 

   
 
 

Functionalist and
Usage-based Approaches
to the Study of Language
In honor of Joan L. Bybee

Edited by
K. Aaron Smith
Dawn Nordquist

   

Functionalist and Usage-based Approaches
to the Study of Language
Studies in Language Companion Series (SLCS)
issn 0165-7763

This series has been established as a companion series to the periodical

Studies in Language.
For an overview of all books published in this series, please see
http://benjamins.com/catalog/slcs

Founding Editor
Werner Abraham
University of Vienna / University of Munich

Editors
Werner Abraham Elly van Gelderen
University of Vienna / University of Munich Arizona State University

Editorial Board
Bernard Comrie Elisabeth Leiss
University of California, Santa Barbara University of Munich
William Croft Marianne Mithun
University of New Mexico University of California, Santa Barbara
Östen Dahl Heiko Narrog
University of Stockholm Tohuku University
Gerrit J. Dimmendaal Johanna L. Wood
University of Cologne University of Aarhus
Ekkehard König Debra Ziegeler
Free University of Berlin University of Paris III
Christian Lehmann
University of Erfurt

Volume 192
Functionalist and Usage-based Approaches to the Study of Language.
In honor of Joan L. Bybee
Edited by K. Aaron Smith and Dawn Nordquist
Functionalist and
Usage-based Approaches
to the Study of Language
In honor of Joan L. Bybee

Edited by
K. Aaron Smith
Illinois State University

Dawn Nordquist
University of New Mexico

John Benjamins Publishing Company

Amsterdam / Philadelphia
 TM The paper used in this publication meets the minimum requirements of
8

the American National Standard for Information Sciences – Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.

doi 10.1075/slcs.192
Cataloging-in-Publication Data available from Library of Congress:
lccn 2017049483 (print) / 2017054662 (e-book)
isbn 978 90 272 0022 8 (Hb)
isbn 978 90 272 6448 0 (e-book)

© 2018 – 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 Company · https://benjamins.com
 Table of contents

Acknowledgments vii
The authors’ reflections on Joan ix
Introduction xv
K. Aaron Smith & Dawn Nordquist
Features of some ergative languages that impact on acquisition 1
Edith L. Bavin
Constructional pressures on ‘sit’ in Modern Greek 17
Soteria Svorou
know and understand in ASL: A usage-based study of grammaticalized
topic constructions 59
Terry Janzen
Traces of demonstrative grammaticalization in Spanish variable subject
­expression: ella ‘she’ vs. él ‘he’ 89
Rena Torres Cacoullos
The company that word-boundary sounds keep: The effect of contextual
ratio frequency on word-final /s/ in a sample of Mexican Spanish 107
Earl K. Brown
Cumulative exposure to phonetic reducing environments marks
the lexicon: Spanish /d-/ words spoken in isolation 127
Esther L. Brown
A usage-based account for the historical reflexes of ain’t in AAE 155
K. Aaron Smith
Gradient conventionalization of the Spanish ­expression of ‘becoming’
quedar(se) + ADJ in seven centuries 175
Damián Vergara Wilson
 Functionalist and Usage-based Approaches to the Study of Language

The evidence add ups: An affix shift study of prefabs 199

Clay Beckner
look up about: Usage-based processes in lexicalization 225
Dawn Nordquist
About the authors 247
Index 249
 Acknowledgments

The editors would like to extend their gratitude to several linguistics scholars
who provided helpful feedback at various stages of the project: David ­Eddington,
Ulrich Utges, Miguel Simonet, Barbara Shaffer, Kenneth Wireback, James Walker,
Penelope Brown, Sandra Thompson and Sebastian Hoffman. A very special
­
­thank-you is owed to Östen Dahl for his typically magnanimous support.
 The authors’ reflections on Joan

Edith Bavin

Joan taught me when I first started my graduate studies in Linguistics in ­Buffalo.

Not surprisingly the first course with Joan was in Phonology. I did not realise then
that I would later be teaching phonology myself – first at Buffalo as a teaching
assistant and then at the University of Oregon. Over the years I was a graduate
student Joan taught in other areas. I remember historical linguistics and typology
lectures, and I also learned about the associations between child language and
language change. When I was in the final year of completing my PhD, Joan recom-
mended me for one of the rare jobs available in Linguistics, a one-year position
at the University of Oregon in Eugene. So my academic career started with a long
drive across the country and a dissertation in the final stages. Joan was on leave
in Berkeley soon after and I was able to meet with her there on several occasions
to discuss progress on my dissertation (and play tennis). This was well before the
days of desktop computers and email. I met Dan Slobin through Joan and he,
like Joan, encouraged my interest in child language. It became my main area of
 Functionalist and Usage-based Approaches to the Study of Language

research when I moved to Australia, first to Linguistics, which gave me opportuni-

ties to conduct field work in central Australia (as well as Kenya and Uganda, where
I worked on Western Nilotic languages) and then to Psychology, where I was able
to set up a lab. I’m glad that Joan could visit Melbourne and accompany me to
indigenous communities in central Australia and Groote Eylandt.
Joan instilled a strong foundation for research – in asking relevant questions and
finding answers. She was also a model in balancing work and social life. Even though
I am now officially retired I continue with my research, built on the strong theoretical
background I received at Buffalo. I hope that, in turn, my own training has benefit-
ted my former graduate students who are now making their own impact in the field.

Soteria Svorou

Teachers have a greater impact on their students’ lives than they ever imagine.
Joan Bybee, my mentor and doctoral thesis advisor, has been a great teacher for
me. Although it’s been almost three decades since approving my dissertation, my
professional life in teaching and research, and to a certain extent my personal life,
can be traced back to what I learned from Joan and under her watch.
Having arrived to the U.S. as a graduate student in linguistics at the University
at Buffalo, I had only a vague idea of what it meant to do research and to write
academic research papers. Through her carefully designed and well-run seminars
in phonology, morphology, and grammaticalization, Joan nurtured me in her craft
in a most gentle way and gave me a model that has inspired my own teaching. She
hired me as her research assistant, which gave me the financial support without
which I would not have been able to continue my studies. She trusted me to work
side by side with her as her apprentice on coding tasks in her universals of verbal
morphological categories project that grew into “The evolution of grammar” book.
The methodology and general theoretical framework of cross-linguistic compari-
son developed in that project formed the basis for the methodology I used in my
doctoral dissertation on cross-linguistic comparison of spatial grams. On Joan’s
recommendation, a version of my dissertation was eventually published as “The
grammar of space” in the Typological Studies in Language series.
In my own linguistics thinking and practice, I strive to follow on Joan’s steps,
albeit far from reaching her, and have benefited from her views and practices. Her
visionary theoretical position that language should be studied in its actual use that
forms its systematicity and history, even before linguists had the means to do so
in a systematic manner or were hesitant to go against the dominant theoretical
current, has finally come to be not only a possibility but by now a necessity for
discovering patterns in language beyond the most common ones. Her clarity of
thought, reflected in her straightforward expression, has further strengthened the
 The authors’ reflections on Joan 

promotion of that position through the decades and I, among a long list of other
linguists, had the fortune to be influenced by it. My current work on usage-based
grammar of constructions evolved naturally from that view.
Thank you, Joan, for being patient with me as I struggled to adapt to a new
educational environment and for giving me the tools and the inspiration to do
linguistics. You have shaped me as a linguist, a teaching professional and academic
and, ultimately, you have influenced my life. I am grateful for all.

Terry Janzen

I may have come to the University of New Mexico’s PhD program in Linguistics
knowing full well that I would learn plenty from Joan, but I could not have antici-
pated the extent to which she would influence me as a scholar and teacher. The
Linguistics PhD program was still new – I was admitted in the second year that it
was up and running, and I joined the three who had been admitted the year before.
Courses with Joan were intense! But I had come to learn from her and that’s what
I got. In her morphology course, she had written the textbook (­“Morphology”), in
grammaticization, again (“The evolution of grammar”), and in phonology, well,
we worked through her then manuscript which became ­“Phonology and language
use”. Among the lessons I learned, it seems, is that if you aren’t happy with the
available literature, write the text yourself. I now regularly teach a course on trans-
lation theory for which I edited, and significantly contributed to, the textbook
(“Topics in signed language interpreting: Theory in practice”), and it was one of
the best things I have ever done. Joan, you taught me well.
Joan had a far-reaching reputation for expecting a great deal from her stu-
dents. I think that at times this might have intimidated some of those in her classes,
whereas I first saw it as a challenge, and tried my best to respond. But soon I under-
stood. Joan herself is, and always has been, an exceptionally hard worker, demand-
ing more of herself than of anyone else, and it seemed to me that her expectation
was simply that if you want to be in the academic and scientific domain, then rigor
is the name of the game. Full stop. And so I saw that Joan was not asking anything
of anyone that she did not similarly ask of herself. Thus for me she inspired, and
continues to inspire, carefulness, thoughtfulness, thoroughness, and fearlessness, in
all aspects of research and teaching, and for this I am forever grateful.

