an /i/. On the
other hand, a process of a-raising, which changes /a/ to [e] preceding a phrase-
final //, underapplies in reduplicative forms. In sum, multiple processes either
apply or fail to apply in order to achieve a unified goal: faithfulness between the
base and the reduplicant.
Examples like Ibibio and Dakota suggest a revised definition: a faithfulness
conspiracy occurs when multiple processes work together to limit how or where
unfaithful mappings can occur. In Amahls grammar, unfaithful mappings cannot
accumulate; in Ibibio, unfaithful mappings cannot occur in root-initial syllables;
and in Dakota, unfaithful mappings cannot occur between a base and its redupli-
cant. Though this is a new way of looking at positional faithfulness or base-redu-
plicant-correspondence, it seems likely that in languages in which the pressure to
retain faithfulness in certain positions or between certain correspondents is high,

multiple processes might conspire together to achieve the goal.
OT was an important development in phonological theory in part because it
directly addressed the issue of conspiracies. The target of a markedness conspiracy
can be represented by a single markedness constraint in an OT grammar. When
the markedness constraint is ranked above two faithfulness constraints, which are
themselves ranked (M >> F1 >> F2), it can trigger a conspiracy one faithfulness
constraint is violated in some cases, and the other is violated in remaining cases.
The target of a faithfulness conspiracy is less clearly captured in a constraint-based
theory, but in all of the cases described here, a family of specific faithfulness con-
straints outrank several more general faithfulness constraints. The lower-ranked
general faithfulness constraints are violated in order to satisfy the high-ranked
family. It is really this notion of families of specific faithfulness constraints that
best corresponds to the faithfulness conspiracy the high-ranked family of con-
straints stipulates the conspiracys target. In the case of Ibibio, the family of posi-
tional faithfulness constraints employed by Beckman (1998) denotes the target. In
Dakota, the family of base-reduplicant constraints (McCarthy & Prince 1995)
would denote the target. By this account, the LC analysis of Amahls multiple CFEs
may provide a more coherent link to other faithfulness conspiracies than the HG
account, as the family of LC constraints suggests that it is multiply-unfaithful
targets that are disallowed.
The next section turns to a discussion of exceptions to a faithfulness conspira-
cy, showing that they are also principled.
4. Other error patterns
Thus far we have shown that in a variety of different repair patterns, Amahl avoid-
ed doubly-unfaithful repairs, choosing fell-swoop repairs or allowing marked
outputs to surface instead. Of course, as any reader who has been paying careful
attention to the data has noticed, there were many instances in which Amahl did
apply multiple processes to a single form. Kisseberth noted the same issue in his
discussion of markedness conspiracies when he described a conspiracy that avoid-
ed producing adjacent stressed syllables, even though some rules do produce the
structure. He argues In such a case, it is not possible to say that what I have called
a derivational constraint holds for every rule in the grammar; rather it holds for
just some rules. And which rules it holds for and which it does not is not predict-
able (p. 305). That is, not all rules will successfully avoid the marked target struc-
ture. In a constraint-based analysis, this type of situation may arise simply as a
result of higher-ranked constraints.
The data in (26) show Amahls repairs for word-initial /l/, which was primarily
disallowed in the earliest stages of his grammar. Amahl substituted other segments
for the /l/, and the substitute harmonized in place with the following consonant.
Note that each repair violates multiple faithfulness constraints: an /l/ followed by a
coronal was produced as [d] (violating Ident[lateral] and Ident[continuant]), be-
fore a labial as [w] (violating Ident[lateral] and Ident[coronal]), and before a
velar as [] (violating Ident[lateral], Ident[continuant] and Ident[coronal]).
Based on the examples above, one might wonder why the fell-swoop repair did not
apply: why was the /l/ not simply deleted, as it clearly carried many marked fea-
tures and underwent many unfaithful changes?
(26) A multiply-unfaithful pattern (Place harmony)
a. /l/ produced as [d] preceding a coronal consonant
lady dejdi: (1) ladder dd (1)
lash dt (1) light dait (1)
b. /l/ produced as [w] preceding a labial consonant
label we:bu (4) lamp wp (5)
left wpt (3) lift wipt (6)
c. /l/ produced as [g] preceding a velar consonant
leg k (1) Lego u: (1)
lick ik (1) like aik (1)
The simplest explanation for this exception to the conspiracy may be that in gen-
eral, Amahl did not often delete onsets. Onsetless syllables are marked relative to
syllables with onsets, both cross-linguistically and in first-language acquisition. It
is possible that a constraint like Onset (all syllables have onsets) is highly ranked
or weighted in Amahls grammar, thus eliminating a fell-swoop repair. Indeed, we
see in (27) that coda /l/s are often deleted, a repair that would not violate Onset.
(27) Coda /l/ is deleted
pedal bgu (1) ball b: (1)
smell mu (4) apple bu (1)
elbow bu: (1) milk mik (1)
In sum, multiply-unfaithful mappings may be allowed to avoid violation of a high-
er-weight markedness constraint.
Another instance in which multiple repairs may be allowed to accumulate is
when the accumulation of two repairs is preferable to the accumulation of three.
In 2.3 above, a CFE involving the deletion of /s/ in /s+sonorant/ clusters was pre-
sented. The repair options for the /s+sonorant/ cluster were fairly straightforward:
deleting the /s/ resulted in an unmarked singleton onset, but deleting the sonorant
did not; therefore, the /s/ was deleted, even though it had the lower sonority pro-
file. However, we did not mention other fricative clusters like /fl/ and /fr/. Amahl
produced neither /f/ nor the liquids in the early stages of his grammar.5 Thus in
each of these clusters, both segments are marked in some way there is no single
repair that will result in a completely unmarked structure. Amahls productions
for these two clusters are shown in (28).
(28) Other fricative clusters
flag w (8) freeze wi:d (10)
floor w: (4) frog w (7)
flower ww (1) friend wn/wnd (2)
While the [w] may at first glance appear to correspond to the liquid (as it did in
(26b)), in fact it corresponds to the fricative. At this stage of his grammar, Amahl
substituted initial /f/ with [w] (29a). The [w] is unlikely to correspond with the /l/
or /r/, because the substitutes for each of these sounds assimilated in place to a
following consonant (as in (26a,c) and (29b)).
(29) Amahls substitutes for the individual components of /fl/ and /fr/ clusters
a. Substitutes for initial /f/
face we:t (4) fog w (5)
five waiv (11)
b. Substitutes for initial /r/
red dt (1) rock k (2)
ram wm (2)
It is clear that Amahls substitutes for /fl/ and /fr/ patterned with his substitutes for
singleton /f/, not with those for the singleton liquids. Thus Amahls repair for an
f-initial cluster always required the violation of at least two faithfulness constraints,
but by retaining a substitute for the /f/ rather than the liquid, he eliminated the
need for a third repair, place assimilation.
Finally, a multiply-unfaithful output may be allowed to surface when the
markedness violations are non-local, and so there is no single fell-swoop repair that
could eliminate the entire marked structure (see Jesney 2011 for an in-depth discus-
sion of this issue). The repairs for /l/ and /r/ before a velar consonant in (26c) and
(29b), respectively, illustrate this possibility. These forms contain two marked struc-
tures: an initial liquid and an alveolar consonant followed by a velar. Amahl exhibited
a pattern of consonant harmony that often induced long-distance place assimilation
to repair alveolar+velar sequences. Examples of velar harmony are in (30).
5. Interestingly, /sl/ clusters as in (15b) above seem to be an exception to this generalization,

as they were often reduced to an [l]. Amahl does not exhibit this repair for /fl/ clusters or for
singleton /l/ until somewhat later.
(30) Velar harmony

dark a:k (1) ding-dong i (1)
take e:k (2) talk :k
like aik (1) ring i (1)
The two marked structures are non-local. Assuming that deletion of the onset is
not an option (as mentioned above), no other fell-swoop repair remains. Replac-
ing the liquid with an alveolar stop would not solve the consonant harmony prob-
lem, and deleting or changing the place of the final velar would not eliminate the
initial liquid. Thus the doubly-unfaithful repair is necessitated.
5. Discussion and conclusion
This chapter has presented a cohesive set of data that indicate that many of the
repairs in Amahls grammar achieved the same end: multiple unfaithful mappings
were not permitted to combine when some other option was available. The con-
cept of a faithfulness conspiracy was introduced to describe cases in which several
repair processes worked together to avoid outputs that are unfaithful in specific
ways or places. Evidence was also presented explaining why multiple repairs were
sometimes allowed to combine. This occurred when a fell-swoop repair was not
available, or when some other process in the grammar was prioritized. Further
research is necessary to determine whether all of Amahls multiple-repair struc-
tures fell into one of these two categories.
It is uncertain how long Amahl retained his avoidance of doubly-unfaithful
repairs determining at what point in his development this changed, and what the
next set of repairs were, could shed light on the acquisition process. Jesney (2011)
argues that in a learning account of HG, fell-swoop constraints like deletion would
be predicted as repairs for doubly-marked segments, and singly-marked segments
should surface faithfully sooner than the doubly-marked counterparts. Farris-
Trimble (2008) suggested that given the gradual learning algorithm (Boersma
1998; Boersma & Hayes 2001), all children would be expected to go through a CFE
stage. Whether it can be documented is another question. Moreover, the question
remains whether other children have a similar tendency (to avoid multiple re-
pairs), or whether this is child-specific.
From a theoretical point of view, Amahls CFEs allowed evaluation of two
competing theoretical constructs, LC and HG. Both provided a straightforward
account of the CFEs. One factor may have tipped the scale slightly in favor of the
LC account, however this account was consistent with the faithfulness conspira-
cy Amahls CFEs and his chain shift can all be explained with one construct.
Because HG is not able to account for the chain shift, it offers a less coherent
account. Additionally, the family of LC constraints converges with the idea of

triggering constraint families in faithfulness conspiracies.
While LC might have had a slight advantage in accounting for Amahls faithful-
ness conspiracy, it is not necessarily the case that all faithfulness conspiracies are
better described with LC. The positional faithfulness and reduplication cases pre-
sented in 3 can be explained in either OT (without recourse to LC) or HG. (Though
Jesney (2011) argues that HG presents a more elegant description of positional effects
than positional faithfulness within OT.) Further research will be needed to determine
the relative merits of the two theories to faithfulness conspiracies in general.
The analysis of Amahls faithfulness conspiracy also raises several theoretical
questions about the structure of Con. In each of Amahls CFEs, segmental deletion
was the fell-swoop repair that avoided a multiply-unfaithful output. In some cases,
though, the pattern was more specific: in 2.2, for instance, obstruent deletion was
necessary to avoid the co-occurrence of nasal deletion and devoicing, and in 2.3,
s-deletion avoided the co-occurrence of sonorant-deletion and stopping and/or
harmony. It was thus necessary to split Max into a family of constraints banning
the deletion of certain types of segments, so that the fell-swoop deletion (obstruent
deletion in both of these cases) could be ranked or weighted separately from con-
straints against other types of deletion. Alongside this family of Max constraints is
a family of Ident constraints, which militate against the change of a specific fea-
ture within a segment. This seems dangerously close to a duplication of constraints,
but both sets are necessary in this analysis. A fell-swoop repair is, by definition, a
drastic repair, as it must eliminate multiple marked structures at once. Deletion is
one of the few repairs that serves this purpose. But not all segmental deletions are
equal. Some eliminate doubly-marked structures, while others eliminate only a
singly-marked segment, leaving another marked structure behind. It is important
to have a way to separate the fell-swoop repairs from the other available repairs.
Finally, the faithfulness conspiracy construct proposed here was supported by
three examples: Amahls avoidance of multiply-unfaithful outputs, Ibibios avoid-
ance of unfaithfulness in the strong root-initial position, and Dakotas avoidance
of unfaithfulness between the base and the reduplicant. There are countless ex-
amples in the literature of markedness conspiracies, and further reflection on the
nature and composition of faithfulness conspiracies is needed to determine wheth-
er they are equally prevalent.
Acknowledgements
Many thanks to Dan Dinnsen for introducing me to child phonology in general and
Amahl in particular, and for making me love phonology, even when I thought it was
harder than syntax. This work was supported in part by grants to the University of
Iowa from the National Institutes of Health (DC000242, DC008089, and DC011669)
and a Presidents Research Start-up Grant from Simon Fraser University.
References
Beckman, J. 1998. Positional Faithfulness. PhD dissertation, University of Massachusetts at