Rena Torres Cacoullos

My respect for Joan, both as a fascinating and inspiring linguist and as a generous
and delightful person, was “at first sight” and still grows. To her I owe the i­ mpetus
 Functionalist and Usage-based Approaches to the Study of Language

for my first publication, an outgrowth of a paper in her grammaticalization semi-

nar at UNM; “The evolution of grammar” is still the most marked-up volume I
own. Her books are the main textbooks for my functional syntax and language
change courses, because they are inviting and get everyone’s brain working. Her
innovative but totally “makes sense” explanations show that “theoretical” linguis-
tics is not synonymous with formalist approaches – from Joan I learned the satis-
faction of integrating diachrony and synchrony, cognition and discourse.

Earl Brown

I first met Joan Bybee in August 2003 when I was a first-semester doctoral student
at the University of New Mexico, specifically in a “Phonological Representation”
course that I took with her. I was fortunate to take two additional courses with her,
and she was gracious enough to serve on my dissertation committee. The notion
that grammar is simply the cognitive organization of experience with language
was a new and fascinating idea that made sense to me. It is the basic premise that
has guided my research since then, and I continue today to apply the principles of
Joan’s theoretical framework as I use empirical methods to study language varia-
tion, especially in Spanish.

Esther Brown

To say Joan Bybee’s work has had an impact on my career feels like the understate-
ment of the century. Her scholarly achievements, made evident in academic titles,
positions, and publications, have earned her recognition as a scholar of renown,
whose impactful work shapes contemporary theory and methods in significant
ways. What may be less widely recognized, however, is that her level of scholarly
achievement is matched by her skill as an educator. I am privileged to be able to
attest from personal experience that she is a remarkable teacher. For close to two
decades, I have genuinely enjoyed thinking about the ideas, examples, hypoth-
eses and theories that Joan Bybee first introduced to me as a graduate student in
her seminars. The curiosity sparked in me by the concepts she imparted in class
(and in her subsequent publications) set me on a path of inquiry that has kept
me busy professionally for years. So much so, in fact, that a simple file search on
my work computer for the word Bybee returns 2,210 documents: more than pho-
nology, New Mexico, and variation. Now, for a former UNM Lobo who considers
herself a researcher of phonological variation, this search result makes evident that
the Bybean effect is strong. I am fortunate to have been introduced to her work
 The authors’ reflections on Joan 

­ rst-hand and to find I enjoy thinking about it as much as I do. With my contribu-
fi
tion to this Festschrift, I am delighted to now have 2,211 files on my computer that
pay tribute to such a spectacular career.

K. Aaron Smith

In the Spring of 1995, I was living and working in New York City when I happened
across a volume of “The evolution of grammar” in a bookstore. The title contained
my three favorite words, “Evolution,” “Grammar,” and “the,” and so I bought the
book. I had decided to take time off between my master’s and my doctorate, as I
couldn’t quite decide what PhD program would be best for me. After reading EoG, I
was no longer in any doubt; The University of New Mexico was my destination and
I wanted to “do” grammaticalization, or as I still like to say, “grammaticization.” For
the next years, I took seminars from Joan and wrote a dissertation under her direc-
tion. I say with very little hyperbole that my debt of gratitude for her scholarly and
professional mentorship and support can not be repaid by niether money nor deed. I
hope, however, that this Volume at least lets her know how very grateful I am to her.

Damián Wilson

As I was taking classes toward my master’s in Hispanic Linguistics during the early
2000s I wanted to know more about this grammaticization phenomenon that kept
coming up in our courses on sociolinguistics and bilingualism. I took that famous
Language Change course in the Department of Linguistics with Joan, which
changed the course of my studies. I had already become fascinated with the way
that language variation indexed social variables such as economic class and gen-
der. It was even more fascinating to learn about how variation had consequences
in contributing to larger evolutionary processes, such as grammaticization. I used
to refer to this area of studies as ‘brain candy’ as it was intellectually stimulating
to ponder the many ways that grammar changes through usage. At some point
Joan told me that I was decent at executing semantic analysis and she encouraged
me to continue the work she and David Eddington had done on Spanish verbs of
‘becoming’ for a final project.
Although I planned on conducting an ethnolinguistic study on Spanish language
transmission patterns in a New Mexican community for my dissertation I kept return-
ing to Joan’s encouragement, especially that I turn that final project into a disserta-
tion. One day it hit me: “If someone like Joan Bybee likes your work…”. Through her
encouragement I was able to successfully execute a diachronic study of the S­ panish
 Functionalist and Usage-based Approaches to the Study of Language

verb of ‘becoming’ quedar(se) and later develop that project into a monograph. The
book was more than just a valuable contribution to studies of language change – it
meant that I could sleep better (yes, it spent some time under my pillow).
I will never forget one of our many meetings at the Frontier when she told
me, “We need to talk about your writing.” I tried to act like nothing as I feared
the worst. She followed, “You start too many sentences with ‘this’ where I can’t tell
what ‘this’ is.” With a sigh of relief I promised to omit all ambiguous uses of ‘this’
and to this day insist that my graduate students to the same.
I am eternally grateful to Joan for her inspiration, kindness, and patience. I
think about her every time I am tempted to write “this”.

Clay Beckner

Over the years, Joan Bybee has been invaluable to me as a mentor, collaborator,
and friend. In particular, I would like to draw attention to her insights on language
as a complex adaptive system. I first met Joan as a student in her Usage-Based
Phonology class. The course provided a rigorous, empirically sound account of
language as a dynamic system, which changes according to bottom-up, domain-
general principles. Joan built on the research of precursors who worked in a com-
plex systems framework, and incorporated her own observations to offer a broad
and novel outlook on language. This course was nothing short of revelatory for
me: it made me realize I could unify my interest in complex systems (from a pre-
vious academic life) with my interest in linguistic research. This refreshing per-
spective was important in my decision to get a Ph.D. in Linguistics – and Joan’s
far-reaching contributions to phonology, morphosyntax, and diachronic linguis-
tics continue to inform how I think about language every day.

Dawn Nordquist

I first met Joan Bybee when I was a master’s student, new to Linguistics, in 1995.
Joan welcomed me to the department by being incredibly generous with her time
and resources, offering me several graduate assistantships, and Joan’s support
and encouragement were key to my decision to pursue a PhD. In fact, Joan’s 1998
seminar in Usage-Based Linguistics was the inspiration for my dissertation and
subsequent studies on collocations and semantic prosody. In addition to being a
supportive and guiding mentor, Joan has been a generous colleague and a friend.
Joan and I enjoy our lunches at a local Albuquerque favorite (The Frontier), and
she has opened up her home and ranch to me on a number of occasions, kindly
introducing me to my first horseback riding adventure!
 Introduction

K. Aaron Smith & Dawn Nordquist

Illinois State University / University of New Mexico

The inspiration for this volume finds itself in Professor Joan L. Bybee’s 2005
­Presidential Address for the Linguistic Society of America. In that address, later
published in Language in 2006, Professor Bybee outlined a usage-based model of
language storage and processing, one in which form is determined by the uses that
we put language to. The theory, which we return to in more detail below, can be
viewed as the culmination of Joan’s long career in linguistics.
Receiving her PhD in linguistics from UCLA in 1973, Joan’s first professional
appointment was as an assistant professor of linguistics (tenure-track) at SUNY
at Buffalo. Joan rose through the ranks of associate and full professor at Buffalo,
and in 1989 moved to the University of New Mexico, where she was named a
Regent’s Professor (from 1996) and Distinguished Professor (from 2005). She
also held important administrative positions at UNM (while staying academically
productive!), serving as Associate Dean of the College of Arts and Sciences from
1992–1993 and Chair of the Department of Linguistics from 1998–2001.
Bybee’s involvement in the field of linguistics, of course, has extended far
beyond the institutions in which she has taught and served. For several years, she
has been a professor in the LSA’s Summer Institute (1976 in Oswego, NY; 1980 in
Albuquerque, NM; 1995 in Albuquerque, NM; 2001 in Santa Barbara, CA; and
most recently as the Collitz Professor in Lexington, KY in 2017). She has also held
a number of fellowship and visiting scholar positions and she has worked with
scholars from various institutions around the world. From 1979–1980, she held a
Social Science Research Council Post Doctoral Research Training Fellowship dur-
ing which she studied child language development with Dan Slobin at the Univer-
sity of California, Berkeley. From 1983–1984, she held the Netherlands Institute
for Advanced Study in the Social Sciences Fellowship. From 1987–1988, she held
a Guggenheim Fellowship to study mood and modality in the languages of the
world and in May of 1990, she was a visiting distinguished fellow in the Depart-
ment of Linguistics at La Trobe University in Melbourne, Australia.
Joan’s work has been seen as important by both internal and external grant-
ing agencies. To date she and her colleagues on various projects/studies have
been awarded more than $300,000, some of which have come from the most

doi 10.1075/slcs.192.int
© 2018 John Benjamins Publishing Company
 K. Aaron Smith & Dawn Nordquist

prestigious (and competitive) granting agencies in the US, including several

grants from the the National Science Foundation. Joan’s influence on the field of
linguistics has also been exercised through her prolific editorial work. She has
served on the board of editors/advisory boards for several major journals and
book series of linguistic research (Linguistics, Studies in Language, Lingüística,
Studies in Language Companion Series, Morphology Yearbook, inter alia).
While many of the above accolades already speak to the field’s appreciation of
Joan, she has been awarded a number of honors that recognize her work as a scholar
and teacher. In 1976, San Diego State University recognized her with the Distin-
guished Alumni Award from the College of Arts and Letters (Joan completed her
master’s in linguistics at San Diego State). In 1997, she was the Astor Visiting Lec-
turer at the University of Oxford. She has been a Research Fellow of the LSA since
2006 and a Fellow of the Cognitive Science Society since 2016. And in September of
2005, the University of Oslo awarded her an honorary Doctorate degree.
Obviously, the kind of impact that a scholar like Joan makes in the field in terms
of grants, honors, awards, teaching and service comes about because of the wide
dissemination of her scholarship through publication. A prolific writer, Joan’s record
includes seven authored books, five co-edited books, some 34 articles in refereed
journals, and another 59 chapters in books. Joan’s scholarly work shows an intel-
lectual journey that spans more than four decades, during which time her work has
been at the fore of some of the most important movements and discoveries in the
science of language study.

Joan’s scholarly contributions in the field of linguistics

In the following, we give a biography of Joan’s scholarly contributions to the field

of linguisitics. In doing so, we necessarily present Joan’s work as being grouped
into various “phases”. To be sure, of course, the ability to segment Joan’s scholarly
activity into “phases” is a convenience of historiography. In a way similar to the fact
that speakers of English on Jan. 1st, 1500 didn’t wake up suddenly speaking Early
Modern English whereas the day before they had been speaking Middle English, no
“phase” of Joan’s work concluded the former, and certainly none precluded the next.

Natural generative phonology

The first publication to appear on Joan Bybee’s CV is “The syllable in phonologi-

cal theory,” which was published in Language in 1972. The article appeared while
Joan was still a PhD student in the Linguistics Department at the University of
 Introduction 

California, Los Angeles. That first publication shows us two things: One, that Joan
was to become an influential scholar in linguistics, and two, that the first area of
linguistics that Joan contributed her research and publishing energies to was in the
area of phonology.
During the early and mid-1970s, Joan published several articles, book chapters
and an edited book on phonology, and in 1976, Joan’s first monograph appeared,
“An introduction to natural generative phonology” (NGP). (NB. Joan’s last name
throughout the 70s appears as “Hooper”. From the early 80s, her publications
appear as “Bybee.”) “NGP” presents the theory of Natural Generative Phonology,
a theory that is evoked and worked out in several of the articles published by Joan
preceding the book treatment. (For a fuller account of the development of Natural
Generative Phonology, one can also see the several papers by Vennemann cited in
“NGP.”)
Natural Generative Phonology emerged as a theory during a time when
the dominant theoretical approach to phonology was generative (Halle 1962;
­Chomsky & Halle 1968), and “NGP” offered a new way of doing phonology,
largely by constraining the theory, particularly in terms of the allowable deriva-
tions between phonological representation and surface realization. On the one
hand this was achieved by limiting abstractness in rules and representations, and
on the other by allowing rules to refer to structure outside of the phonological
component, for example morphology. As Bybee argues, the advantage of differ-
ent kinds of rules for phonological alternations is that it more likely reflects the
representation of those forms and the forces that operate on them in the mind
of a native speaker (Hooper 1976: 11), which at that time she still referred to as a
speaker’s competence.
Re-reading “NGP,” it’s somewhat amusing to hear Joan Bybee arguing for tra-
ditional generative-type rules and referring to speaker’s competence versus their
performance (she would later abandon many of the tenets of generative linguistics
and see below). However, already in “NGP” we can see many of the concerns about
the importance of language use, diachrony and especially psychologically plau-
sible accounts for theoretical claims in linguistic theory.

Child language acquisition

Much of Joan’s work in Natural Generative Phonology presaged work that would
appear in the next arc of her scholarly work, which involved a period of focus on
child language acquisition. As mentioned previously, Joan spent the academic year
from 1979–1980 working with Dan I. Slobin on issues of child language develop-
ment. Out of that period of research, there appear a number of publications on
 K. Aaron Smith & Dawn Nordquist

the topic of child language acquisition. In a joint book chapter with Dan ­Slobin,
“Why small children cannot change language on their own: Suggestions from
the English past tense,” Bybee and Slobin compare innovative past tense forms in
various English verb classes produced by young children, 3rd grade children and
university-­aged adults to on-going changes in the English language. Their data
show that it is not only the innovations made by the very young children that cor-
respond to changes on-going in English but that adults and older children follow
the same patterns; in fact their data showed that in some instances the young chil-
dren do not show innovations observable in the history of English while the older
children’s and adults’ language does. They conclude then that the locus of language
change cannot be attributable in all cases to the imperfect learning of language by
young children acquiring their first language.
We have chosen this study by Bybee and Slobin as representative of Bybee’s
work in child language acquisition because it illustrates and adumbrates aspects
of her work that will appear in her later publications. (We should be clear that the
aspects of Joan’s work that we have chosen to discuss here in no way represents the
range of her contributions nor the “most important” – to attempt to do so would
be silly.) On the one hand, the challenge of such a subtle finding to the powerfully
dominant paradigms and tenets of linguistics at the time should not be overlooked.
At the time, the site of language change was believed to be located in the child lan-
guage acquisition process and the mere suggestion that some changes might not be
attributable to the language acquisition process and that those changes might even
be a function of language use by adults was (and still is for some) unadmissable. In
Joan’s work, both in Natural Generative Phonology and child language acquisition,
we continue to see that organization of morphological patterning played a central
role in Joan’s theoretical exposition.

Morphology

While Joan’s research has, since the beginning, always had some focus on morphol-
ogy, we define the period of Joan’s work between 1982–1985 as the “morphology”
phase, mostly because it is during that time that Joan worked on and published
one of her most cited and influential works, “Morphology: A study of the relation-
ship between meaning and form.” One of the study’s most impressive assets is the
design and compilation of a database of 50 languages, randomly selected with the
constraint that no two languages were of the same linguistic phylum nor “from
the same cultural or geographic area” (Bybee 1985: 25). The sampling design was
developed by Revere Perkins and described by him in Perkins (1980). Perkins also
chose the languages for Bybee’s study. The design of the database was important
 Introduction 

because it removed several of the known biases in database compilation, although

it was not free from its own problems, not the least of which was finding adequate
linguistic descriptions for the randomly selected languages in the sample. (NB.
This is still at issue when carrying out linguistics research and it is a call for why
we need more people to do language documentation, particularly on endangered
languages that are likely to disappear within the next generations.) That data-
base method would be expanded upon in Bybee, Perkins & Pagliuca (1994, and
see below).
In “Morphology,” Joan introduces and/or further articulates many of the
important aspects of the theories she had developed and would continue to
develop. One central theoretical aspect of the theory expounded in “Morphol-
ogy” was variable lexical representation. Up to the publication of “Morphology,”
the view of the lexicon was fairly static from the dominant generative approach.
Membership in the lexicon was an either-or proposition and much argumentation
concerned the question whether a lexeme belonged in the lexicon or not. As Bybee
(1985: 116) says in reaction to this view and in introducing her own model of
mental storage, “I now propose to abandon this restricted, binary way of thinking
about lexical storage, and treat the problem as the complex psychological problem
that it is.” That complexity, as described by Bybee, involves varying strength of
representation and connections that the word has with others in the lexicon.
As with much of Joan’s theorizing, part of her concern (and source of evidence
for her explanations) is to be found in the diachronic realm. As anyone who has
attempted to account for language change within generative grammar knows, the
theory does not lend itself easily to models of language change (particularly in its
earlier versions). In fact, and perhaps in abreaction to the 19th century focus on
language change, generative theorizing tended toward synchronic description; the
60s and 70s may be fairly characterized as a pointedly ahistorical time in linguis-
tics – but the rumblings of the importance of language change for linguistic theory
continued to sound, and sound loudly in Joan’s work. However it is in “Morphol-
ogy” that one might say that language change comes to take the central position
that it will have in Joan’s work for the next several decades.
Another aspect of “Morphology,” and one that will become increasingly domi-
nant in Joan’s work, is the interaction of form and meaning, both in diachronic and
synchronic description. The second half of “Morphology” tests various hypotheses
on the correlation of form and meaning within the domains of tense, aspect and
mood, and a good portion of that correlation is accounted for in the diachronic
develpment of tense/aspect/mood forms. More specifically, grammatical meaning
that is more general is shown to be older or more grammaticalized, and the forms
for expressing that more grammaticalized meaning will be more phonologically
reduced and more fused with the stems they inflect. It is the ­cross-linguistic v­ alidity
 K. Aaron Smith & Dawn Nordquist