Amherst.
Boersma, P. 1998. Functional Phonology: Formalizing the Interaction Between Articulatory and
Perceptual Drives. The Hague: HAG.
Boersma, P. & Hayes, B. 2001. Empirical tests of the gradual learning algorithm. Linguistic In-
quiry 32: 4586.
Clements, G.N. 1990. The role of the sonority cycle in core syllabification. In Papers in Labora-
tory Phonology, 1: Between the Grammar and Physics of Speech, J. Kingston & M. Beckman
Dinnsen, D.A. 2011. On the unity of childrens phonological error patterns: Distinguishing
symptoms from the problem. Clinical Linguistics and Phonetics 25: 968974.
Dinnsen, D.A., OConnor, K.M. & Gierut, J.A. 2001. The puzzle-puddle-pickle problem and the
Farris-Trimble, A.W. 2008. Cumulative Faithfulness Effects. PhD dissertation, Indiana Univer-
sity at Bloomington.
Farris-Trimble, A.W. 2009. Weighted constraints and faithfulness cumulativity in phonological
acquisition. In Proceedings of the 33rd Annual Boston University Conference on Language
Development, J. Chandlee, M. Franchini, S. Lord & G.-M. Rheiner (eds), 151162. Somerville
Farris-Trimble, A.W. 2010. Nothing is better than being unfaithful in multiple ways. In Proceedings
from the Annual Meeting of the Chicago Linguistics Society 44, M. Bane, J.J. Bueno Holle, T.
Grano, A.L. Grotberg & Y. McNabb (eds), 7993. Chicago IL: Chicago Linguistics Society.
Gnanadesikan, A. 2004. Markedness and faithfulness constraints in child phonology. In Fixing
Priorities: Constraints in Phonological Acquisition, R. Kager, J. Pater & W. Zonneveld (eds),
73108. Cambridge: CUP.
Jesney, K. 2011. Cumulative Constraint Interaction in Phonological Acquisition and Typology.
PhD dissertation, University of Massachusetts at Amherst.
Jesney, K. & Tessier, A-M. 2011. Biases in Harmonic Grammar: The road to restrictive learning.
Natural Language and Linguistic Theory 29: 251290.
Kirchner, R. 1996. Synchronic chain shifts in optimality theory. Linguistic Inquiry 27: 341350.
Kisseberth, C. 1970. On the functional unity of phonological rules. Linguistic Inquiry 1: 291306.
Legendre, G., Miyata, Y. & Smolensky, P. 1990a. Harmonic Grammar A formal multi-level
connectionist theory of linguistic well-formedness: Theoretical foundations. In Proceedings
of the Twelfth Annual Conference of the Cognitive Science Society, 388395. Hillsdale NJ:
Lawrence Erlbaum Associates.
Legendre, G., Miyata, Y. & Smolensky, P. 1990b. Harmonic Grammar A formal multi-level
connectionist theory of linguistic well-formedness: An application. In Proceedings of the
Twelfth Annual Conference of the Cognitive Science Society, 884891. Hillsdale NJ: Lawrence
Erlbaum Associates.
Legendre, G., Sorace, A. & Smolensky, P. 2006. The Optimality Theory Harmonic Grammar
connection. In The Harmonic Mind: From Neural Computation to Optimality-Theoretic
Grammar, P. Smolensky & G. Legendre (eds), 903966. Cambridge MA: The MIT Press.
ubowicz, A. 2002. Derived environment effects in optimality theory. Lingua 112: 243280.
Marantz, A. 1982. Re reduplication. Linguistic Inquiry 13: 435482.
McCarthy, J.J. & Prince, A.S. 1995. Faithfulness and reduplicative identity. In University of Mas-
sachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory, J. Beckman, L.
Walsh Dickey & S. Urbanczyk (eds), 249384. Amherst MA: GLSA.
McCarthy, J.J. 2002. A Thematic Guide to Optimality Theory. Cambridge: CUP.
McMurray, B., Cole, J.S. & Munson, C. 2011. Features as an emergent product of computing
perceptual cues relative to expectations. In Where Do Phonological Features Come From?
[Language Faculty & Beyond 6], G.N. Clements & R. Ridouane (eds), 197236. Amsterdam:
John Benjamins.
McMurray, B. & Jongman, A. 2011. What information is necessary for speech categorization?
Harnessing variability in the speech signal by integrating cues computed relative to expec-
tations. Psychological Review 188: 219246.
Moreton, E. & Smolensky, P. 2002. Typological consequences of local constraint conjunction. In
Proceedings of the 21st West Coast Conference on Formal Linguistics, L. Mikkelsen & C. Potts
(eds), 306319. Cambridge MA: Cascadilla Press (Available on Rutgers Optimality Archive,
ROA-525).
Pater, J. 1999. Austronesian nasal substitution and other NC effects. In The Prosody Morphology
Interface, H. van der Hulst, R. Kager, & W. Zonneveld (eds), 310343. Cambridge: CUP.
Pater, J. 2001. Austronesian nasal substitution revisited. In Segmental Phonology in Optimality
Theory: Constraints and Representations, L. Lombardi (ed.), 159182. Cambridge: CUP.
Pater, J. & Barlow, J. 2003. Constraint conflict in cluster reduction. Journal of Child Language 30:
487526.
Prince, A.S. & Smolensky, P. 1993/2004. Optimality Theory: Constraint Interaction in Generative
Grammar. Malden MA: Blackwell.
Smolensky, P. 1995. On the structure of the constraint component Con of UG. Ms, University of
California at Los Angeles. ROA 86.
Smolensky, P. 1996. The initial state and richness of the base in Optimality Theory. Technical
report JHU-CogSci-96-4, Department of Cognitive Science, The Johns Hopkins University,
Baltimore MD.
Smolensky, P. 2006. Optimality in phonology II: Harmonic completeness, local constraint con-
junction, and feature domain markedness. In The Harmonic Mind: From Neural Computa-
tion to Optimality-Theoretic Grammar, P. Smolensky & G. Legendre (eds), 27160. Cam-
bridge MA: The MIT Press.
Toscano, J. & McMurray, B. 2010. Cue integration with categories: Weighting acoustic cues in speech
using unsupervised learning and distributional statistics. Cognitive Science 34: 434464.
Superadditivity and limitations on syllable
complexity in Bambara words

University of Maryland CASL and Indiana University
This paper explores superadditivity effects in natural language by considering

three interrelated phenomena in Colloquial Bambara (CB). The premise of
superadditivity is that although marked structures are accommodated in a
system, particular structures cannot co-occur in a given domain. This arises
because the simultaneous, additive violation of constraints within a domain
arguably incurs an additional penalty. Thus, languages may limit the number
of phonologically complex structures in a domain. We consider superadditivity
in CB, which places strict limitations on the type and distribution of complex
syllables within a word. We also discuss these data as they relate to models of
phonological acquisition which maintain that outright bans on multiple complex
structures do not occur in adult language yet are frequently encountered in
developing languages.
0. Introduction
Languages may act to limit the type and number of phonological complexities
permitted in a word. While a language may not disallow all instances of phono-
logical complexity, complex structures may occur less frequently than otherwise
expected. Such tendencies and restrictions have been attributed to detrimental
superadditivity effects (e.g. Albright 2008, 2009) that would arise when certain
combinations of markedness constraints are violated within a single domain (e.g.
a word). The general premise of superadditivity is that although marked structures
may be individually accommodated in a system, there exist some marked struc-
tures that cannot co-occur in a given domain. For example, Albright (2009) dem-
onstrates this in Lakhota, which avoids lexical roots containing multiple instances
of structures that are relatively marked in the language (e.g. fricatives and conso-
nant clusters).
Scholars have argued that superadditivity arises because the simultaneous, ad-
ditive violation of constraints on certain marked structures within a single system
(or domain) incurs an additional penalty. That is, the sum of the violations plus
this additional penalty is superadditive in that 1+1 > 2. This proposition diverges
from standard conceptions of Harmonic Grammar (HG: e.g. Smolensky &
Legendre 2006) in which phonological constraints on markedness (and faithful-
ness) are assigned weights that sum only additively (rather than superadditively)
to yield a total violation score for a particular candidate under evaluation. Additive
models like HG account for gang effects or cumulativity effects (e.g. Coetzee &
Pater 2008; Farris-Trimble 2008; Green & Farris-Trimble 2010, among others)
where the violation of two constraints is more costly than the individual violation
of one constraint or the other, i.e. 1+1=2.1
HG is an alternative to Standard Optimality Theory (OT: Prince & Smolensky
1993/2004), a core tenet of which is strict domination of constraints. Strict domi-
nation is such that only a single constraint is ultimately responsible for evaluating
a candidate, effectively masking the effect of constraints ranked lower in the hier-
archy. Because of strict domination, OT does not easily address gang effects. This
limitation prompted the proposal of Local Constraint Conjunction (LCC: e.g.
Downing 1998; ubowicz 2002, 2003, 2005; Moreton & Smolensky 2002; Smolen-
sky 1995, 2006) wherein two constraints are conjoined to create a new constraint
that is violated when both its constituent constraints are violated within a defined
local domain. The formal mechanism of LCC (still in the spirit of strict domina-
tion) requires that the locally-conjoined constraint be ranked above its component
constraints in order to have its intended effect. Some scholars have argued that
instantiations of LCC are not grammatically intuitive and may (in some instances)
predict unattested phonological grammars (e.g. Farris-Trimble 2010; Ito & Mester
1998; Pater, Bhatt & Potts 2007). While our analysis of superadditivity effects em-
ploys a type of local conjunction of constraints, we aim to show that these con-
straints do not wield quite the same power that analogous constraints have in
typical LCC analyses.
This paper has several aims. First, we shed light on the types of superadditive
effects found in natural language by considering three interrelated phenomena in
1. There is growing interest in methods of additive and superadditive evaluation in formal

phonology. Examples of recent work include Albright, Magri & Michaels (2008); Coetzee &
Pater (2008); Farris-Trimble (2008); Khanjian, Sudo & Thomas (2010); Legendre, Sorace &
Smolensky (2006); Pater, Bhatt & Potts (2007); and Pater (2009, to appear).
Superadditivity in Bambara words
Colloquial Bambara (CB).2 These phenomena relate to the fact that while certain
complex syllables can co-occur in CB, the language strictly limits the type and
distribution of these syllables within a word. That is, CB avoids words that contain
multiple instances of some, but not all, complex syllable types. To demonstrate
this, we employ an HG framework supplemented by superadditively conjoined
constraints to demonstrate that the additive effect of violating certain combina-
tions of constraints is more costly, phonologically-speaking, than the sum of their
individual parts. Second, we propose that it may be necessary to redefine Legendre,
Sorace and Smolenkys (LSS: 2006) characterization of a particular type of super-
additivity effect known as superlinearity to include faithfulness constraints in light
of our CB data. Lastly, we discuss how these CB data relate to models of phono-
logical acquisition (e.g. Albright, Magri & Michaels 2008) which maintain that
outright bans on multiple complex structures do not occur in adult language yet
are frequently encountered in developing languages (i.e. in language acquisition).
We argue that additive interactions resulting in such a ban do occur in CB but are
most likely unstable and will probably fail to persist.
1. Overview of minimization in Colloquial Bambara
Green and Diakite (2008) first reported that CB has synchronically developed
CCV, CVC, and CVV syllables. CB differs from Standard Bambara (SB) where syl-
lables generally conform to a maximal CV template. Complex syllables enter CB
via two processes, Vowel Syncope (VS) and Velar Consonant Deletion (VCD),
which are driven by the interaction of competing constraints on segmental
markedness (*Peak) vs. faithfulness (Max) and syllable margin phonotactics
(Green 2010). Bambara has a typical seven-vowel system in which high vowels are
preferentially, but not exclusively, deleted.3 Via VS, a short vowel of any type is
removed from a syllable in any word position under the appropriate conditions, as
2. CB is an urban variety of Bambara (Bamana or Bamanankan, ISO:bam) spoken in Bamako,

Mali. This language, described in Green (2010), has emerged or diverged from the standard,
more phonologically-conservative urban variety of the language. Standard Bambara (SB) gener-
ally allows only maximal CV syllables. CB, on the other hand, permits a wider variety of syllable
shapes, including CCV, CVC, and CVV. The data in this paper are drawn from Green (2010) and
our subsequent research on Bambara.
3. VS can remove vowels of any height from a particular syllable or word position, but pho-
notactics and metrical restrictions govern the process (Green 2010). There is independent evi-
dence that high vowels may delete in certain Standard Bambara words to create phonetic CCV
syllables (e.g. Dumestre 2003).
long as phonotactic constraints on syllable margins are not compromised.4 Sylla-

bles formed via VS must have a sonorant as the second member of a branching
onset (in CCV syllables) or as a singleton coda (CVC syllables).
VCD operates alongside VS, producing CVV syllables with derived long vow-
els by removing intervocalic velar stops flanked by identical vowels. Derived CVV
syllables differ from syllables with phonemic long vowels which are limited in their
distribution to morpheme-initial positions and may be absent in the synchronic
phonologies of some speakers (Dumestre 2003). Whether by the action of VS or
VCD, there is an overall drive towards minimization in CB that typically results in
a single segmental deletion, effectively reducing a word by one syllable. We illus-
trate below that in a restricted set of instances, a second deletion is permitted.
Typical instances of single deletion via VS are in (1). For each example,
we provide a word in SB with its CB counterpart alongside an English gloss.
Where appropriate, we give illicit forms for comparison, marked by *, follow-
ing convention.
(1) Vowel syncope patterns from input trisyllabic words
Standard (SB) Colloquial (CB) Gloss
a. [ka.bi.la] [ka.bla] *kbi.la tribute
b. [s.f.n] [sa.fn] *sfa.n soap
c. [de.li.ko] [del.ko] *dli.ko habit
d. [fa.rC.mA] [far.mA] *fri.ma brave
e. [si.la.m] [sla.m] *sil.m Muslim
f. [du.l.ki] [dl.ki] *dul.ki shirt
g. [ca.pa.lo] [ca.plo] millet beer
h. [na.ma.sa] [nam.sa] banana
i. [a.la.ma] [la.ma]/[al.ma] spoon
j. [ke.le.ku] [kel.ku]/[kle.ku] to stumble
k. [su.ru.ku] [sur.ku]/[sru.ku] hyena
l. [ka.ba.no] [ka.ba.no] asylum
m. [sa.ba.ti] [sa.ba.ti] stable
n. [d.s.kG] [d.s.kG] heart
These data show a variety of possible reductions that rely on syllable margin pho-
notactics. They show (i) word-internal CCV syllables created by [+hi] vowel loss
4. We assume that the non-syncopated form of words characteristic of SB are lexical and are
the CB input. A non-syncopated form of a given word will surface in certain constructions when
syncope cannot occur for reasons of metrical structure (Green 2010). Tonal melodies on synco-
pated forms are also clearly derived from those in SB. For more on the properties of syllable
structure and phonotactics in CB and its relation to the Split Margin Approach to the syllable,
see Green (2010) and Green, Davis, Diakite & Baertsch (2012, to appear).
(1ab); (ii) word-internal syllable contact sequences created by [+hi] vowel loss
(1cd); (iii) word-initial CCV syllables created by [+hi] vowel loss (1ef); and (iv)
analogous instances involving [hi] vowel loss (1gi). Examples (1jk) show vari-
ation between CCV and CVC outcomes where identical vowels can be removed to
create syllables with licit margin phonotactics. Examples (1ln), however, show
that VS will not occur when appropriate phonotactic conditions are not met. These
reductions occur analogously in shorter words, as in (2).
(2) Vowel syncope from input disyllabic words
Standard (SB) Colloquial (CB) Gloss
a. [si.rA] [srA] *sir to scar
b. [t.n] [tn] *tn taboo
c. [bo.li] [bol] *bli to run
d. [se.li] [sel] *sli prayer
e. [bi.li] [bli]/[bil] roof
f. [di.bi] [di.bi] *dbi darkness
g. [ki.ti] [ki.ti] *kti trial
Examples (2ab) show that monosyllabic CCV words are created via VS through
high or non-high vowel loss. (2cd) show that a CVC syllable with a word-final
[-nasal, -continuant] coda is also accommodated, i.e. [l]. (2e) shows that variation
between these two reductions occurs when the appropriate conditions are met. (2fg)
show disyllabic words that cannot be reduced because of margin phonotactics.
There are other instances in CB where either VS cannot remove a vowel or
where the process competes with VCD. Words in (3) show VCD removing an in-
tervocalic velar stop flanked by identical vowels. In these words, phonotactic re-
strictions preclude VS.
(3) Velar Consonant Deletion
Standard Colloquial Gloss
a. [si.i] [sii] to sit
b. [m.k] [m] person
c. [t.] [t] name
d. [sa.a] [saa] sheep
e. [du.u] [duu] village
f. [s.k.li] [s.li] *[s.kli] infection
g. [s.k.ma] [s.ma] *[s.kma] morning
Green, Davis, Diakite and Baerstch (GDDB: 2012) argue that the domain of ap-
plication for VCD is a maximally disyllabic prosodic foot constructed at the left
edge of a word. Thus, in (3), VCD acts on an intervocalic velar stop within this
domain when it is flanked by identical vowels, e.g. 3f: (s.k) (li) s.li. VS could
not have applied in the words in (3ae), as it would create either (i) a disallowed
obstruent-obstruent onset in a CCV syllable or (ii) an impermissible obstruent
coda in a CVC syllable.
GDDB also suggest that VS and VCD compete with one another to remove
segmental content within the foot domain. In some instances, this competition
ends in a stalemate in which either process (but not both) may apply. This is shown
in (4).5 The crucial difference between the examples in (4) where there is variation
and those in (3fg) where there is no variation is that in (4), the vowels in the ini-
tial foot are both high, but they are non-high in (3fg).
(4) Variation between Velar Consonant Deletion and Vowel Syncope
a. [(si.ki) (lA)] [sii.lA]/[si.klA] *ski.la chair
b. [(su.ku) (na)] [suu.na]/[su.kna] *sku.na urine
c. [(su.u) (ri)] [suu.ri]/[su.ri] *su.ri pre-fasting meal
Additional data suggest a role for prosodic feet in VS. This relates to variation in
the creation of CCV or CVC syllables, as in (5). Examples (5ab) illustrate that
when identical vowels within a prosodic foot are targets for VS, there is variation
between CCV and CVC reductions. In (5c), when deletion targets are separated by
a foot boundary, only a single outcome is possible.
(5) Variation in Vowel Syncope
a. (b.r) (t) br.t/br.t to tear apart
b. (mu.so) (k.r) (ba) mu.so.kr.ba/mu.so.kr.ba wise woman
c. (d.ki) (li.da) d.kli.da, *d.kil.da to dance
The data above show the basic mechanism and outcomes of VS and VCD and their
proposed link to prosodic foot structure, which are critical to the discussion in
sections below. They also offer insight into the distribution of complex syllables in
CB. By the action of VS, CB admits CV.CCV, CVC.CV, and CCV.CV words. By
VCD, CB introduces CVV.CV words.6 To delve more deeply into the possibility of
complex syllable co-occurrences, we turn to longer words, such as compounds
and other more morphologically complex derivatives.
5. Others offer independent perspectives on certain characteristics of foot structure in