of that correlation that will occupy much of Joan’s work for the next several years.
Although many pages in “Morphology” dealt extensively with the development of
grammatical expression, the term “grammaticization/grammaticalization” appears
infrequently, and it not listed in the subject index at all. (NB. that earlier in Joan’s
writing and in that of her students, the preferred terms was “grammaticization,”
while many other linguists preferred the term “grammaticalization”. To date, the
field has decided on the term “grammaticalization,” and for consistency we have
adopted the term here, as has Joan in her more recent writing.) The absence of that
term in “Morphology” may seem remarkable for a scholar who would contribute
so much to the area of grammaticalization studies.

Grammaticalization

Grammaticalization, which in very general terms is concerned with the ways

that grammatical structure develops (often from lexical material) and continues
to develop over time, is hardly a 20th-century concern. Questions about the
source of grammar reach far back into history, and have been asked by think-
ers on language within many different traditions around the world at different
times. The more specific term “grammaticalization” in the western tradition of
linguistics is usually attributable to Antoine Meillet from about 1912. However,
despite the long interest in the problem of the origins of grammatical forms,
theories of linguistics developed in the US during the early and middle part of
the 20th century would not provide fertile ground for the study of diachrony,
grammaticalization included.
Throughout the latter half of the 1980s and into the 1990s, Joan (and col-
leagues) produced a prodigious amount of scholarship that we will characterize
here as falling under the name “grammaticalization.” We do not want, however,
to circumscribe or limit the more general theoretical importance of any of the
work done under that designation, and as we will discuss below, Joan’s work in
grammaticalization both built upon work she had been doing since the 70s and
laid the groundwork for the next “phase” of her scholarship, usage-based gram-
matical theory.
In 1985, the same year as the publication of “Morphology,” Östen Dahl from
the University of Stockholm published a book titled “Tense and aspect systems.”
Dahl’s “Tense and aspect” intersected with “Morphology” at several points. Aside
from the obvious focus on verb systems, both had, more or less, explicit diachronic
approaches. In 1989, Joan and Professor Dahl published a comparison of their
two studies in the journal Studies in Language, titled “The creation of tense and
aspect systems in the languages of the world.” The title is telling. On the one hand,
 Introduction 

“creation” signals clearly that the engagement is diachronic, and on the other that
the hypotheses being developed and tested are intended to make universal claims.
(Universals and typology are omnipresent in the work we characterize here as
“grammaticalization.”) The paper is of importance to the study of grammatical-
ization because it validated the findings of both “Morphology” and “Tense and
aspect” since despite very different methodologies and sources of data (Dahl’s study
was based on a questionnaire), the two studies showed stikingly similar results;
very specific diachronic patterns WERE universal in scope. Also, it validated the
approach that Joan had taken to studying diachronic patterns in the development
of tense, aspect and mood. Her next book, would expand her method and scope of
investigation in the grammaticalization of verbal morphosyntax, and become an
exceedingly influential book in modern linguistics.
“The evolution of grammar” (with Revere Perkins and William Pagliuca) came
out in 1994 and established itself as a seminal text within the area of grammatical-
ization studies. In the book, Bybee et al. report on the grammaticalization of tense,
aspect and mood from their database (Gramcats), a stratified sampling of 76 maxi-
mally unrelated and non-contiguous languages. Their findings uphold many of
those reported on in “Morphology” and in Bybee and Dahl (1989). One important
focus of “Evolution” is its expansion of the mechanisms involved in the kinds of
semantic changes to be regarded in grammaticalization. Those mechanisms, how-
ever, were not only interesting in light of grammaticalization, but many of them
have been found to operate on language more generally, or in other words seem
to point to more general ways that the mind stores, processes and uses language.
It was then fitting that in the next stage of her career, Joan’s theories would turn to
address larger questions concerning an explicit theory of grammar in the mind.

Usage-based theory

We round out this discussion of Joan’s illustrative career with her contributions
to Usage-Based Linguistics. Langacker (1987: 494) first defined a usage-based
approach as one in which “substantial importance is given to the actual use of
the linguistic system and a speaker’s knowledge of this use; the grammar is held
responsible for a speaker’s knowledge of the full range of linguistic conventions,
regardless of whether these conventions can be subsumed under more gen-
eral statements.” Langacker’s use of the term was in direct response to rule-list
approaches to language that assigned unpredictable forms to the lexicon while
a separate grammar rule was posited to generate all regular forms. Langacker
argued that evidence for the primacy of a disembodied, maximally-general rule or
schema was lacking. Since then, of course, usage-based linguistics has developed
 K. Aaron Smith & Dawn Nordquist

into a theory (not just an approach) which posits that the intersection of language
use and domain-general cognitive abilities are centrally responsible for the inher-
ent gradience and variability of all structures observed in language.
The preceding sections highlight that Joan has embraced usage-based lin-
guistics throughout her career in some form or another and under one guise or
another in that she has always/consistently viewed grammar as the cognitive orga-
nization of language users’ experience with language (Bybee 2006). Her recogni-
tion of the role that language use plays in phonological representation (e.g., Bybee
2002), degrees of morphological paradigmatic productivity (e.g., Bybee 1995), and
grammatical categories’ emergence and organization (e.g., Bybee 1998) all rest
on the notion that usage patterns, especially frequency of use patterns, interact
with processing abilities whereby repeated linguistic events are recognized and
tracked by language users endowed with general cognitive capabilities. Indeed,
Joan’s research on language change has demonstrated that her contribution to the
field of linguistics under the general terminology of usage-based theory is her fin-
est articulation of a functionalist theory of grammar, particularly in phonology
and morphology. However, in more recent years, Joan has turned her usage-based
lens to questions of constituent structure and the emergence of morphosyntac-
tic constructions, culminating in her 2010 book “Language, usage and cognition”.
Of particular import is the claim that constituency, the hallmark of grammatical
structure, is neither domain-specific nor a priori. In this way, Joan has advanced
usage-based linguistics as a theory that has as its starting point that language at
all levels of structure may be explained through domain-general cognition that
applies to the linguistic experiences that speakers/signers track, chunk, store, and
organize within a network of phonetically- and semantically-specific exemplars
along with associated higher order generalizations tied to these richly-detailed lin-
guistic memorial representations.

Linguistics and beyond

This survey of Joan’s career would not be complete without recognizing her contri-
butions to another field that she has embraced whole-heartedly: land conservation
and sustainable ranching in New Mexico. As with her service to the field of lin-
guistics, Joan has similarly dedicated herself to environmentally-sound ranching.
Since before her retirement, Joan has teamed with a number of land management
and ranching organizations to host and present at workshops on restoration of
natural lands, riparian restoration, and sustainable grazing practices. From 2006–
2014, Joan served on the Board of Directors for the Quivira Coalition, a non-profit
organization whose mission is to develop economic and ecological resilience for
 Introduction 

western lands, and she was the Chair from 2008–2013. She also currently sits on
the Board of Directors for Southwest Grassfed Livestock Alliance. On her own
­Mesteño Draw Ranch, Joan raises cattle where she espouses environmentally-
friendly practices. Her cattle graze on native grasses and forbs and consequently
her ranch offers beef that is free of hormones and antibiotics while reducing the
carbon footprint. It is no surprise, then, that, in 2004, Joan was recognized as Out-
standing Conservation Rancher by the Claunch-Pinto Soil and Water Conserva-
tion District in New Mexico; it is noteworthy that a linguistics scholar who has
long recognized that language use impacts language form and function, has also
embraced the fact that land use impacts the form and function of the environment.