Mande languages (Vydrine 2010) and in Bambara tonology (e.g. Leben 2002, 2003; Weidman
& Rose 2006). These and the current approach differ most apparently in their assumptions of
foot headedness.
6. While these outcomes speak only to the distribution of a complex syllable alongside an-
other simplex syllable within a single word, note that *CV.CVV words are conspicuously absent
from these data. Green (2010) motivates this gap in terms of the avoidance of iambic (light +
heavy) foot structure.
2. Distribution of complex syllables in Colloquial Bambara
VS and VCD are constrained by Bambara foot structure and reference the pro-
sodic foot domain when selecting and acting upon their respective targets. The
examples above consider the outcomes of VS and VCD when acting alone or in
competition with one another for a deletion target within a single disyllabic foot.
Compounds and certain morphological derivatives offer a look beyond a single
domain where VS and VCD have deletion targets located in adjacent feet. Longer
words offer a variety of potential deletion targets and provide the opportunity to
witness additional ways that VS and VCD interact with and/or compete with one
another. By exploring such CB words, we can observe the potential for and per-
missibility of speakers to introduce multiple deletions and therefore multiple com-
plex syllables into Bambara. Data below illustrate that the introduction of multiple
complex syllables to a single word is highly regulated due to interactions between
certain phonological constraints.
When a word contains VS and VCD targets in adjacent feet, a single deletion
occurs by the action of VCD, regardless of the foot occupied by the VCD target.
Thus, VCD applies to the exclusion of VS whenever possible; it is the preferred
process of reduction in CB. Data in (6) illustrate this choice and show that both
CVV.CV.CV and CV.CV.CVV words are possible in CB. In the former, VCD acts
within the leftmost of two disyllabic feet, while in the latter, VCD acts on a target
in the rightmost foot.
(6) Competition between Velar Consonant Deletion and Vowel Syncope
a. (se.li) (sa.ka) (se.li) (saa) *sel.sa.ka, *sel.saa sacrificial sheep
b. (n.r) (mu.u) (n.r) (muu) *nr.mu.u, *nr.muu yellow
c. (si.i) (b.l) (sii) (b.l) *si.i.blo, *sii.blo institution
d. (.k) (s.r) () (s.r) *.k.sr, *.sr to find one another
Words containing combinations of complex syllables that would be generated by
VCD and VS acting together (e.g. CVC.CVV or CVV.CCV) are not found. This
may be surprising considering that in some instances (e.g. 6a and 6d), individual
constituents of a compound are reduced by their respective means in isolation, as
in (7). However, when the constituents are compounded in (6), VCD applies to the
exclusion of VS.
(7) VCD is preferred over VS

a. (se.li) (sel) prayer
b. (sa.ka) (saa) sheep
c. (se.li) (sa.ka) (se.li) (saa) *sel.sa.ka, *sel.saa sacrificial sheep
d. (.k) () together
e. (s.r) (sr) to find
f. (.k) (s.r) () (s.r) *.k.sr, *.sr to find one another
We can say more about the preference for reduction via VCD by considering a wid-
er range of word shapes. When compounds have potential deletion targets for VCD
in adjacent feet, only a single reduction by VCD is permitted. We show this in (8).
(8) Single occurrence of VCD
a. (si.i) (sii) to sit
b. (.k) () together
c. (s.) (K.kK) (sii) (K.kK) neighbor
d. (npo.ko) (npoo) cloth
e. (ti.i) (tii) master
f. (npo.ko) (ti.i) (npoo) (ti.i) maiden
The data in (7) show the preferentiality of reduction via VCD, yet (8) shows that
VCD is limited in its application; it does not act twice within a word. The inability
for VCD to act twice suggests that although VCD is a preferred process of reduc-
tion (cf. 6 and 7), CVV.CVV words are disallowed in CB. The preferred reduction
yields a complex syllable in the leftmost foot, closer to the left edge of the word.
One possibility to account for this is that higher level metrical restrictions pre-
clude the co-occurrence of adjacent heavy syllables.
We turn next to words that have adjacent feet with potential VS deletion tar-
gets. What we encounter is more complex, yet similar, to that observed in (8) for
VCD. For words with multiple VS targets, the patterns of complex syllable forma-
tion rely on the types of vowels available and selected for reduction. The simplest
cases concern words with adjacent domains containing non-high vowel VS tar-
gets. In these words, like the representative case of (9), only a single deletion gen-
erating a single complex syllable is permitted. The complex syllable generated is
again in the leftmost foot, creating complexity at the left edge of the word. Like the
individual constituents in (8ab, de), the constituent words in (9ab) are free to
reduce by VS in isolation. However, when the constituents are compounded in
(9c), only a single reduction is possible.
(9) Standard Colloquial Gloss

a. (n.r) (nr) nr tree
b. (ko.lo) (klo) pit
c. (n.r) (ko.lo) (nr.ko) (lo) *(nr.klo) nr pit
This outcome is different from words like those in (10a,d) where either (i) a high
vowel and a non-high vowel or (ii) two high vowels are deletion targets for VS,
respectively. Here, CB permits multiple deletions within a single word. The out-
come is CCV.CCV. Note that this outcome is possible in both monomorphs (10a)
and compounds (10d).
(10) Standard Colloquial Gloss
a. (bi.la) (ko.ro) (bla.kro) young boy
b. (bi.la) (bla) to let go
c. (si.ra) (sr8) road
d. (bi.la) (si.ra) (bla.sra) to travel a short distance with someone
(9) and (10) show that CB permits multiple deletions to create a single CCV.CCV
word; however the process is tightly constrained such that a second deletion is al-
lowed only when at least one of the two VS targets is a high vowel. This result
raises several important questions. First, if VCD is preferred to VS to achieve min-
imization, why is it that VCD can only apply once within a word, while VS can
apply twice? Secondly, how does Bambara regulate the highly selective process of
minimization such that only words containing certain combinations of vowels are
potentially subject to deletion? Lastly, what formal mechanism is best employed to
capture these complex relationships? To address these questions, we entertain po-
tential analyses of these minimizations in OT vs. HG.
3. Formalizing restrictions on multiple complexities
Other work on CB argues that minimization is the result of the antagonistic rela-
tionship between constraints on syllable peak markedness (*Peak) and corre-
sponding faithfulness constraints (Max) that resist minimization (Green 2010;
Green et al. to appear). The ranking of *Peak constraints above Max constraints
compels CB reduction. Furthermore, *Peak[+hi] is ranked above *Peak[hi],
capturing the preference to remove high vowels via VS. The role of phonotactics in
VS is attributed to constraints on permissible syllable margins, following the Split
Margin Approach to the syllable (SMA: e.g. Baertsch 2002; Baertsch & Davis 2009;
Davis & Baertsch 2011). These constraints capture the parallel preferences for con-
sonants of specific sonorities to be found in certain syllable margin positions. M 1
(Onset) Rhyme
Nucleus (Coda)
M1 (M2) M2 (M1)
Figure 1. Split-Margin syllable (adapted from Baertsch 2002)
and M2 constraints correspond to syllable margin positions in a split margin syl-

lable where M1 is a syllable-initial consonant (favoring lower sonority) and M2 is a
coda or second member of an onset (favoring higher sonority). For example,
*M2/L is a constraint banning liquids from an M2 margin position. M1 and M2
constraints may also be conjoined in a local domain (i.e. syllable or word) thereby
dictating specific consonant-consonant sequences permitted in complex onsets
and across syllable contact sequences, respectively. Relevant to our tableaux is
*M1&*M2/L, which bans complex syllable margins containing some consonant in
an M1 position followed by a liquid in M2 position. An adaptation of the split mar-
gin syllable is in Figure 1.
While CB favors minimization, it places strict upper limits on the types and
number of complex syllables allowed in a word. These limits are problematic for
strict domination, which effectively overpredicts segmental deletion and the cre-
ation of complex syllables in CB. In Standard OT, strict domination analysis pre-
dicts that optimal outcomes in CB have fewer syllable peaks in all instances.
Tableau (11) shows that strict domination incorrectly predicts a doubly reduced
candidate () while the attested winner () incurs a fatal violation.
(11)
/nrkolo/ *CCC
*Peak Max [*M1& Wd[*M1& *M2/L
[hi] [hi] *M2/L *M2/L
a. nr.ko.lo ***! * * *
b. n.r.ko.lo ***!*
c. nr.klo *! ** ** * ** **
d. nr.klo ** ** ** ** **
This approach differs from HG (e.g. Smolensky & Legendre 2006) which assigns
constraints language-specific weights rather than a ranking. Note that HG con-
straint weights are arbitrary; however the ratio between them is the key factor in
the analysis (Farris-Trimble 2008:19).7 The cumulative violation of lower-weight
constraints can outweigh or gang up on a higher-weight constraint, thereby mask-
ing its effects. This better captures what occurs in CB, where the cumulative viola-
tion of margin markedness constraints is a less favorable (or less harmonic)
outcome than what is attested, i.e. more syllable peaks but fewer complex syllables.
The candidates in (11), now evaluated in HG, are shown in (12).
(12)
*Pk Max [*M1& Wd[*M1&
*CCC *M2/L
/nrkolo/ [hi] [hi] *M2/L *M2/L H
w=5.5 w=.2
w=4.25 w=2.5 w=.6 w=.6
a. nr.ko.lo 3 1 1 1 16.05
b. n.r.ko.lo 4 17
c. nr.klo 1 2 2 1 2 2 21.2
d. nr.klo 2 2 2 2 2 16.3
Tableau (12) more accurately schematizes the competition between CB constraints.

Although minimization is driven and controlled by the relationship between the
*Peak and Max constraints, the conflict between *Peak constraints (creating
complex syllables) and *M1 and *M2 (resisting complex syllable creation) now
contributes to the observed outcome.
While (12) correctly predicts the observed outcome in words where introduc-
ing multiple complexities are avoided, we must now attend to the opposing state of
affairs where words with multiple complexities are permitted. We begin with words
for which multiple reductions arise via the removal of high vowels, as in (13). By
employing HG, we view the effect of constraints banning different syllable peaks
(high vs. non-high vowels) interacting with other constraints in the grammar. Be-
cause high vowels are preferred targets for VS, *Peak[+hi] has a comparatively
higher weight than *Peak[hi]. Likewise, there is a lower penalty to remove high
vowels than non-high vowels, so Max[+hi] has a comparatively lower weight than
7. It is convention in HG that weights are assigned positive numbers (whole integers or deci-
mals). Candidates accumulate violations indicated by whole negative numbers which sum to yield
a total harmony score H. The candidate with the lowest absolute value harmony score is optimal.
The weights proposed in this paper were verified in OT-Help (Becker, Pater & Potts 2007).
Max[hi]. These relationships are key in correctly predicting outcomes like (13d),
where by removing two high vowels, CB creates a word with two CCV syllables.
(13)
[*M1& Wd[*M1&
*Pk[+hi] *Pk[hi] Max[+hi] *M2/L
/bilasira/ *M2/L *M2/L H
w=5 w=4.25 w=2 w=.2
w=.6 w=.6
a. bi.la.si.ra 2 2 18.5
b. bla.si.ra 1 2 1 1 1 1 16.9
c. bi.la.sra 1 2 1 1 1 1 16.9
d. bla.sra 2 2 2 2 2 15.3
These constraints, weighted as above, also effectively account for the second out-
come involving multiple complex syllables in (14). This involves removing one
high and one non-high vowel to create a CCV.CCV word, where the benefit of
removing a non-high peak outweighs retaining it, if only slightly.8
(14)
*Pk *Pk Max Max [*M1& Wd[*M1
*M2/L
/bilakoro/ [+hi] [hi] [hi] [hi] *M2/L & *M2/L H
w=.2
w=5 w=4.25 w=2.5 w=2 w=.6 w=.6
a. bi.la.ko.ro 1 3 17.75
b. bla.ko.ro 3 1 1 1 1 16.15
c. bi.la.kro 1 2 1 1 1 1 17.4
d. bla.kro 2 1 1 2 2 2 15.8
Thus far, evaluation by HG accurately predicts the attested outcomes in (13) and
(14) where multiple instances of VS create CCV.CCV words. Compared to Stan-
dard OT, HG also arrives at the attested outcome in (12) where only a single reduc-
tion is possible in words containing two non-high vowel deletion targets.
While this is a good start, we must probe HGs predictions for additional word
shapes. For example, there are words like (7f), in which, all else being equal, one
might predict a double deletion, yet only a single deletion is permitted. We propose
8. An additional detail relevant here is that M2 consonants in adjacent CCV syllables must not
be identical. While it is beyond the scope of this paper to explore this, the restriction is likely due
to a requirement for dissimilar M2 consonants within a word.
that what is at issue in this and related instances is the particular combination of
constraint violations that would be necessary to accommodate this double reduc-
tion (in this case, violations of Max[hi] and Max-K). We argue below that this
combination of constraints and two others in CB cannot co-occur within particu-
lar domains, and as a result, their violations combine superadditively to create a
more harmonically complete phonological grammar.
4. Superadditive ordering
Smolensky (2006) and LSS (2006) entertain the hypothesis that some grammars are
non-harmonically complete such that neither strict domination nor a weighted har-
monic analysis accurately predict all attested outcomes. As such, neither ranking nor
simple summation of constraint violations arrives at correct, grammatical predic-
tions. These authors suggest that non-harmonically complete phonologies necessi-
tate the introduction of local constraint conjunction via superadditive ordering
relationships between some constraints. In superadditive relationships, the penalty
for violating a combination of individual constraints is greater than the sum of its
parts, e.g. 1+1>2. Although we return to discussion of such constraint conjunctions
below, it is important to note that the constraints proposed here introduce subtle,
rather than sweeping, effects of cumulativity in CB. To begin, consider first the pre-
dicted yet unattested HG outcome in (15) utilizing the constraints employed above.9
(15)
*Pk Max [*M1& Wd[*M1&
Max-K *M2/L
/ksr/ [hi] [hi] *M2/L *M2/L H
w=1.5 w=.2
w=4.25 w=2.5 w=.6 w=.6
a. .s.r 3 1 14.25
b. .k.s.r 4 17
c. .sr 2 1 1 1 1 1 13.9
d. .k.sr 3 1 1 1 1 16.65
(15) shows that the doubly reduced (15c) is predicted, rather than the attested
winner (15a). It appears, therefore, that HG encounters the same issue of overpre-
dictability observed in standard OT. The outcome in (15), and in particular the
9. We assume that VCD creates a derived long vowel with a single peak (but composed of two
moras), thus violating a *Peak constraint only once. Along these same lines, we suggest that no
vowel is deleted via VCD, and thus no violation of Max-Vowel is assessed.
difference between the total harmony scores for the predicted vs. attested candi-
dates, becomes more intriguing when comparing it to the winner and runner-up
from (14), as in (16).
(16)
*Pk Max Max Max [*M1& Wd[*M1& *M2/L
/ksr/ [hi] [hi] [+hi] K *M2/L *M2/L H
w=.2
w=4.25 w=2.5 w=2 w=1.5 w=.6 w=.6
a. .s.r 3 1 14.25
b. .sr 2 1 1 1 1 1 13.9
/bilakoro/ to lose, sum must be > *Peak[hi]
c. bla.ko.ro 3 1 1 1 1 16.15
d. bla.kro 2 1 1 2 2 2 15.8
to win, sum must be < *Peak[hi]
The relationship (weight ratio) between CB constraints is such that a harmony

score difference of .35 separates the winner (16d) from the runner up (16c). This
same difference (but importantly in the opposite direction) separates the incor-
rectly predicted winner (16b) from the attested winner (16a). In the former case,
for (16d) to lose, the combined violation score of Max and the syllable margin
markedness constraints violated with it (H = 5.3) would have to sum to less than
the penalty for violating *Peak[hi] another time (H = 4.25). Analogously, for the
doubly reduced candidate (16b) to win, the combined violation score (H = 3.9)
would need to be greater than the penalty for violating *Peak[hi] another time.
Thus, it is impossible (via simple summation of violation scores) to arrive at both
attested outcomes in (16), even though the weight ratios between these constraints
have successfully predicted attested outcomes straightforwardly for other word
types, e.g. (1214). CB, therefore, appears to be non-harmonically complete. For
non-harmonically complete systems, Smolensky (2006) proposes that local con-
junction of select constraints in HG introduces harmonicity into the grammar. We
follow Smolensky in exploring this possibility in CB.
(16cd) demonstrate that violating Max[hi] and Max[+hi] within a word is
unproblematic in CB. Likewise, combinations of Max[+hi] and Max[+hi] within
the word in (13) are permitted. This suggests that CB prefers violating a lower-
weight Max[+hi] constraint to avoid the costly violation of a high-weight
*Peak[+hi] constraint. The tradeoff between violating faithfulness and satisfying
peak markedness is also favorable enough (in the case of Max[hi] and Max [+hi])
to allow a second deletion. Violating both Max[hi] and Max-K within a word
(16b), however, is problematic.
We know that CB prefers (when phonotactically possible) to delete high vow-
els; however it will delete a non-high vowel only when the alternative is not to re-
duce at all. Furthermore, given viable VS and VCD deletion targets in adjacent
domains, reduction via VCD is preferred. Thus, the inability for Max[hi] and
Max-K violations to co-occur is not necessarily unexpected, yet formally, it is un-
predicted. This disharmonic relationship suggests a superadditive constraint rela-
tionship. That is, while violating one or the other of these constraints in a word is
normally unproblematic, violating them both presents a situation where their
combined violation is equal to their sum plus an additional penalty. We argue that
this additional penalty relates to the difference in total harmony in (16ab), H =
.35, that separates the predicted and attested winners.
To address this disharmony, we propose the addition of a superadditively con-
joined constraint Max[hi]&Max-K with a local domain of the word, as in (17).10
For analytical purposes below, we assign the conjoined constraint a weight of .4,
although importantly this could be any amount greater than .35; recall that in HG,
it is the ratio between constraints that matters. The addition of this constraint in-
troduces harmonicity into the grammar, such that the grammar now accurately
predicts certain CB outcomes wherein some doubly unfaithful forms are permit-
ted (17d), while others are disallowed (17a).
(17)
Wd[Max *Pk Max Max Max [*M1 Wd[*M1

[hi] & *M2/L
/ksr/ [hi] [hi] [+hi] K & *M2/L & *M2/L H
Max-K w=.2
w=4.25 w=2.5 w=2 w=1.5 w=.6 w=.6
w=.4
a. .s.r 3 1 14.25
b. .sr 1 2 1 1 1 1 1 14.3
/bilakoro/
c. bla.ko.ro 3 1 1 1 1 16.15
d. bla.kro 2 1 1 2 2 2 15.8
10. Although we have no words in our corpus illustrating the interaction of Max[+hi]&Max-
K (where Max[+hi] would act on a high vowel in the second of two disyllabic feet and Max-K
on a velar consonant in the first foot) our analysis predicts that a CVV.CCV word formed by
these constraints may be possible. We expect the reverse to be precluded for reasons of illicit
iambic structure, i.e. *CCV.CVV.
LSS (2006) define this type of superadditive interaction as superlinear conjunction,

which they predict may be more common for combinations of markedness con-
straints. The premise is that if one markedness violation yields a marginal struc-
ture, then multiple markedness violations become even worse. The authors suggest
that combinations of faithfulness constraints, on the other hand, could be ordered
in a sublinear relationship such that faithfulness violations yield dissimilarity, and
thus the penalty for being unfaithful to an already dissimilar structure might be
less costly. They point out, however, that this proposition may be problematic.
Indeed, the instances discussed by LSS concern featural rather than structural
faithfulness. For our purposes in CB, proposing a sublinear relationship between
faithfulness constraints fails to capture the data. In (17) and below, a superlinear
relationship between relevant faithfulness constraints better captures the interac-
tion between illicit constraint combinations. Via *Peak markedness constraints
that have higher weight than opposing Max constraints, CB actively removes in-
dependent structures (vowels). When Max is violated to remove a peak, the other
(remaining) structures that Max still protects against removing are no less repre-
sentative of faithful structure than they were previously. We propose, therefore,
that faithfulness constraints, too, can be superlinearly conjoined.
Recall from (12) that outcomes reduced by two non-high vowel deletions are
also unacceptable in CB. For expository purposes, we first accounted for this for-
mally via an undominated *CCC constraint militating against triconsonantal se-
quences. That this constraint is, in fact, too powerful becomes clear from a very
limited number of words that accommodate multiple deletions to create CVC.
CCV words. Importantly, as in other words, this involves the deletion of two high
vowels and non-identical M2 consonants. For example, [ji.ri.bu.lu] leaf in SB re-
duces to [jir.blu] in CB. An undominated *CCC constraint would incorrectly pre-
clude such words. By lowering the weight of *CCC considerably, an interesting
result emerges. (18) illustrates that an unattested winner is now predicted. The
weight ratios between constraints again yield a harmony score difference of .35
between this candidate and the unpredicted yet attested winner.
(18)
*Pk[hi] Max[hi] [*M1& Wd[*M1 *M2/L

/nrkolo/ *M2/L & *M2/L H
w=4.25 w=2.5 w=.2
w=.6 w=.6
a. nr.ko.lo 3 1 1 1 16.05
b. n.r.ko.lo 4 17
c. nr.klo 2 2 1 2 2 15.7
d. nr.klo 2 2 2 2 2 16.3
As in (17), at issue in preventing the double reduction is CBs unwillingness to vio-

late Max[hi], except as a last resort. Reduction by non-high vowel loss is the least
preferred means of minimization in CB, and Max[hi] is the highest weight faith-
fulness constraint (of those considered) in the hierarchy. CB disallows multiple
deletions of non-high vowels within a word, suggesting another case of super-
additivity. That is, while removing one non-high vowel is unfavorable, removing
more than one is far worse; so much so that it is prevented. Admitting a self-
conjoined constraint Wd[Max[hi]2 (shorthand for Wd[Max[hi]&Max[hi]) into
the hierarchy adds harmonicity into the system, and the attested outcome is once
again accurately accounted for. This is shown in (19).
(19)
Wd[Max *Pk Max [*M1& Wd[*M1& *M2/L

/nrkolo/ [hi]2 [hi] [hi] *M2/L *M2/L H
w=.2
w=.4 w=4.25 w=2.5 w=.6 w=.6
a. nr.ko.lo 3 1 1 1 16.05
b. n.r.ko.lo 4 17
c. nr.klo 1 2 2 1 2 2 16.1
d. nr.klo 1 2 2 2 2 2 16.7
A final phenomenon arguably related to the instances of disharmony above and

the avoidance of removing non-high vowels centers upon certain words that con-
tain high vowels that are ineligible for deletion, whether due to phonotactics or
issues of metrical structure. What makes these words unusual is that they also
contain a non-high vowel whose removal would appear to both satisfy the drive
towards minimization in CB and would not compromise syllable margin phonot-
actics. This non-high vowel, however, is never removed by VS. The result is CB
words that are fully faithful to their SB input. Representative examples include: SB
[ki.ba.ru] CB [ki.ba.ru] book; SB [fu.a.ri] CB [fu.a.ri] worthless person;
SB [du.k.n] CB [du.k.n] courtyard. In these words, removing the first syl-
lable high vowel would create impermissible obstruent-obstruent complex onsets,
while removing the word-final vowel would create a disallowed coda. Creating a
word-internal obstruent-sonorant complex onset is acceptable in other instances,
but here, the option is not possible. Tableau (20) shows that the grammar overpre-
dicts non-high vowel deletion.
(20)
*Pk *Pk Max Max [*M1& Wd[*M1&
*M2/ Obs *M2/N
/dukn/ [+hi] [hi] [hi] [+hi] *M2/N *M2/N H
w=5.5 w=.2
w=5 w=4.25 w=2.5 w=2 w=.6 w=.6
a. du.k.n 1 2 13.5
b. dk.n 1 2 1 16
c. du.kn 1 1 1 1 1 1 13.15
The constraint ratios again yield a difference in harmony scores of .35 between the
predicted and attested winners. The recurrence of this value and the ratio of con-
straint violations that it represents suggests that here, too, disharmony is at play.
The disharmony once again appears related to the dispreference in CB to violate
Max[hi] except as a last resort. In (20), the incorrectly predicted candidate avoids
creating an impermissible complex onset by instead removing a non-high vowel.
The combination of constraints violated to do so is Max[hi] and *Peak[+hi]. The
attested winner, however, avoids a Max[hi] violation, instead allowing a second
*Peak[hi] violation. Resisting the drive toward minimization by total preserva-
tion of peak faithfulness is preferred in this situation. Although this seems prob-
lematic formally, it is intuitive in terms of cross-linguistic generalizations on
markedness.
In reference to harmonic completeness, Smolensky (2006) discusses that the
presence of a marked structure in an inventory entails the presence of an un-
marked structure, relatively speaking. The outcome in (20) supports this assertion,
i.e. one would not predict CB to maintain a marked high vowel at the expense of
removing an unmarked non-high vowel instead. CB behaves as predicted by bar-
ring against this marked state of affairs. This again suggests a superadditive
relationship between Max[hi] and *Peak[+hi]. An important detail, however,
concerns the domain of conjunction for these particular combined constraints.
While the constraints in (17) and (19) had a local domain of the word,
Max[hi]&*Peak[+hi] have a more restricted domain, i.e. a foot. This is clear
when comparing outcomes like (20) (SB [(du.k) (n)] courtyard CB [(du.k)
(n)]) to words like SB [(s.b) (l)] patience which reduces to [(s.bl)] in
CB. The latter shows that violations of Max[hi] and *Peak[+hi] can co-occur
within a word, while the former demonstrates that such violations cannot co-occur
within the more restricted foot domain. Admitting a superlinearly conjoined
Ft[Max[hi]&*Peak[+hi] constraint with a local domain of the foot introduces
harmonicity to the grammar, as in (21).
(21)
Ft[Max
*M2/ *Pk *Pk Max Max [*M1& Wd[*M1 *M2/
[hi]&*Pk
/dukn/ Obs [+hi] [hi] [hi] [+hi] *M2/N & *M2/N N H
[+hi]
w=5.5 w=5 w=4.25 w=2.5 w=2 w=.6 w=.6 w=.2
w=.4
a. du.k.n 1 2 13.5
b. dk.n 1 2 1 16
c. du.kn 1 1 1 1 1 1 1 13.55
In this section, we presented three instances in CB where superadditive (in this