References

Bybee, Joan L. 1985. Morphology: A study of the relation between meaning and form. Amsterdam:
John Benjamins. doi: 10.1075/tsl.9
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: CUP.
Bybee, Joan. 2006. From usage to grammar: The mind’s response to repetition. Language 82(4):
711–733.
Bybee, Joan. 2002. Word frequency and context of use in the lexical diffusion of phonetically
conditioned sound change. Language Variation and Change 14: 261–290.
Bybee, Joan. 1998. A functionalist approach to grammar and its evolution. Evolution of
­Communication 2: 249–278.
Bybee, Joan. 1995. Regular morphology and the lexicon. Language and Cognitive Processes 10:
425–455.
Bybee, Joan L. & Dahl, Östen. 1989. The creation of tense and aspect systems in the languages of
the world. Studies in Language 13: 51–103.
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar: Tense, Aspect,
and Modality in the Languages of the World. Chicago IL: University of Chicago Press.
Chomsky, Noam & Halle, Morris. 1968. The Sound Pattern of English. New York NY: Harper
and Row.
Halle, Morris. 1962. Phonology in generative grammar. Word 18: 54–72.
Hooper, Joan B. 1976. An Introduction to Natural Generative Phonology. New York NY:
­Academic Press.
Langacker, R. 1987. Foundations of Cognitive Grammar, Vol. 1. Stanford CA: Stanford University
Press.
Perkins, Revere D. 1980. The evolution of culture and grammar. SUNY/Buffalo Dissertation.

The papers

It is to be expected that Professor Bybee’s students would become engaged in the various
research areas that characterize her long career as a result of the time that they studied under
her tutelage. The organization of this book reflects those areas (see our discussion of Professor
 K. Aaron Smith & Dawn Nordquist

Bybee’s ­contributions to the areas of child language acquisition, phonology, morphology,

grammaticalization, and usage-based linguistics above). Thus, the first chapter of this book,
titled “Features of some ergative languages that impact on acquisition”, is a study of child
language acquisition in which Edith Bavin investigates the role of cues in children’s acquisition
of ergative or split ergative verb patterning. If a child is acquiring a verbal system in which there
are both ergative-absolute and nominative-accusative patterns, one might hypothesize that
acquisition of such patterning would take longer. However, as Bavin argues from the literature
and from her own research, there appear to be no delays in verbal morphosyntax acquisition.
Bavin argues that the available structural cues in those languages are sufficient to build up adult-
like (or input) representations of the relevant constructions in the child’s grammar.
The next section of the book contains three papers that fall largely under grammaticaliza-
tion studies. The first paper of the section by Soteria Svorou, titled “Constructional pressures on
‘sit’ in Modern Greek” considers the incipient constructionalization of the Greek postural verb
kaθome ‘sit’ in syntactic conjunction with other verbs, which she schematizes as Vkaθome ke V2.
Svorou’s analysis reveals a range of semantic types of the Vkaθome ke V2 construction that arise
from the interplay of lexical semantics of kaθome, the lexical semantics of the V2, the morpho-
logical aspect of the verbs in the construction, and adverbial expressions, inter alia. The mean-
ings of the emerging Vkaθome ke V2 types, however, cannot be derived compositionally from
the elements of the construction, but are better accounted for as meanings that have become
conventionally associated with the constructions over time.
The second paper of this group by Terry Janzen is “know and understand in ASL:
A usage-based study of grammaticalized topic constructions”. Janzen analyzes the use of cogni-
tive verbs know and understand in American Sign Language (ASL) to mark certain topics.
Utilizing a corpus of conversational ASL, Terry Janzen identifies that know, when it appears in
topic marking constructions, does not reference mental activity, but rather serves the function
of marking the topic phrase as signaling domain-level, rather than specific, referential identifi-
able information. Moreover, as is typical of grammaticalizing structures, know tokens show a
type of phonetic reduction (undershoot), such that the target location of know token articula-
tions are often found lower on the face than source lexical know which is signed higher on the
face, making contact with the forehead. understand, also a grammaticalizing topic marker in
ASL, differs from know in some of its specific topic marking functions. understand also lacks
the phonetic reduction of undershoot that the know tokens exhibit. Janzen argues that this
phonetic difference is likely a result of the difference in the production of the two signs canonical
lexical forms, but that there also may be some frequency effects. The paper highlights the gradi-
ent nature of grammaticalization (as is the notion of topic) as it lies at the center of pragmatic,
structural and semantic factors that characterize interactive communication.
The final paper of this section by Rena Torres Cacoullos treats the grammaticalization of
third person subject pronouns in Spanish. Cacoullos’ study (“Traces of demonstrative grammat-
icalization in Spanish variable subject expression: ella ‘she’ vs. él ‘he’”) focuses on an important
but sometimes overlooked aspect of Bybee’s contribution to our knowledge of how grammatical-
ization works. While many studies invoking a grammaticalization framework focus on semantic
generalization and the correlative effects of distributional widening, much of Bybee’s work in
the area has shown that even with the loss of some of the specificities of meaning, grammatical-
ized elements retain certain aspects of their source semantics. Cacoullos, noting the undisputed
origin of Spanish 3rd person singular pronouns, él and ella, as having their source in Latin distal
demonstratives, seeks evidence in earlier Spanish for retention of some distal meaning. Her
analysis finds support for the source retention hypothesis in that greater distance of previous
 Introduction 

referent, syntactic function of previous referent in non-subject role, and feminine gender, which
represent increased distal reference and reduced topicality in the texts examined, influence the
selection of a 3rd person singular overt subject pronoun.
The next section of this book turns to the area of usage-based phonology. Both papers in
this section provide analyses of reductive phonological processes in Spanish. In Earl Brown’s
“The company that word boundary sounds keep: The effect of contextual ratio frequency on
word final /s/ in a sample of Mexican Spanish”, the author examines how raw lexical frequency
and the frequency with which a word appears in a reduction-favoring context (contextual ratio
frequency) predict reduction of /s/ as measured by duration and center of gravity. While raw
lexical frequency does not predict either shorter duration or decreases in center of gravity for
word-final /s/, contextual ratio frequency, for this data set, does operate as a significant inter-
active predictor for /s/ reduction so long as the following context is favorable (i.e., a following
consonant). Given the lack of predictive power of raw lexical frequency in this study and ceiling
effects for lexical frequency in other studies, Brown contributes to the ongoing research into the
complexities of usage-based frequency effects by investigating the relatively under-researched
area of contextual ratio frequency.
The second paper in this section also investigates the role of contextual ratio frequency or
frequency in reducing contexts (FRC) and complements the findings of the first paper. Esther
Brown’s “Cumulative exposure to phonetic reducing environments marks the lexicon: Spanish
/-d/ words spoken in isolation” applies a usage-based approach to phonology in which words
are stored in a “cloud” of exemplars that may differ based on a speaker’s experience with the
word. These experiences include the phonetic shape of words with reduced segments. In other
work with her colleagues, Brown has found the frequency in reducing contexts (FRC) to be a
strong indicator of the appearance of a reduced variant of the word. However, it is not clear that
the reduced variants of the word represent stored exemplars in the lexicon or simply represent
a planning strategy for a reduced articulation because the speaker has recognized a potential
reducing context. Within a usage-based account, there may be any number of factors, both
linguistic and extra-linguistic, that would favor a reduced variant, presumably reinforced by
former experience with the reduced variant and those reducing factors. In this study, Brown
designs an experiment to test whether the reduced variant is more likely to be stored in the
lexicon or is indicative of on-line manipulation of the word by the speaker. The results of her
experimentation suggest that when contextual information is minimized or removed, the FRC
effect remains, suggesting that those reduced forms are being pulled from the lexicon.
The final section of the book explores topics within the area of usage-based grammar, that
is, grammar as an emergent phenomenon of language use. The first paper in this section by
K. Aaron Smith titled, “It ain’t over yet: Usage-based pressures on the spread of auxiliary ain’t in
African American English,” provides a grammatical analysis of the auxiliary verb ain’t in certain
varieties of English in which it is underdetermined for tense and aspect meaning. On the one
hand, Smith provides some evidence that the increased use of auxiliary ain’t in certain tense
domains may be promoted by frequent exemplars within that doman, but he also argues that the
ain’t auxiliary, because of its broad distribution across many tense and aspect domains, has con-
tinues to lose specificity. Smith concludes that this apparent simplification in grammar is offset
by the assoiciation of the form with community practices, thus pushing usage-based representa-
tion beyond the realm of linguistic structure to include sociolinguistic identity.
The next paper in this series treats emergent productivity of constructions involving verbs
of ‘becoming’ with adjectival complements in Spanish. In “Gradient conventionalization of
the Spanish expression of ‘becoming’ quedar(se) + ADJ in seven centuries”, Damián Wilson
 K. Aaron Smith & Dawn Nordquist