case, superlinear) conjunction of constraints introduces harmonicity into the
grammar. We suggest that these constraints, Wd[Max[hi]&Max-K, Wd[Max[
hi]&Max[hi], and Ft[Max[hi]&*Peak[+hi], arise from a restriction in CB pro-
tecting against removing non-high vowels, except in the most extreme cases.
While our intent is to explore apparent instances of superadditivity in CB, rather
than to argue the merits and shortcomings of constraint conjunction, we offer
some brief comments on the latter here. First, regarding motivation for constraint
conjunction, we provide motivation in Section 5 that the proposed conjunctions
are arguably intuitive within the scope of the CB grammar; all three relate to
subtle, yet related, phonological effects in the languages phonology. Next, unlike
constraints in OT LCC, the power of the proposed superadditively conjoined
constraints is comparatively minimal; their weight is comparable to syllable mar-
gin constraints. Relating to concerns in Pater, Bhatt and Potts (2007) regarding
locality of conjunction, we have described the effects of Wd[Max[hi]&Max-K
and Wd[Max[hi]&Max[hi] within the broader definition of the grammatical
word thus far; however if future research (perhaps on compounds and more com-
plex derivatives) uncovers that these effects are more restricted to the prosodic
word domain, it may be necessary to redefine the local domain of these con-
straints. A more restricted domain is already clear in the case of Ft[Max[
hi]&*Peak[+hi]. Finally, the larger question as to the types of constraints that
should be permitted to conjoin is still an open area of debate (e.g Ito & Mester
1998; ubowicz 2005), and therefore we set this point aside for the present time.
The next section discusses some further theoretical and analytical concerns re-
lated to our findings.
5. Summary and conclusion
CB exhibits a drive towards minimization by removing segments, and therefore

syllables, from its SB input. Generally, this involves violating faithfulness con-
straints (i.e. Max) to avoid violating more costly markedness constraints (i.e.
*Peak). CB data show that high vowels are marked and easily removed, and while
non-high vowels are not exempt from removal, this is a less preferred option. Con-
straints on singleton and conjoined syllable margins also determine the reductions
that are phonotactically accommodated by the language and even preclude mini-
mization in notable instances. Overall, minimization transforms SBs maximal CV
syllable template to one in CB that permits complex syllables (e.g. CCV, CVC,
CVV). Complex syllables are permitted in CB within the bounds of phonotactics
and metrical structure, yet the number and co-occurrence of these syllables in a
given word is restricted. We have shown that some words accommodate more
than one complex syllable; however their co-occurrence is lexically gradient. That
is, words in which two high vowels (or alternatively one high vowel and one non-
high vowel) are removed allow two complex syllables in some combinations. Other
combinations (i.e. where two non-high vowels would be removed or where certain
combination of VS and VCD would act simultaneously to remove segments) are
disallowed. Similarly, in a foot domain, removing a non-high vowel while leaving
behind a high vowel is not permitted. We next consider further motivation for
these bounds and the state of affairs that they militate against.
The principle of harmonic ascent asserts that languages violate Faith only to
reduce markedness (e.g. McCarthy 2000; Moreton 2004). Thus, an output unfaith-
ful to its underlying form must emerge less marked. Because *Peak constraints are
critically ranked above Max in CB, removing a vowel decreases Peak markedness;
however by creating a complex syllable, this subsequently increases syllable mar-
gin markedness, albeit to a smaller, non-equivalent degree. Adding to this is the
fact that the penalties for violating constraints on different types of marked syllable
structures are themselves nonequivalent.
Removing two marked high vowels, while doubly unfaithful, increases the
harmony of the system to a degree that overcomes the opposing effect of introduc-
ing multiple low level violations of margin markedness constraints. We can say the
same for the removal of one high and one non-high vowel, although the positive
effect is decreased, and the possible outcomes are more restricted. In CB, remov-
ing a non-high vowel is a last ditch effort to minimize when no other means is
possible. This straightforwardly explains the restriction related to Wd[Max[hi]2.
If removing one non-high vowel is generally unfavorable, removing a second does
not contribute to harmonic ascent. For Ft[Max[hi]&*Peak[+hi], the principle of
harmonic ascent precludes the language being unfaithful to unmarked structure
while allowing a marked structure to remain intact. Lastly, in reference to