looks at the diachronic evolution of two categories of adjectives used in the quedsar(se) + ADJ
­construction: adjectives that describe changes in human company and adjectives that describe
‘happiness’ emotions. Relying on type frequency as one predictor of a specific categorie’s pro-
ductivity, Wilson considers other usage-based factors that can impact a category’s productiv-
ity over time. In ­particular, Wilson proposes an operationalization of conventionalization that
relies on corpus factors, such as type frequency and token frequency, in addition to usage fac-
tors, such as widespread usage and previous usage. Wilson provides a detailed analysis of each
adjectival category from the 1200s to the 1900s and examines the way that category structure
interacts with evolutionary trends: highly coherent categories that are structured around a single
highly conventionalized adjective type endure more robustly than more scattered categories.
In the third paper, Clay Beckner examines how usage impacts storage of English multi-
word expressions. In his “The evidence add ups: An affix shift study of prefabs”, Beckner exam-
ines the premise that multi-word units are stored holistically and therefore processed holistically.
To test this hypothesis, the author collects experimentally-induced affix positioning errors for
prefabs, predicting that the more unit-like a prefab, the more likely a speaker may externalize
the third-person singular agreement suffix, a behavior which is congruent with the notion that
the prefab was accessed as a whole without processing its internal structure. In line with research
on entrenchment and holistic storage, Beckner’s stimuli include data which vary in terms of
token frequency and in terms of relative frequency. Beckner’s experimental speech error data
distribute with respect to both usage factors in predicted and unexpected ways. The presented
analysis argues that both the bigram frequency of the prefab and the individual frequency of the
prefab’s component words are relevant usage factors which affect structure, reflected in holistic
processing as evidenced by the distribution of affix shift errors in the data.
The final paper of this section, and the book, is fittingly written by Professor Bybee’s final
dissertation student before her retirement. Dawn Nordquist’s “Look up about: Usage-based
processes in lexicalization” offers an account of emergent lexicalization of a new phrasal-­
prepositional verb in English. Drawing from online data, Nordquist presents evidence that
the string acts as a new verb, the semantics of which is largely synonymous with the verb “to
google”. The paper then addresses how the multi-word lexical expression may have emerged.
Although the exact source of the verb cannot be identified, Nordquist presents evidence of a
likely source construction and argues that Bybee’s (2002) Linear Fusion Hypothesis which states
that “sequential co-occurrence is more basic than constituency” can explain how the verb might
have emerged from the identified source construction. The paper therefore contributes to the
idea that usage-based experiences are basic to the emergence of language form.
 Features of some ergative languages
that impact on acquisition

Edith L. Bavin
La Trobe University

Bybee’s (e.g., 2006, 2010, 2013) view is that grammar is the cognitive organisation
of one’s experience with the language. Evidence from the language acquisition
field supports this approach. A major task for the child in acquiring a language
is detecting distributional properties of the language and the co-occurrence of
features, that is, identifying the form-function units in the language input. In this
chapter, I briefly discuss ergative alignment and present examples from research
on the acquisition of ergative languages to show that different cues to ergativity
are provided. The examples illustrate that children quickly attune to contexts for
ergative marking, identifying regularity and recurring patterns as they construct
a grammar.

Introduction

Bybee has argued that grammar is the cognitive organisation of one’s experience
with the language (e.g., Bybee 2006, 2010, 2013). When acquiring a language, two
important aspects of experience are: the language structures in the input and the
social context in which the child develops (Hoff 2006). In acquiring a language
the main task for children is to detect the distributional patterns in the input lan-
guage – that is, how forms are distributed and their function – and form men-
tal representations of these patterns in order to generalise to new instances. An
assumption is that the child becomes sensitive to co-occurrence probabilities. The
view presented by Thiessen (2016) is that becoming sensitive to central features of
a set of exemplars represents statistical learning; it involves “the extraction of ele-
ments of the input into memory traces, and subsequent integration across those
memory traces that emphasize consistent information” (p. 1). From research find-
ings on the development of specific features within a grammar and comparisons
of development across languages, it is evident that children are rapidly attuned
to distributional properties of the language(s) and the co-occurrence of features,
that is, identifying the form-function units (the frequent sequences of words
and morphemes, e.g., Bybee 2006; Goldberg 2006) that form processing units

doi 10.1075/slcs.192.01bav
© 2018 John Benjamins Publishing Company
 Edith L. Bavin

(Bybee 2010).1 Noun + case marking, verb + inflections and syntactic patterns of
verb + nominals are examples of types of constructions. Thus the ordering of ele-
ments, case marking, and verb inflections are among the features that can serve as
cues to the grammar (see Bates & MacWhinney 1989). In recognising recurring
patterns in the language, children draw on their perceptual and cognitive skills,
including categorization and analogy (Tomasello 2003, 2015), and they rely on
memory skills to form representations; these skills need to be sufficiently devel-
oped to process the input.

Cues to acquisition

As discussed by Bates and MacWhinney (1989), cues compete and apply differ-
ently across languages; their value depends on ‘cue validity’, that is their ‘avail-
ability’ (the fact that a cue is readily available for the child to use) and ‘reliability’
(which is established if a cue leads to the correct functional choice). When word
order is variable it is not a reliable cue to the role of the arguments in a sentence,
and if there is frequent ellipsis of core arguments in a case marking language, the
case forms will not be overt and so are not available.
A large body of research on the acquisition of languages has focussed on verb
argument structure and children’s sensitivity to particular verb argument con-
structions. In English word order is a valid cue for identifying the associations
between a verb and its arguments; the subject appears before the verb and the
object follows. Using intermodal preferential looking experiments with toddlers
acquiring English, researchers (including Naigles 1990; Naigles, Bavin & Smith
2005; Gertner, Fisher & Eisengart 2006) have shown that even before children
produce Noun-Verb-Noun sequences, they attribute a meaning to them: one par-
ticipant (the first noun) acting on another. When this meaning is assigned with
novel verbs in the test sentences it is evident that the children have generalised

. From the data available it is clear that neither token frequency nor distributional fre-
quency can be the sole determining factor of how rapidly a child acquires specific construc-
tions in the input language. Support for this view is presented by contributing authors to the
Journal of Child Language special issue on frequency (2015). As argued by O’Grady (2015),
frequency interacts with other factors so adding to processing costs. O’Grady provides the
example of pronouns in English: identifying the referent for personal pronouns takes longer
for children than identifying the referent for reflexives, and yet personal pronouns are more
frequent in the input. The explanation provided is the proximity of the antecedent in the
clause. If two associated elements are distant and it take longer to identify the association,
memory is implicated.
 Features of some ergative languages that impact on acquisition 

established associations between syntax and semantics. However, word order is

variable in many languages, frequently influenced by discourse constraints: the
arguments may appear in different positons in relation to the verb within a sen-
tence and so position of a nominal is not a reliable cue to its role in the sentence.
In addition, core arguments are often ellipsed; in those contexts the case markings
are not overt, that is, not available. In contrast to languages that rely on the order
of words to distinguish the core arguments, as in English, in Mayan languages it
is the ordering of morphemes within the verb cluster that fills this purpose (Brown,
Pfeiler, de Leon & Pye 2013).
In the following sections of the paper I discuss some features of specific lan-
guages that children seem to use to detect distribution patterns in the language; I
use examples from ergative languages because much of the acquisition literature
and theoretical discussion of language acquisition is based on English and other
European languages, but theories need to be able to incorporate languages that
are typologically different. A brief overview of some of the features that have been
shown to impact on children’s acquisition of the ambient language should be of
value to students learning about child language development (from various dis-
ciplines) and also to linguists whose focus is not language acquisition. Important
for theory development is comparison of acquisition across typologically different
languages as well as closely related languages (see Pye 2017). In the next section I
give an overview of ergative alignment; this is followed by a section with examples
from some ergative languages.

Ergative alignment

A distinction is made between ergative-absolutive and nominative-accusative

languages. The difference is in how the languages treat core arguments: subjects
of transitive (A) and intransitive verbs (S) and objects of transitive verbs (O).
In nominative-accusative languages, A and S are grouped together somehow,
by word order, case marking, agreement or control; the A and S arguments are
distinguished from O arguments. However, in the canonical pattern of ergative
languages (see Comrie 2013), S and O are grouped together and these arguments
are distinguished from A arguments. Variation from this pattern is often shown
in ergative languages. For example Hindi has what is referred to as a split sys-
tem. Ergative alignment is not always used; it is used with perfective aspect, but
not imperfective (Narasimhan 2005, 2013), and there are other conditions for
using or not using ergative alignment. In the next section I use some of Nara-
simhan’s examples to illustrate the split system. Because the subject of a transitive
(two arguments) or ditransitive (three arguments) verb is distinguished from the
 Edith L. Bavin

s­ ubject of an intransitive (one argument) verb in the canonical ergative pattern,

a main issue for acquisition is whether the child learner initially fails to make the
distinction.
Each of the ergative languages discussed in the next section imposes particu-
lar challenges for acquisition The descriptions of Hindi are based on Narasim-
han (2005, 2013). My own research is the basis for the examples from Warlpiri;
the research was conducted in Yuendumu in central Australia (Bavin 2000, 2004,
2013). A brief discussion of research findings from Mayan languages reported by
Brown et al. (2013) is also included. (Also see Brown 2007, 2008; Pye, Pfeiler, de
León, Brown & Pedro Mateo 2007).2 The data reported in the sources for all the
languages I discuss are naturalistic, collected in the child’s natural environment,
not based on questionnaires or experiments. In their discussion of the acquisition
of three ergative languages of Papua New Guinea, Rumsey, San Roque and Schei-
ffelin (2013) illustrate that interactional routines may provide both structural and
situational templates for the use of the ergative. Structural aspects are the focus of
the current paper.