Wd[Max[hi]& Max-K, because VCD is a preferred means of minimization, there
is little motivation to remove an additional unmarked vowel.
Superadditive effects are often found in instances where constraints represent
relative dispreference for certain structures (i.e. structures that are not banned
outright), independent of their relationship to other constraints (Albright 2009).
While Albright refers specifically to superadditive combinations of markedness
constraints, we propose that these principles may be applicable to other constraint
combinations, as shown above in CB. Indeed, Albright (2008), arguably in support
of our CB observations, suggests that additive interactions are confined to par-
ticular subsets of constraints, or violations within a restricted domain. This is cer-
tainly true of CB, where superadditive effects are limited to constraint violations in
conjunction with Max[hi] and are restricted either to the domain of the word
(for Max[hi] with another faithfulness constraint) or the foot (for Max[hi]
with *Peak[+hi]).11 Furthermore, the restrictions and superadditive effects ob-
served alongside the gradient permissibility of certain combinations of complex
syllables within a word can be attributed to a critical ratio or well-formedness
threshold (Albright 2009).
The reoccurring value, H=.35, between predicted and attested outcomes dis-
cussed above represents a critical ratio or harmony threshold between constraints
on markedness vs. faithfulness. This value corresponds to the difference between
retaining a non-high vowel vs. removing another segment and subsequently creat-
ing a complex syllable. This tradeoff, and hence this ratio, emerges in each instance
of superadditivity considered above, whereby CB maintains strict checks and bal-
ances on the types of markedness it will retain or permit, while still facilitating a
drive towards minimization. CB permits a certain amount of markedness via the
low (cumulative) weight of margin constraints, thereby facilitating reduction and
the introduction of complex syllables; however it places limits on the types of re-
ductions that can occur to satisfy this drive. The role of superlinear conjunction in
11. We thank Amalia Gnanadesikan (personal communication) for pointing out that the way
in which CB bars against dispreferred structures and even the recurrence of Max[hi] in super-
additive conjunctions is reminiscent of a phonological conspiracy (e.g. Kisseberth 1970), al-
though not in all respects. In typical conspiracies, rules act together yet in different ways to
achieve the same outcome. In CB, however, while the superadditively conjoined constraints
function in similar ways to protect against removing non-high vowels, there is no constraint or
combination of constraints that bans the removal of these vowels outright. Moreover, while
conspiracies typically yield transparent, surface-true outcomes, the outcomes in CB are arguably
opaque. A deeper comparison between the effects of superadditivity and conspiracies must
await further research.
CB, in effect, is to place an upper limit on reductions and thereby to ensure har-
monic ascent.
Finally, we consider how these complex, additive relationships might arise in
CB and how or why they might eventually be resolved. Albright, Magri & Michaels
(AMM: 2008), drawing upon data from Dutch acquisition (Levelt, Schiller & Levelt
2000), offer a point of departure in their study modeling doubly-marked lags.
They offer that at certain points in language acquisition, there is a noticeable gap in
structures (e.g. syllables) containing two marked characteristics (e.g. complex on-
sets and codas), even though structures containing one marked structure or the
other are readily attested (i.e. syllables with a complex onset or syllables with a
coda). Although such bans on doubly marked structures are not uncommon in
early stages of language acquisition, they fail to survive in stable, end-state adult
language. In their error-driven Split Additive Model, AMM demonstrate, that lan-
guages predictably pass through stages banning doubly-marked structures, but that
such restrictions are quickly eliminated before the language reaches its end state.
The CB data presented in this paper may offer some support to this proposi-
tion, although the details are quite different. Consider first that the ranking of
constraints in CB generally compels minimization via reduction through the high-
er weight of Markedness constraints (*Peak). This drive is kept closely in check by
Faithfulness (i.e. Max). The additive interactions at play in CB, however, are dif-
ferent from those discussed in AMM (2008). While AMM are concerned with
combinations of Markedness constraints (*Coda and *#CC), which act together
to overcome Max, the additive interactions of note in CB are either between Faith-
fulness constraints (i.e. Wd[Max[hi]& Max-K and Wd[Max[hi]2) in the word
domain or between Faithfulness and Markedness constraints (i.e. Ft[Max[
hi]&*Peak[+hi]) in the foot domain. In CB, these additive interactions generally
act to preserve Faithfulness to a less marked vowel while subsequently permitting
the retention of overall Markedness by retaining an extra syllable peak. What is
interesting here and more in line with Dutch doubly-marked lags is the relation-
ship between additive interactions in the CB word domain and the introduction of
complex syllables into the language. While we have seen that CB readily permits
complex syllables in most instances, our data illustrate that the language permits
multiple complex syllables within a single word far less often.12 When multiple
complex syllables are permitted, certain co-occurrence restrictions come to light,
and moreover, as a result of these additive interactions, only certain combinations
of segments can be removed to create them. Via these interactions, CB effectively
bans the introduction of some, but not all, doubly marked structures (i.e. multiple
12. The details of Ft[Max[hi]&*Peak[+hi] are somewhat different but follow from the same
generalization.
complex syllables) within a word. Recall that this state of affairs, while grammati-
cal, is not thought to be characteristic of an end-state grammar.
Green (2010) suggests that CB is an emergent language variety stemming from
a more phonologically-conservative, normative variety of Bambara with stricter
syllable phonotactics. The data in this paper, viewed in light of AMM (2008), ap-
pear to support the proposition that CB is in a state of flux, having phonological
characteristics akin to that of a developing language. It is clear that CB has relaxed
its phonotactics in comparison to Standard Bambara, yet it restricts the minimi-
zation processes involved in this relaxation. We expect, therefore, that CB phonol-
ogy may ultimately (and perhaps quickly) remove its remaining restrictions on
multiple syllable complexity and come to more closely approximate the more le-
nient syllable phonotactics of what has yet been anecdotally described for some
varieties of Maninka (B. Diakite, personal communication), a close cousin of
Bambara, via the removal of the superadditive relationships outlined above. The
outcome must clearly await future research.
What we have done is to provide insight into the types of superadditive rela-
tionships between constraints that may be encountered in natural language and to
provide some evidence for the role that such interactions play in the emergent
phonology of CB. The data presented suggest that HG with the addition of super-
linear, locally-conjoined constraints is necessary to predict several attested gram-
matical outcomes in the language that are otherwise formally unpredicted. As a
complement to earlier work on superlinearly conjoined markedness constraints,
the analysis in this paper demonstrates two instances in which faithfulness con-
straints combine in a superlinear relationship, as well as an instance in which a
faithfulness constraint and a markedness constraint combine in such a relation-
ship. Important to the discussion in this paper has been the identification of a
consistent manifestation of disharmony in CB, represented by a critical ratio be-
tween attested vs. formally predicted outputs. We illustrated that each instance of
disharmony is linked to violations of Max[hi] alongside some other constraint,
which effectively act to limit the drive towards minimization in the language. Fi-
nally, we considered how the synchronic ban on some instances of multiple com-
plex syllables within a single word relates to propositions about the acquisition
and resolution of doubly-marked lags discussed elsewhere in the literature.
References
Albright, A. 2008. A universally gradient co-occurrence restriction? Paper presented at the 16th
Manchester Phonology Meeting.
Albright, A. 2009. Cumulative violations and complexity threshold: Evidence from Lakhota.
Paper presented at the 17th Manchester Phonology Meeting.
Albright, A., Magri, G. & Michaels J. 2008. Modeling doubly marked lags with a Split Additive
Model. In BUCLD 32 Proceedings, Vol. 1., H. Chan, H. Jacob & Enkeleida Kapia, 3647.
Somerville MA: Cascadilla.
Baertsch, K. 2002. An Optimality Theoretic Approach to Syllable Structure: The Split-margin
Hierarchy. PhD dissertation, Indiana University.
Baertsch, K. & Davis, S. 2009. Strength relations between consonants: A syllable-based OT ap-
proach. In Strength Relations in Phonology, K. Nasukawa & P. Backley (eds), 293324. New
York NY: Mouton de Gruyter.
Becker, M., Pater, J. & Potts, C. 2007. OT-Help 1.2 software package, University of Massachusetts,
Amherst. <http://web.linguist.umass.edu/~OTHelp/>
Coetzee, A. & Pater. J. 2008. Weighted constraints and gradient restrictions on place co-occur-
rence in Muna and Arabic. Natural Language & Linguistic Theory 26: 289337.
Davis, S. & Baerstch, K. 2011. On the relationship between codas and onset clusters. In Hand-
book of the Syllable, C. Cairns & E. Raimy (eds), 7198. Leiden: Brill.
Downing, L.J. 1998. On the prosodic misalignment of onsetless syllables. Natural Language &
Linguistic Theory 16: 152.
Dumestre, G. 2003. Grammaire fondamentale du bambara. Paris: Karthala.
Farris-Trimble, A. 2008. Cumulative Faithfulness Effects in Phonology. PhD dissertation, Indi-
ana University.
Green, C. 2010. Prosodic Phonology in Bamana (Bambara): Syllable Complexity, Metrical
Structure, and Tone. PhD dissertation, Indiana University.
Green, C. & Diakite, B. 2008. Emergent syllable complexity in Colloquial Bamana. Journal of
West African Languages 35(1): 4556.
Green, C. & Farris-Trimble, A. 2010. Voice contrast and cumulative faithfulness in Luwanga
nouns. Studies in African Linguistics 39(2): 183233.
Green, C., Davis, S., Diakite, B. & Baertsch, K. 2012. Domain-restricted reduction: A proposal
for segmental feet in Bamana. In Selected Proceedings of the 41st Annual Conference on Af-
rican Linguistics, B. Connell & N. Rolle (eds), 19. Somerville MA: Cascadilla Proceedings
Project.
Green, C., Davis, S., Diakite, B. & Baertsch, K. To appear. On the role of margin phonotactics in
Colloquial Bamana complex syllables. Natural Language and Linguistic Theory.
Ito, J. & Mester, A. 1998. Markedness and word structure: OCP effects in Japanese. ROA-255.
Khanjian, H., Sudo, Y. & Thomas, G. 2010. Modeling doubly marked lags with exacerbation.
Paper presented at the 18th Manchester Phonology Meeting.
Kisseberth, C. 1970. On the functional unity of phonological rules. Linguistic Inquiry 1(3): 291
306.
Leben, W. 2002. Tonal feet. In Proceedings, typology of African prosodic systems, U. Gut & D. Gib-
bon (eds), 2740. Bielefeld: Bielefeld Occasional Papers on Typology.
Leben, W. 2003. Tonal feet as tonal domains. In Trends in African Linguistics 5: Linguistic Typol-
ogy and Representation of African Languages, J. Mugane (ed), 129138. Trenton NJ: Africa
World Press.
Legendre, G., Sorace, A. & Smolensky, P. 2006. The Optimality Theory-Harmonic Grammar
connection. In The Harmonic Mind: From Neural Computation to Optimality-theoretic
Grammar, P. Smolensky & G. Legendre (eds), 339402. Cambridge MA: The MIT Press.
Levelt, C., Schiller, N.O. & Levelt, W.J. 2000. The acquisition of syllable types. Language Acquisi-
tion 8: 237264.
ubowicz, A. 2002. Derived environment effects in Optimality Theory. Lingua 112: 243280.
ubowicz, A. 2003. Local conjunction and comparative markedness. Theoretical Linguistics 29:
101112.
ubowicz, A. 2005. Locality of conjunction. In Proceedings of the 24th West Coast Conference on
Formal Linguistics, J. Alderete, A. Kochetov & C.-h. Han (eds), 254262. Somerville MA:
Cascadilla Proceedings Project.
McCarthy, J.J. 2000. Harmonic serialism and parallelism. In Proceedings of NELS 30, M. Hiro-
tani, A. Coetzee, N. Hall & J.-y. Kim (eds), 501524. Amherst MA: GLSA.
Moreton, E. 2004. Non-computable functions in Optimality Theory. In Optimality Theory in
Phonology: A reader, J. McCarthy (ed.), 141164. Malden MA: Blackwell.
Moreton, E. & Smolensky, P. 2002. Typological consequences of local constraint conjunction. In
Proceedings of the 21st West Coast Conference on Formal Linguistics, L. Mikkelsen & C. Potts
(eds), 306319. Somerville, MA: Cascadilla Proceedings Project.
Pater, J. 2009. Weighted constraints in generative linguistics. Cognitive Science 33(6): 9991035.
Pater, J. To appear. Universal grammar and weighted constraints. In Harmonic Grammar and
Harmonic Serialism, J. McCarthy & J. Pater (eds). London: Equinox.
Pater, J., Bhatt, R. & Potts, C. 2007. Linguistic optimization. ROA-924.
Prince, A.S. & Smolensky, P. 1993/2004. Optimality Theory: Constraint Interaction in Generative
Grammar. Malden MA: Blackwell.
Smolensky, P. 1995. On the structure of the constraint component Con of UG. ROA-86.
Smolensky, P. 2006. Optimality in Phonology II: Harmonic completeness, local constraint con-
junction, and feature domain markedness. In The Harmonic Mind: From Neural Computation
to Optimality-theoretic Grammar, P. Smolensky & G. Legendre (eds), 27160. Cambridge
MA: The MIT Press.
Smolensky, P. & Legendre, G. 2006. The Harmonic Mind: From Neural Computation to Optimal-
ity-theoretic Grammar. Cambridge MA: The MIT Press.
Vydrine, V. 2010. Le pied mtrique dans les langues mand. In Essais de typologie et de linguis-
tique gnrale: Mlanges offerts Denis Creissels, F. Floricic (ed.), 5362. Lyon: ENS
ditions.
Weidman, S. & Rose, S. 2006. A foot-based reanalysis of edge-in tonal phenomena in Bambara.
In Proceedings of the 25th West Coast Conference on Formal Linguistics, D. Baumer, D.
Montero & M. Scanlon (eds), 426434. Somerville MA: Cascadilla Proceedings Project.
Author index
A Boersma, P. 35, 45, 83, 84, 86, 88, Coetzee, A. 224, 246
Abrahamson, A. 179181, 196 219, 221 Cole, J.S. 207, 222
Abrahamson, J. 179181, 196 Bollt, A. 73, 87 Colina, S. 135, 151
Adriaans, F. 84, 86 Booij, G. 182, 196 Colson, B.G. 111
Aitchison, J. 158, 170 Borowsky, T. 135, 137, 148, 151 Compton, A. 73, 86
Albright, A. 223225, 243246 Bradley, T.G. 26, 42, 45 Connelly, M. 19, 22
Altenberg, E. 25, 45 Brenner, C. 99, 111 Conway, C.M. 108, 112
Alter, J. 18, 22, 134, 151 Broselow, E. 25, 45, 115, 117, Cooper, J.A. 77, 89
Anaya, E. 108, 112 118, 128 Core, C. 119, 130
Apoussidou, D. 83, 88 Brown, R. 12, 22, 58, 137, 142, 151 Costello, J. 71, 86
Aslin, R.N. 84, 88 Brown, B.L. 72, 88 Cox, F. 13, 17, 23
Avery, P. 71, 88 Buller, B. 176, 196 Crawford, J.S. 101, 111
Buller, E. 176, 196 Crosbie, S. 11, 22
B Burkholder, R.A. 104, 105, 110 Cruttenden, A. 55, 56, 65
Baddeley, A. 104, 105, 110, 111 Burton, M. 33, 47 Crystal, T.H. 33, 46
Baerg, M. 160, 170 Culbertson, J. 18, 22, 134, 151
Baertsch, K. 226, 231, 232, 246 C Curtis, W.J. 98, 110
Bain, B.A. 74, 86 Calabrese, A. 116, 129
Barkley, R.A. 101, 110 Camarata, S.M. 134, 151 D
Barlow, J.A. 1, 7, 9, 10, 49, 50, Cancino, H. 30, 46 Dalal, R.H. 137, 151
113, 115, 125, 127, 129, 133, 134, Carlisle, R.S. 25, 46 Davidson, L. 99, 111
148150, 173, 174, 192, 196, 200, Carman, J.N. 157159, 171 Davis, R.A.O 100, 111
213, 221, 222 Carter, A.K. 105, 110, 111 Davis, S. 117, 128, 149, 151, 173,
Barton, D. 2628, 47 Casali, R.F. 135, 151 176, 196, 223, 226, 227, 231, 246
Bastiaanse, R. 116, 129 Casserly, E.D. 106, 110 Dayley, J. 185, 186, 196
Becker, M. 233, 246 CDaCI Investigative Team, 112 Dean, E.C. 72, 73, 83, 86
Beckman, M. 122, 127, 137, 150 Charles-Luce, J. 4, 9, 10, 113, 114 De Jong, D. 135, 151
Beckman, J. 175, 196, 215, 216, Chen, S. 25, 45, 115, 128 de Lacy, P. 115, 128, 175, 190, 197
221 Chen, M. 72, 86 Delforge, A.M. 26, 42, 45
Bedore, L.M. 73, 86 CHILDES database, 18, 19, 22, Demuth, K. 920, 2224, 134,
Ben-David, A. 119, 120, 123, 54, 55 136138, 143, 144, 147149,
128, 130 Chin, S.B. 1, 6, 7, 49, 50, 54, 65, 151153
Bender, J. 159, 170 71, 86, 113, 114, 173, 174, 192, 196 de Villiers, J. 137, 142, 151
Berend, N. 162, 168, 170, 171 Cho, T. 29, 46 de Villiers, P. 137, 142, 151
Berent, I. 115, 128 Cho, M.-H. 173, 174 Diakite, B. 225227, 245, 246
Bergeson, T. 96, 110 Chomsky, N. 2, 7 Dillon, C.M. 105, 110, 111
Bernhardt, B.H. 54, 65, 84, 86, Cicchetti, D. 98, 110 Dinnsen, D.A. vii, 110, 21, 26,
149, 153 Clark, A. 93, 110 28, 46, 4951, 54, 65, 66, 71, 72,
Bhatt, R. 73, 88, 224, 241, 247 Clark, G.M. 98, 110 74, 75, 77, 84, 86, 87, 92, 113,
Birdsong, D. 111, 115, 129 Cleary, M. 97, 102, 103, 105, 107, 114, 134, 150, 154, 163, 170, 171,
Blair, C. 100, 110 110112 173, 174, 192, 196, 200, 201, 213,
Blevins, J. 134, 148150, 177, 196 Clements, A. 54, 65 220, 221
Bloomfield, L. 155, 169, 171 Clements, G.N. 71, 86, 124, 128, Dodd, B. 11, 22, 54, 65, 66
Blumstein, S.E. 33, 47 135, 137, 151, 210, 221 Dollaghan, C.A. 74, 86
Dorman, M.F. 98, 112 Gennari, S. 19, 22 Hsieh, L.I. 16, 20, 22, 137, 152
Downing, L.J. 224, 246 Gerken, L. 12, 13, 18, 22, 24, 73, Hua, Z. 11, 22, 54, 65
Dubasik, V.L. 49, 53, 54, 56, 84, 85, 87, 122, 130, 148, 151 Hualde, J.I. 26, 42, 46
6466 Gerrits, E. 119, 121, 128, 130 Huber, W. 116, 129
Ducker, G. 73, 88 Gibbon, F. 26, 47 Hulme, C. 104, 111
Dumestre, G. 225, 226, 246 Giegerich, H.J. 117, 128, 135, 151 Hume, E.V. 71, 86
Gierut, J.A. 3, 710, 26, 28, 46, Hyams, N. 12, 23
E 4951, 71, 72, 7477, 84, 86, 87, Hyman, L. 176, 186, 197
Echols, C.H. 194, 196 92, 113115, 117, 128, 148151,
Eckman, F.R. 9, 10, 25, 27, 29, 42, 173, 174, 201, 213, 221 I
43, 46, 113, 114 Gioia, G.A. 101, 108, 111 Iglesias, A. 139, 152
Edwards, M.L. 54, 65, 72, 89, Glicksman, L. 33, 47 Ingram, D. 11, 23, 49, 5357,
122, 127 Gnanadesikan, A. 192, 197, 200, 6466, 72, 88, 134, 152
Edwards, J. 137, 150 221, 243 Ingram, K. 55, 56, 64
Eisenberg, L.S. 112 Goad, H. 15, 22, 55, 65, 125, 128 Isquith, P.K. 101, 108, 111
Elbert, M. 310, 26, 46, 4951, Goedemans, R. 176, 197 It, J. 149, 152, 224, 241, 246
54, 65, 66, 7175, 77, 83, 86, 87, Goehl, H. 27, 46 Iverson, G. 25, 27, 29, 42, 43, 46
92, 113, 114, 134, 154, 174 Goldsmith, J.A. 3, 8, 135, 151
Ellis, D. 19, 22 Gordon, M. 176, 178, 179, J
Espy, K.A. 101, 111 186188, 191, 192, 197 Jacewicz, E. 27, 46
Ettlinger, M. 137, 146, 149, 151 Gouskova, M. 115, 128 Jackendoff, R. 133, 152
Evans, K. 20, 24 Green, A.D. 149, 151 Jacobs, A. 116, 129
Everett, D.L. 176, 188, 196 Green, C. 113, 114, 173, 174, Jakobson, R. 135, 152, 163
Everett, K. 188, 196 223228, 231, 245, 246 Jarosz, G. 84, 88
Ezrati, R. 120, 123, 128 Green, L.J. 135, 136, 138, 139, 151 Jeare, A.R.
Guijarro-Fuentes, P. 133, 153 Jesney, K. 207, 218221
F Guy, S.C. 108, 111 Johnson, B.W. 136, 152
Farris-Trimble, A.W. 1, 7, 8, 49, Guy, G. 136, 151 Johnson, D.C. 160162, 171
50, 84, 87, 113, 114, 173, 174, 199, Johnson, V.E. 148, 152
200, 202, 203, 205, 207, 219, H Jongman, A. 164, 207, 222
221, 224, 233, 246 Hall, T.A. 117, 128 Jusczyk, P.W. 83, 88
Farwell, C.B. 77, 87 Halle, M. 2, 7, 129, 135, 151
Fee, E.J. 54, 65 Hamann, S. 83, 88 K
Ferguson, C.A. 77, 87 Hammerly, H. 25, 46 Kager, R. 84, 86
Fernndez Fuertes, R. 5355, 66 Hammond, M. 13, 24, 122, 130 Karimi, S. 118, 128
Fikkert, P. 117, 128, 193, 194, 197 Hardcastle, W.J. 26, 47 Karpicke, J. 112
Finer, D. 115, 128 Harris, J. 26, 42, 46, 116, 128 Kaull, P. 162, 169171
Fink, N.E. 112 Hayes, B. 84, 86, 176, 186, 197, Keare, A.R. 134, 139, 150, 152
Flege, J.E. 25, 46 219, 221 Kearns, K.P. 75, 88
Fletcher, P. 26, 47 Hegarty, M. 136, 153 Keel, W. 114, 155, 160, 162, 165,
Fong, C. 137, 153 Henning, S.C. 108, 111, 112 166, 171
Forrest, K. 9, 10, 26, 46, 49, 51, Hewlett, N. 28, 46 Kehoe, M. 13, 23, 194, 195, 197
71, 87 Hickey, R. 163, 171 Kenstowicz, M. 117, 128
Fox, R.A. 27, 46 Hitch, G.J. 105, 111 Kenworthy, L. 108, 111
Fudge, E.C. 135, 151 Hodge, M. 9, 10, 26, 46 Keren-Portnoy, T. 79, 89
Fujimura, O. 4, 135, 151 Hogg, R. 116, 128 Keyser, S.J. 135, 151
Holm, A. 11, 22 Khanjian, H. 224, 246
G Hooper, J. 115, 128 Khramova, M. 160, 162164, 171
Gabas Jr., N. 178, 197 Hoover, J.R. 149, 151 Kiparsky, P. 29, 46, 135, 152
Gandour, J. 73, 86, 134, 151 Horn, D.L. 96, 100, 111, 112 Kirchner, R. 205, 221
Garcia, A. 26, 45, 46 Horohov, J.E. 136, 137, 146, 152 Kirk, C. 13, 14, 23, 24, 84, 89, 134,
Gathercole, S.E. 105, 111 House, A.S. 33, 46 138, 149, 152
Geers, A.E. 9799, 102, 104, Householder, F.W. 3, 8 Kirk, K.I. 96, 97, 107, 110, 111
111, 112 Howell, J. 72, 86
Author index
Kisseberth, C. 149, 152, 199, 200, Marler, P. 97, 111 Newhoff, M. 54, 66
215, 216, 221, 243, 246 Marshall, C.R. 14, 23, 136, 138, 152 Newport, E.L. 84, 88, 194, 196
Klingberg, T. 101, 111 Martin, A.J. 105, 111 Nicholas, J. 111
Konishi, M. 97, 111 Martnez-Gil, F. 26, 47 Niparko, J.K. 96, 112
Kornfeld, J.R. 27, 46 Masapollo, M. 19, 22
Kristoffersen, K. 119, 121, 129, 130 Mattheier, K. 156, 171 O
Kronenberger, W.G. 101, 106, Maxwell, E.M. 4, 8, 26, 47 Oberauer, K. 100, 112
108, 111, 112 Maye, J. 84, 88 OConnor, K.M. 49, 51, 113, 114,
Kuhl, P.K. 16, 24, 137, 153 McCarthy, J.J. 148, 149, 152, 204, 173, 174, 201, 213, 221
Kurowski, K. 33, 47 216, 222, 242, 247 Oetting, J.B. 136139, 146, 147,
Kwiatkowski, J. 72, 88 McCullough, E. 12, 18, 22 152, 153
McCully, C.B. 116, 128 Oh, M. 41, 47
L McDonald, J.L. 139, 152 Ohala, J.J. 27, 47
Labov, W. 72, 88, 136, 152 McEvoy, S. 54, 65, 66 Ohala, D.K. 115, 129, 192, 197
Ladefoged, P. 163, 171 McGarrity, L.W. 7, 8, 49, 51, Onstine, J. 71, 86
Lado, R. 44, 46 173176, 197
Lambert, M.B. 169, 171 McGregor, K.K. 148, 151 P
Lawrence, A. 104, 111 McIntosh, B. 12, 18, 22 Parker, S. 115, 129
Leben, W. 228, 246 McLeod, S. 119, 121, 130 Pater, J. 18, 23, 84, 88, 115, 125,
Lee, S. 27, 46 McMurray, B. 207, 222 129, 200, 207, 222, 224, 233,
Legendre, G. 200, 206, 221, 222, McNab, F. 101, 111 241, 246, 247
224, 225, 233, 246, 247 McReynolds, L.V. 71, 72, 75, Patrolia, M. 19, 22
Lehiste, I. 14, 23, 34, 47 87, 88 Payne, D. 183, 197
Lennert, T. 128 Mealings, K.T. 15, 17, 20, 23 Peperkamp, S. 84, 89
Leonard, L.B. 16, 22, 54, 66, 72, Meindl, J. 157, 171 Peters, A.M. 85, 88
73, 86, 88, 137, 152 Mnard, L. 24 Peters, S. 97, 111
Levelt, C. 20, 23, 84, 86, 244, 247 Mesalam, L. 54, 66 Phillips, B.S. 72, 88
Levelt, W.J. 20, 23, 244, 247 Mester, A. 224, 241, 246 Pierpont, E.I. 98, 112
Lewis, B. 54, 66 Metsala, J.L. 75, 89 Pinker, S. 115, 129
Lewis, W. 84, 87 Michaels, J. 224, 225, 244, 246 Pisoni, D.B. 50, 91, 93, 96, 97, 99,
Liceras, J. 5355, 66 Mildner, V. 119, 120, 123, 129 100, 102108, 110112
Lightfoot, D. 53, 66 Miles, K. 13, 23 Polite, E.J. 137, 138, 146, 147,
Lim, J.-A. 41, 47 Miller, J. 139, 152 149, 152
Lin, S. 13, 23 Miyamoto, R.T. 96, 100, 111 Post, R. 161, 171
Lle, C. 14, 18, 23 Miyata, Y. 200, 206, 221, 222 Potts, C. 224, 233, 241, 246
Locke, J. 73, 88 Moloi, F. 19, 22 Powell, T. 4951, 54, 65, 71, 73,
Lockhead, G. 73, 88 Moog, J.S. 98, 111 77, 86, 87, 113, 114
Loeb, D.F. 137, 151 Moran, M.J. 135, 152 Price, P.J. 137, 153
Logan, J.S. 71, 88 Morelli, F. 115, 129 Prince, A.S. 13, 23, 115, 129, 148,
Lovins, J. 117, 118, 129, 135, 151 Moreton, E. 205, 222, 224, 242, 149, 152, 153, 200, 216, 222,
ubowicz, A. 200, 204, 222, 224, 247 224, 247
241, 247 Morgan, J.L. 84, 89 Pruitt, S.L. 113, 133, 136, 138, 139,
Lunte, G. 160, 171 Morris, S.R. 136, 152 147, 153
Luria, A.R. 100, 111 Morrisette, M.L. 4951, 71, 72, Putnam, M. 161, 171
75, 77, 87, 88, 113, 114, 173, 174
M Muir, C. 104, 111 Q
Machobane, M. 19, 22 Munson, J. 134, 152 Quinn, P.C. 73, 88
Macken, M.A. 2628, 47 Munson, C. 207, 222 Quittner, L. 112
MacWhinney, B. 54, 55, 66,
143, 152 N R
Magri, G. 224, 225, 244, 246 Nauta, W.J.H 100, 112 Radford, A. 12, 14, 23
Major, R. 25, 47 Needham, A. 73, 88 Razza-Peters, R. 100, 110
Marantz, A. 215, 222 Nespor, M. 12, 23 Redmond, S.M. 136, 153
Marecka, M. 119, 120, 123, 130 Neurburger, H. 102, 112 Reed, C. 156, 171
Reichmann, F.F.L. 168, 171 Sorace, A. 206, 222, 224, 225, V

Reid, J. 72, 86 246 Vago, R. 25, 45
Rice, K. 71, 88 Spahr, A.J. 98, 112 Valian, V. 17, 20, 24
Rice, M.L. 149, 151 Sporns, O. 98, 100, 112 van der Lely, H.K.J. 14, 23, 136,
Rich, F. 183, 197 Stampe, D. 3, 5, 8 138, 152
Rickford, J.R. 136, 153 Steele, J. 15, 22 Van Riper, C. 74, 89
Ritchie, W. 25, 47 Stemberger, J.P. 84, 86, 116, 129, Veneziano, E. 20, 24
Roark, B. 20, 23 136, 149, 153 Venneman, T. 115, 129
Roman, A.S. 112 Stenneken, P. 116, 129 Vergnaud, J.-R. 135, 151
Romani, C. 116, 129 Steriade, D. 115, 129 Vihman, M. 79, 89
Rosansky, E. 46 Stevens, K.N. 33, 47 Vogel, I. 12, 23
Rose, S. 228, 247 Stites, J. 13, 24 Vydrine, V. 228, 247
Rose, Y. 125, 128 Stockman, I.J. 135, 136, 153
Rothman, J. 133, 153 Stoel-Gammon, C. 13, 23, 57, 66, W
Rvachew, S. 83, 88 77, 83, 89, 194, 195, 197 Walley, A.C. 75, 89
Storkel, H.L. 149, 151 Wang, C. 25, 45, 115, 128
S Stuhlman, N. 120, 128 Wang, W.S.-Y. 72, 86
Salmons, J. 27, 46 Sub, H.M. 100, 112 Wang, N.Y. 112
Schach, P. 169, 171 Sudo, Y. 224, 246 Waters, D. 72, 86
Schiller, N.O. 20, 23, 244, 247 Sundara, M. 15, 16, 24, 137, 138, Weenink, D. 35, 45
Schirmunski, V. 156, 162, 164, 147, 148, 153 Weidman, S. 228, 247
168, 171 Svirsky, M.A. 102, 112 Weiner, F.F. 72, 73, 89
Schumann, J. 46 Swanson, K.A. 7, 8, 173, 174 Weismer, G. 4, 5, 810, 26, 46,
Scobbie, J.M. 26, 28, 47 Swanson, L.A. 16, 22, 137, 152 47, 54, 66, 92, 113, 114, 134, 154
Sedey, A. 111 Swartzlander, P. 73, 87 Weis, B. 161, 171
Seeger, G.S. 159, 172 Weiss, D.J. 84, 88
Selkirk, E.O. 12, 15, 23, 24, 115, T White, K.S. 84, 89
129, 135, 153 Tamura, S. 177, 197 White, L. 15, 22
Senturia, M.B. 135, 153 Tenenbaum, J.B. 73, 89 Wildfeuer, A. 156, 157, 172
Service, E. 105, 111 Teoh, S.W. 102, 112 Wilhelm, O. 100, 112
Sharma, A. 98, 112 Tesar, B. 84, 89 Williams, A.L. 83, 89
Shattuck-Hufnagel, S. 15, 20, 24, Tessier, A.-M 207, 221 Wilson, R. 84, 87
137, 149, 153 Thal, D.J. 112 Wiltshire, C.R. 148, 149, 154
Shiraishi, H. 177, 197 Theodore, R.M. 15, 17, 24, 137, Wittmann, W.W. 100, 112
Shriberg, L.D. 54, 65, 66, 72, 88 138, 145, 147, 149, 153 Wolfram, W. 136, 154
Shrier, M. 165, 168, 172 Thomas, E.R. 135, 136, 138, 140,
Simonsen, H. 119, 121, 129, 130 149, 154 X
Sinclair, H. 20, 24 Thomas, G. 224, 246 Xu, F. 73, 89
Smit, A.B. 11, 13, 24 Thompson, L. 54, 66 Xu, Z. 115, 128
Smith, B.L. 27, 47 Thomson, N. 104, 110, 111
Smith, C.L. 34, 47 Tobey, E.A. 102, 111, 112 Y
Smith, J. 175, 178180, 183, 184, Tomic, D. 119, 120, 123, 129 Yava, M. 13, 24, 113, 115117, 119,
190, 191, 197 Topintzi, N. 176180, 182184, 120, 123, 125, 126, 129, 130
Smith, N.V. 77, 88, 193, 199201, 186188, 192, 197
207, 213, 222 Toscano, J. 207, 222 Z
Smolensky, P. 13, 23, 84, 89, 115, Treiman, R. 116, 129 Zamuner, T.S. 13, 24, 122, 130
128, 129, 148, 149, 153, 200, Tremblay, A. 19, 20, 22 Zapf, J. 137, 146, 149, 151
204206, 221, 222, 224, 233, Trommelen, M. 117, 129 Zewen, F.-X. N. 186, 197
235, 236, 240, 246, 247 Tsukamoto, A. 190, 197 Ziemer, S. 75, 87
Snow, C. 54, 55, 66. Tyler, A.A. 72, 89 Zoll, C. 180, 197
Snyder, T. 72, 88 Zumach, A. 119, 121, 128
Someillan, M. 117, 130 U
Song, J.Y. 15, 16, 19, 20, 22, 24, 25, Ullman, M.T. 98, 112
137, 138, 143145, 147149, 153 U.S. Census Bureau, 158, 172
Subject index
A consonant inventories 53, 54, medial 17, 40, 138