Acquiring ergative languages: Some examples of influencing factors

Hindi
The Hindi split ergative system is illustrated in (1) (from Narasimhan 2013). In
(1a) ergative case marking (ne) is used with the perfective verb, but in the non-
perfective sentence in (1b) there is no ergative case marker, and nor is there in (1c)
which has a one-argument verb. The subject is in nominative case.3

1. a. laRkii-ne kitaab-Ø paRh-ii.

girl.sg.f-erg book.sg.f-nom read-sg.f.pfv
‘(The) girl read (a/the) book.’

. Bavin and Stoll (2013) provide a more detailed work on the acquisition of ergative lan-
guages.
. The abbreviations used are:
sg singular
f female
erg ergative
pfv perfective
nom nominative
pres present
 Features of some ergative languages that impact on acquisition 

b. laRkii-Ø kitaab-Ø paRh-tii hae.

girl.sg.f-nom book.sg.f-nom read-sg.f.ipfv be.3sg.pres
‘(The) girl reads (a/the) book.’
c. laRkii-Ø gir-ii.
girl.sg.f-nom fall-sg.f.pfv
‘(The) girl fell.
We might expect that learning the restrictions on ergative alignment in the lan-
guage, as with other languages with split ergative systems, would lengthen the
acquisition period and overgeneralizations would be evident. That is, children
would initially overextend ergative-absolutive alignment before identifying the
specific contexts in which it does/does not apply. However, this is not supported;
Narasimhan (2013) reports that overgeneralisations were not evident in the longi-
tudinal data collected from three young children who were acquiring Hindi.
In a corpus of 5412 verbal utterances in the input to the three children (when
they were aged 1;5–1;7 years for one child, 3;1–3;2 years for a second, and 2;3–2;4
years for the third), the number of tokens of ergative case was just 1.76% (95/5412).
However, an influencing factor was that perfective verb forms were not used as
frequently in the input as imperfective forms (also see Narasimhan, Budwig &
Murty, 2005). Interactions with young children generally focus on the immediate
context rather than past events, and this was so for the Hindi data. The result was
that the frequency of ergative contexts in the input was reduced. In addition, core
arguments were frequently ellipsed, so the ergative case marker was not overt. This
affected its availability as a cue to ergative contexts.
The verb + aspect construction was readily acquired by the children stud-
ied, which suggests that verb inflections could serve as a cue to ergative contexts
(­Narasimhan 2013). The verb in Hindi agrees with the gender of one of the core
arguments but which one controls agreement depends on the case markers used
on the subject and object. If two non-nominative arguments (e.g., ergative and
dative) appear in a sentence, masculine singular agreement on the verb is used
as a default (Narasimhan 2013). In the input data reported, perfective verbs with
masculine singular agreement frequently co-occurred with the ergative ne case
marking. This co-occurrence of aspect marking and ergative case could form the
basis for generalising to other perfective contexts, even without overt ergative case
marking. While no overextensions of ergative case marking were found in the
child data, there were some errors of omission of ne in obligatory contexts. How-
ever, for the youngest of the three children, the ergative case form appeared in all
obligatory contexts by age 2;1 (Narasimhan 2013: 228).
Also proposed by Narasimhan (2013) as a possible cue to ergative alignment
is the use of specific ‘light’ verbs (which have aspectual function) with the main
verbs. The two forms occurring with ergative alignment structures contrast with
 Edith L. Bavin

one light verb used in non ergative contexts. Thus, when a perfective light verb
appears with a main verb it provides a cue to alignment.
In summary, a number of language specific cues to alignment are available to
children acquiring Hindi. A arguments are frequently ellipsed and case markers
are then not available, but verb morphology is a cue to ergative alignment, as is the
co-occurrence of specific light verbs.

Warlpiri
Warlpiri is a Pama-Nyungan language of central Australia (Hale 1982; Laughren,
Hoogenraad, Hale & Granites 1996; Nash 1986). Verbs are distinguished by their
case frame; the system is split in that the ergative case marking system represents
ergative alignment but cross referencing of core arguments follows a nominative-
accusative system. Most two- or three-argument verbs require an ergative argument
as the primary argument (with ergative/absolutive or ergative/absolutive/dative
case frames) and when overt it is case-marked as ergative. The primary argument of
non ergative case-framed verbs is nominative (with zero case marking). Although
morphologically ergative in terms of case marking, for the cross-referencing sys-
tem for core arguments in Warlpiri the same clitic form is used to cross-reference
an A and an S argument; a different form is used for an O. An example of a cross
reference marker (-rna 1sg sbj) is shown in 2b. There is no distinction in the cross-
referencing morphemes for ergative and non ergative verbs.
The language has variable word order and frequent ellipsis of core arguments.
In spite of the variable word order there are ordering restrictions within construc-
tions. For example, the clitic cluster, positioned in second position of the sentence,
forms a template. It includes morphemes related to tense/aspect/mood, and affixes
that indicate the person and number (singular, dual or plural) of the subject and
object, and there are reflexive forms. Examples of argument cross reference mor-
phemes are given in (2). They follow any aspect/mood markers; if none of these
are included, the agreement clitics are attached to the first linguistic element of the
clause, regardless of its function. With a few exceptions, object clitics follow the
subject clitics, and dative objects take precedence over absolutive objects (as in 2f).
Other examples of argument cross reference morphemes are given in (2),
from published sources and my own data. In (2a) the future marker kapu is fol-
lowed by the third person plural subject clitic. There is no overt subject nominal in
(2a) nor in (2b), but the clitic rna in (2b) registers a first person singular subject;
this follows the aspect marker ka. In (2c), the subject clitic rna is attached to the
verb, the first element of the clause, and in (2d) the third person dual subject clitic
pala references the non-overt subject argument. In (2e) the third plural object
clitic follows the third person plural subject morpheme. In (2f) the dative argu-
ment marlu-ku is cross referenced with the third person dative form rla.
 Features of some ergative languages that impact on acquisition 

2. a. kapu-lu nga-rni
fut-3pl.sbj ingest-npst4
‘They will eat.’
b. nya-nyi ka-rna marlu
see-np ipfv-1sg.sbj kangaroo
‘I see a/the kangaroo.’
c. nya-ngu-rna marlu
see-pst-1sg.sbj kangaroo
‘I saw a/the kangaroo.’
d. marlu ka-pala luwa-rnu
kangaroo ipfv-3du.sbj shoot-pst
‘Two (people) shot a/the kangaroo.’
e. luwa-rnu-lu-jana
shoot-pst-3pl.sbj-3pl.obj
They shot them (plural)
f. warru-rnu-rla wati-ngki marlu-ku
look-pst-3sg.dat man-erg kangaroo-dat
‘The/a man searched for the/a kangaroo.’

Because they follow a nominative-accusative pattern, the cross-reference markers

in the clitic cluster are not sufficient to indicate the case frame of the verb. The
child requires other information to categorize which verbs require an A argument.
When a third person nominal is overt, the case markers help identify its role;
an overt argument with ergative case marking indicates the verb has an ergative
case frame. However, given the extensive ellipsis of core arguments, the ergative
case marker is often unavailable. A null case-marked nominal in a clause could
­represent either an object of an ergative verb or a subject of a non-ergative verb.