adjunct see consonant cluster 5661, 6770 obligatory 11, 21
African American English consonant phonological 11, 12, 21, 32,
(AAE) 135, 136, 138143, cluster see consonant cluster 133143, 162
146150 final 14, 133135, 138, 164 preceding 133, 137, 138,
Ainu 177, 179 see also obstruents, sonorant 140142, 145, 147, 149
allophones 26, 2830, 4144 conspiracy/ies 7, 173, 199202, preconsonantal 140
allophonic split(s) 28, 4244 204, 207, 208, 210, 212217, prevocalic 143
Alyawarra 177 219, 220, 243 prosodic 1114, 1619
Amahl 173, 193, 199203, 207220 constraint(s) treated 7174
Arabela 179, 183, 184, 186, 187, Con 205, 220 untreated 71
191, 192, 195 constraint weights 206, 233 covert contrast 9, 10, 2528, 33,
article(s) 1719, 135, 164168 duplication of 220 36, 38, 40, 41, 43, 45
fixed hierarchy of, 208 cumulative faithfulness effect(s)
B Local Constraint Conjunc- (CFE) see faithfulness
Banaw 176, 177, 180 tion (LC) 200, 204, 205,
basic analysis 56, 59, 64, 67, 68 209, 212, 214, 216, 219, 220, D
Bislama 188 224, 235 Dakota 215, 216, 220
markedness 175, 200, 203, deaf children 6, 91, 93, 94,
C 204, 211, 216, 217, 223, 233, 96102, 104109
chain shift(s) 213, 214, 215, 219 236, 238, 242245 deafness 85, 93, 94, 96, 99, 103,
Childes Database 18, 19, 54, 55 ranking paradox 203, 204, 108, 109
Chimalapa Zoque 175 208, 209, 211, 212 derived environment
cluster see consonant cluster strict domination of, 202, effect(s) 2931, 38, 44,
cochlear implants vii, 6, 50, 91, 224, 232, 235 dialect 17, 114, 135, 136, 139, 140,
9395, 108 superadditively con- 143, 144, 148, 149, 159, 160,
cognitive control processes 100, joined 225, 241, 243 162165, 168170
106108 superadditive interaction domain
complexity 14, 17, 56, 116, of, 224, 235, 237, 238, 240, cognitive domains 109
133135, 137, 138, 140, 142, 148, 241, 243, 245 domain-general cognitive
173, 223, 230, 245 see also superlinear(ity) processes 93, 98, 100, 101
consonant cluster 17, 207 weighted 206, 207, 217, 220, domain-specific cognitive
adjunct clusters 117 234, 235 processes 98, 100, 101, 108
adjunct 1, 117, 125, 148 context 1019, 21, 38, 40, 43, domain-specific phonological
coda cluster(s) 13, 14, 15, 7275, 107, 133150, 155, 162, processes 202, 204, 205,
24, 212 163 209, 212, 223, 224, 227230,
cluster reduction 15, 210, 212 basic 38, 40, 44 232, 235, 237, 240244
medial cluster simplifica- derived 29, 38, 44 processing domains 106
tion 214 final 133, 145, 148
onset cluster(s) 210212, 218 fixed-sentence vs. varied- E
s-cluster(s) 115, 117, 118, sentence 107 executive function 100, 101, 106
210212 following 133, 138, 140, 142,
consonant correctness 53, 54, 143, 145147, 149 F
56, 57, 5961, 6770 intervocalic 147 factorial typology 125, 126, 149
faithfulness 148, 173, 193, input 2, 14, 24, 71, 73, 8285, 127, plural 1416, 20, 2224, 114,
194, 196, 199201, 203205, 137, 148, 203, 204, 209212, 133, 134, 136138, 140147,
207, 209, 211213, 215220, 226, 227, 239, 242 161, 165, 166, 168
224, 225, 231, 236, 238240, interlanguage 25, 38 zero marking 141
242245 Iowa-Oto 177 morphophonology 133, 134
conspiracy 173, 199, 215, 216, multidimensional analyses 56
219, 220 J
cumulative faithfulness Jma 179, 180, 182, 183, 191 N
effect(s) (CFE(s)) 200 Nankina 188
204, 208211, 212, 214, 215, K Niuafoou 179, 183, 190194
216, 219, 220, 224 Karo 178, 179, 183, 186, 187 non-local 218, 219
multiply-unfaithful map- noun case morphology 164, 168
pings 207, 212214, L
216218, 220 language processing 91, 95, 96, O
fell-swoop 202, 208, 210213, 98, 100102, 105, 106, 109 obstruents 3, 33, 116, 117, 122,
215220 language separation 54 124, 136138, 140143, 145, 148,
French 19, 135 learning 11, 16, 19, 21, 25, 2729, 163, 178, 186, 201, 202, 203,
frequency effects 20, 24, 122, 127 4245, 71, 73, 75, 8285, 91, 207213, 220, 228, 239
9698, 100, 102, 104, 105, 109, see also consonant
G 219 Old Order Amish 157, 158, 160
gang effects 202, 207, 212, 224, algorithm 71, 84, 219 opacity 7, 205, 213, 243
233 lexical gradience 242 optimality theory 7, 84, 163,
geminate 178, 186189 Low German 157159, 162, 173175, 200, 224
generalization see phonological 168169 see also constraints
generalization Lyon Corpus 19
German 18, 19, 117, 155165, P
168170 M Pennsylvania Dutch 157
see also Pennsylvania Dutch, Mainstream American English Pennsylvania German 156158,
Pennsylvania German (MAE) 138, 143, 145147, 150 168, 169
varieties in Kansas 156, 162 markedness 24, 115, 116, phonemic contrasts 26, 27, 135
Gitanjali 192, 200 120, 175, 199, 200, 203, 204, phonological coding 91, 103
208, 211, 215218, 220225, phonological complexities 223
H 231, 233, 236238, 240, phonological disorders 26,
harmonic ascent 242, 244 242245 28, 49, 7174, 82, 8487, 92,
harmonic completeness 235, conspiracy 215, 216 113115, 118, 119, 126, 134
236, 240 constraint(s) 175, 200, 203, treatment of 6, 4951, 7179,
Harmonic Grammar (HG) 200, 204, 211, 216, 217, 223, 233, 8185, 91, 93, 94, 150, 207
206, 207, 209, 210, 212, 214, 236, 238, 242245 phonological generalization 72
216, 219, 220, 224, 225, 231, positional 175 74, 76, 77, 79, 83, 84
233237, 245 Marshallese 186188 across-class generaliza-
harmonicity see harmony minimization 225, 226, 231233, tion 71, 74, 7680, 82
harmony 217220, 233, 239, 240, 242245 lexical diffusion 71, 72, 82, 88
236238, 239243 morpheme 11, 12, 1417, 20, 21, trajectory of 74, 79, 80
Hawaiian 205 29, 31, 32, 133, 134, 136138, within-class 74, 76, 79, 80, 82
headedness 125, 126, 228 140146, 150, 177, 226 phonological mean length of
heritage societies 160 3rd-person singular (3s) 133, utterance (pMLU) (see whole
heritage speakers 157, 158, 160 134, 136142, 144147 word measures)
see also rememberers, semi- grammatical morpheme 11, phonological phrase 12, 1820
speakers 12, 14, 17 phonological processes see
hiatus 177, 180, 182 overt marking 113, 133, domain; processes
see also vowel sequences 135143, 145150 phonological similarity 49,
past tense 14, 17, 133, 134, 136, 5355, 57, 64, 65
I 137, 150, 162 Pirah 188
Ibibio 215, 216, 220 positional effects 13, 220
Subject index
position reduplication 115, 215, 220 sonority 115120, 122, 124131,

coda 59 simplification 15, 135137, 214 135138, 140, 143, 148, 175, 176,
final 13, 16, 17, 20, 33, 34, 135, stopping 201, 202, 204, 206, 178, 179, 183, 184, 190, 192196,
137, 141, 142, 146 210, 211, 213, 214, 220 210, 218, 232
initial 13, 27, 3234, 36, 37, 39, velar consonant deletion sonority distance 116, 120, 135
40, 43, 73, 75, 192, 226 (VCD) 225231, 235, 237, sonority index 116, 126
intervocalic 33 242, 243 Sonority Sequencing
medial 15, 16, 21, 32, 3641, velar harmony 218, 219 Principle 115, 137, 210
43, 142 velarization 155, 160, 162, 163, Spanish 1, 2, 4, 14, 18, 19, 2527,
onset 59, 117, 190 213, 214 2931, 34, 35, 38, 4043, 4547,
positional effects 13, 14, 31, vowel lengthening 175 54, 55, 6466, 116, 126, 135
141, 144, 148, 162, 220 vowel syncope (VS) 225231, speech-island 155157
positional augmentation 183, 233, 234, 237, 239, 242 speech perception 16, 84, 91, 97,
191 word-final devoicing 201, 100103, 106109
positional faithfulness 215, 202 Split Margin Approach 226, 231
216, 220 production accuracy 71, 72, 74, see also syllable
see also faithfulness 7679, 81, 82, 85, 115, 120 strict domination 202, 224,
positional markedness 175, prosody 232, 235
176 foot 12, 13, 1820, 177, 178, superlinear(ity) 173, 225, 238,
see also markedness 182, 183, 185, 189, 193, 194, 241, 243, 245
prominent 175 227230, 237, 240, 242244 see also constraints
see also strong position metrical restrictions 225, 230 syllable 1114, 16, 1823, 56, 115,
strong 220 Prosodically Licensed 11, 13, 117, 133135, 137, 138, 140, 147,
see also prominent 19, 21 148, 173, 175196, 201, 202,
position prosodic contexts 12, 13, 209, 211, 212, 215, 223, 225228,
syllable-margin 231, 232 16, 18 230233, 236, 239, 241245
predictors 94, 96, 104 Prosodic Licensing 911, 20, adjunct 1, 117, 125, 148
prepositional case 155, 165, 168 21, 23 appendix 134
processes prosodic word 11, 12, 14, 15, coda(s) 11, 1316, 20, 2224,
assimilation(s) 26, 155, 157, 1719, 134, 241 5764, 6770, 135, 136, 147,
163, 164, 215, 218 Providence Corpus 18, 19, 143 148, 176, 201, 203, 207, 209,
deaffrication 214 psycholinguistic theories 82, 211, 212, 213, 217, 226228,
deletion 126, 135, 194, 207 232, 239, 244
199203, 208, 209, 211213, extrasyllabic 135, 148
217, 219, 220, 225232, 234, R margin phonotactics 225
237239 rememberers 157 227, 239
disharmony 237, 239, 240, 245 see also heritage speakers, Split-Margin 232
domain of, see domain semi-speakers see also Split Margin
epenthesis 117, 118, 124, 182, root maximization 215 Approach
183, 199
glide formation 177 S T
medial cluster simplifica- s-cluster(s) see cluster Tmpisa Shoshone 179, 183,
tion 214 semantic transfer 155, 168, 169 185189, 191, 192, 195
metathesis 214 semi-speakers 157, 160 twins 5355, 58, 6266
nasal deletion 208, 209, 220 see also heritage speakers;
nasalization 135, 163, 164 rememberers U
obstruent deletion 208, 209, sensory, cognitive and linguistic unfaithful 173, 199, 200, 207,
212, 213, 220 processes 98 212220, 237, 238, 242
place harmony 217 Sesotho 19, 22
reduction 15, 75, 115, 119, 124, siblings 49, 5355, 6264 V
125, 127, 164, 167, 168, 199, sonorant 27, 116118, 125, 126, variability 1012, 2023, 26,
209, 210, 212, 226228, 137, 140142, 144146, 148, 91, 93, 94, 98, 99, 101106,
229231, 233, 234, 235, 237, 178, 208, 210212, 217, 220, 109, 127
239, 243, 242244, 226, 239 voice onset time (VOT) 26, 27
Volga Germans 160, 170 sequences of vowels 177, whole word measures 56, 5961,
vowel 180, 182 6770
duration 13, 28, 3335 see also hiatus phonological mean length
see also vowel length sonority of, 116, 178 of utterance (pMLU) 56,
height 116, 177, 190, 225, 227, syllabification of, 135 5863, 6770
228, 230, 231, 233, 234, whole word proximity 56, 64
237243 W word shape measures 60, 61
length 13, 135, 175, 178, Wargamay 175 working memory 21, 91, 99, 100,
185188, 225, 226, 235 well-formedness threshold 243 102104, 106108
see also vowel duration Wenker Sentences 161, 162
nasalization 135, 163, 164 Western Aranda 177 Y
reduction 17, 168, 227, 230 Yawelmani Yokuts 199

2014 Peresectives On Phonological Theory and Development

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2014 Peresectives On Phonological Theory and Development

Uploaded by

Copyright:

Available Formats

Perspectives on Phonological Theory and Development

Language Acquisition and Language Disorders

Volumes in this series provide a forum for research contributing to theories of

John Benjamins Publishing Company

theAmerican National Standard for Information Sciences Permanence

Library of Congress Cataloging-in-Publication Data

Perspectives on Phonological Theory and Development : In honor of Daniel A. Dinnsen /

2014 John Benjamins B.V.

Foreword and tabula gratulatoria vii

section 1. Representations and contrast: What does the learner know? 9

Prosodic Licensing and the development of phonological

Covert contrast in the acquisition of second language phonology 25

section 2. Sources of individual differences in phonological acquisition 49

Sibling rivalry: Comparing phonological similarity between

Abstracting phonological generalizations: Evidence from

Rapid phonological coding and working memory dynamics in children

section 3. Cross-linguistic approaches to phonological acquisition 113

What guides childrens acquisition of #sC clusters? A cross-linguistic account 115

The role of phonological context in childrens overt marking of -s

German settlement varieties in Kansas: Some unusual phonological

section 4. Theoretical advances in the field:

The role of onsets in primary and secondary stress patterns 175

A faithfulness conspiracy: The selection of unfaithful mappings

Superadditivity and limitations on syllable complexity in Bambara words 223

Author index 249

ra Birna sgeirsdttir Jeannette S. Leonard

(1) [sonorant, continuant] [voice]/___#

Shortly after meeting Mary Elbert, however, Dinnsen commenced pursuit of

qualifications, methodology, approach, testimony, and general demeanor (593

Representations and contrast

Much of Dinnsens early research into phonological acquisition was concerned

Demuth reviews a number of findings that suggest that much of childrens

Dinnsen, D.A. 1985. A re-examination of phonological neutralization. Journal of Linguistics 21:

There has been much research on the development of phonological representa-

be more reliably produced in some phonological/prosodic contexts than others.

Interactions at the segmental/prosodic interface

A childs use of a particular segment (phoneme) may be influenced by the prosodic

Interactions at the morphology/syllable structure interface

Interactions at the morphology/phrasal interface

Interactions at the morphology/prosodic word interface

What, then, do we make of the 50% use of a segment or morpheme? If it is

Brown, R. 1973. A First Language. Cambridge MA: Harvard University Press.

Fred R. Eckman, Gregory K. Iverson and Jae Yung Song

1. In describing the extent of the fricative-voicing phenomenon in Spanish, Martnez-Gil

fourteen participants produced a reliable distinction between these fricatives that

As we have detailed in other work, there is evidence that learning to contrast TL

involves suppressing the NL allophonic rule and results in implicationally related

The derived environment effect holds implications for IL phonologies in which

The second important aspect of (2) is that evidence of a derived environment

Another important correlate of the voiced-voiceless contrast for fricatives is

(a) Full (100%) overlap (b) Partial overlap

(c) No (0%) overlap

Table 1. Subjects performance, expressed as a percentage of target-like productions,

Subjects Word-initial Word-medial Inter-morphemic

[s] [z] [s] [z] [s] [z]

Contrast in basic environment

Contrast medially & in derived environment

Subjects Contrast initial Contrast medially Contrast inter-morphemically

Contrast in basic environment

3024 No Yes Yes

Position Mean SD Mean SD Statistics

Word-initial 2.27 5.66 43.90 38.76 t(13) = 5.87, p < 0.00

Next we examined which of the individual speakers showed a difference in the