. The abbreviations used for Warlpiri are:

pl plural
sbj subject
npst non past
sg singular
dat dative
erg ergative
du dual
ipfv imperfective
pst past
ins instrumental
loc locative
obj object
el elative
 Edith L. Bavin

Knowing which verbs are classified with an ergative case frame is essential to dis-
tinguish the nominal’s function.
There are other functions of ergative markers, so ergative case forms may be
in the input even if not marking an A argument. Ergative case forms are used for
instrumental function (as in 3a), but only when the clause has an ergative verb.
They also appear on words associated with an ergative argument of the clause,
whether that arguments is overt or ellipsed. In (3b) wita-ngku ‘small-ERG’ refers
to the dog, not the thing eaten; when the A argument and a modifier/associated
noun are both overt they do not have to be contiguous.5
Ergative case markers can also follow spatial case markers to indicate the loca-
tion of the ergative argument as in (3c) (from Hale 1982). The ergative case form
follows the elative case on pirli ‘hill’ indicating that the ergative argument (wirriya
‘boy’) is at the hill/ throwing from the hill. These additional functions of ergative
case forms provide more information for identifying the ergative verbs.
3. a. wirriya-rlu paka-rnu kurlarda-rlu
boy-erg hit-pst spear-ins
‘The boy hit (something) with a spear.’
b. maliki-rli nga-rnu wita-ngku
dog-erg eat-pst small-erg
‘The small dog ate (something).’
c. wirriya-rlu ka kiji-rni watiya pirli-ngirli-rli
boy-erg npst throw-npst stick hill-el-erg
The boy is throwing a stick from the hill (the boy is on the hill)

Cross sectional data were collected from young Warlpiri children – 11 pairs of
children aged 2–4;8 (see Bavin 2013 for details). They were playing together
outside with adults in the immediate context. Children under two years were
also recorded interacting with another child and with one or more adults.
The data show that young Warlpiri children included ergative case marking
on the subjects arguments of ergative verbs (e.g., pakarni ‘hit’, luwarni ‘shoot’
and ngarni ‘ingest’) by the age of three years. The data do not show overgen-

. Note that at the time of data collection four allomorphs for ergative were used in the
community: the rl/rli forms for words of two or more syllables and the ngku/ngki forms for
one-syllable words. The vowel is conditioned by vowel harmony; following a high front vowel
ngki or rli are used and following a back vowel the ‘u’ is used.
However, some young adults were heard to use the ngku/ngki forms for ergative case
rather than the expected rlu/rli forms, which indicates change in progress. O’Shanassy (2009,
2016) has recently discussed other changes in the form of the language spoken in communi-
ties of Warlpiri people.
 Features of some ergative languages that impact on acquisition 

eralisation of ergative case marking to non ergative verbs. The use of ergative
morphology with agreement or instrumental functions appeared in the child
data by age 4:0 years. Ergative arguments were not evident from two-year olds
since their utterances were very short and included no verb-noun combinations
that require ergative case marking, Their utterances typically comprised kin
terms, deictic words (e.g., nyampu ‘here/this one’), nominals and diminutives.
In answer to ‘where’ questions yali ‘there’ was the frequent response; when talk-
ing about their activities the pronominal clitic was included, as in pakarnu-rna,
hit-1sgg ‘I hit it’.
The restriction of ergative case to verbs with an ergative case frame by age
three provides evidence that the children at that age were aware of different case
frames. An example from a three year old is given in (4); it includes a first person
singular pronoun, ngaju, with ergative case marking. While the case marker form
ngku is appropriate for two syllable words, older Warlpiri speakers (Hale 1982)
would use a long form of ngaju, that is, ngajulu, plus the ergative allomorph for
words longer than two syllables, rlu, giving ngajulu-rlu. However, the use of the
long form was very infrequent in the adult input.
4. luwa-rnu-rna yali yuwarli-rla ngaju-ngku
shoot-past-1sg.sub there/that one house-loc 1sg.sub-erg
‘I shot that one at the house.’ (or ‘I shot (something) there’)

More overt ergative arguments appeared in the older children’s speech as the
number and type of verbs increased, but ellipsis of arguments was common, as
in the input. While the children did use ergative case forms with overt arguments
they did not always use the appropriate allomorph, indicating some uncertainty
about where to use the rlu/rli and ngku/ngki forms and also indicative of change
in progress.
Ergative case marking is a reliable cue for classifying ergative verbs but its
validity as a cue is weakened by the issue of availability. Although ergative argu-
ments, and thus ergative case marking, were not found to be frequent in the input,
semantic cases were frequent; they included locative, allative, elative, and perlative
case. Thus case marking is a salient feature of the language and children will be
attuned to hearing nominal+suffix constructions so when ergative arguments are
overt in the input, we would assume the children’s attention will be drawn to the
N+ergative case construction.
Since categories are formed on the basis of an overlap in shared features, recog-
nising patterns of verb endings could help children categorise those verbs that are
associated with ergative case. There are five morphological verb classes in Warlpiri,
identified by suffixes which differ for tense (past and non-past tense), infinitive,
imperative and irrealis mood, and there is a strong correlation between verb class
 Edith L. Bavin

and case frame. The number of verb roots in the language is not large (Nash 1986
lists about 127), although additional verbs can be derived (see Nash 1982), and a
change of case frame can modify the meaning of a verb. The majority of verbs are
in classes 1 and 2. With very few exceptions class 1 verbs have an absolutive subject
[nonpast = -mi (or zero), past = -ja, imperative = -ya, irrealis = -yarla, and infini-
tive = -nja]. In contrast, with very few exceptions class 2 verbs have an ergative
case frame [nonpast = -rni, past = -rnu, imperative = -ka, irrealis = -karla, and
infinitive = -rninja]. Categories are formed based on co-occurrence of features
and commonalities of the features. (e.g. see Gentner & Namy 2006). Thus, if an
initial association is established between a few verbs in class 2 based on different
endings, and an association is made with co-occurring ergative case marking, by
analogy other verbs with an overlap in suffixes would be categorised similarly. As
an example, if a verb has the same past tense and imperative endings as another
verb and the first verb appeared with an ergative-marked argument, an assump-
tion could be that the other verb could also appear with an ergative-marked argu-
ment, while verbs with different endings for past and imperative would not be
classified as the same type of verb. Verbs in class 3 and 4 are also ergative, each
class with distinct verb suffixes. There is only one verb in class 4 (ngarni ‘ingest’)
and only 7 verbs in class 3 (including the frequent verbs nyanyi ‘perceive’, yinyi
‘give’, kanyi, ‘transport’/‘carry’, and pinyi ‘act on’ /‘attack’). Categorisation of class
4 verbs as ergative could also be made on the same basis of shared morphologi-
cal and co-occurrence features. In summary, the validity of ergative case mark-
ing is affected by the lack of availability, but verb morphology, perhaps, reinforces
other cues as to which verbs have an ergative case frame. Verb semantics has been
discussed extensively in relation to argument structure; however, verb morphol-
ogy might also be a possible source of information to children about categories of
verbs with different case frames.

Mayan languages

Brown et al. (2013) report on the acquisition of four Mayan languages, K’iche’,
Tzeltal, Tzotzil, and Yukatek, all ergative languages spoken in Mexico. In these
languages the verb cluster provides a template with restrictions on the ordering
of morphemes. A brief summary of the acquisition patterns reported by Brown
et al. is included here to illustrate that the position of the argument cross refer-
ence markers in the verbal cluster together with perceptual salience influenced the
order of acquisition and developmental patterns across the languages.
In addition to distinct argument cross reference markers in the verb cluster,
there are morphemes marking aspect and mood and a ‘status suffix’ that encodes
 Features of some ergative languages that impact on acquisition 

transitivity, modality and aspect (Brown et al. 2013: 295). Examples of verb clusters
are given in (5). The aspect marker appears initially in the verb cluster as shown
[Examples 1-4 from Brown et al.] The verb stem is in bold and all examples start
with the incompletive aspect marker. The ergative morpheme is positioned before
the verb root, with distinct forms for consonant initial verb roots and vowel-initial
verb roots. For plural ergative subjects there is a separate plural affix (except in
K’iche’) added at the end of the verb complex. A difference across the languages is
that the absolutive cross referencing marker is a prefix in K’iche’, a suffix in Tzeltal
and Yukatek, and either a prefix or suffix in Tzotzil, depending on the person of
the subject (see 5c). Primary stress is typically on the final syllable. Yukatek dif-
fers from the other three in having a split ergative system, which is conditioned
by aspect.6
5. a. K’iche’
k-at-ki-q’aluu-j
inc-2abs-6erg-hug-status  (Note: 6 erg = 3pl)
‘They hug you.’
b. Tzeltal
ya s-mey-at-ik
inc 3erg-hug-2abs-pl
‘They hug you.’
c. Tzotzil
i. ch-a-s-mey-ik
inc-2abs-3erg-hug-pl
‘They hug you.’  [3rd person plural subject, 2nd person object]
ii. ch-a-mey-ik-on
inc-2erg-hug-pl-1abs
‘You-pl hug me.’  [2nd person plural subject, 1st person object]
d. Yukatek
k-u méek’-ech-o’ob
inc-3erg hug-2abs-3pl
‘They hug you.’

To identify the argument markers and other morphemes in the verbal cluster the
child needs to segment the forms and map functions to each. Longitudinal data
reported from two children from each language shows that the children were

. The abbreviations used are:

inc incompletive
abs absolutive
pl plural
erg ergative
 Edith L. Bavin

r­ apidly attuned to the distinction between transitive and intransitive verbs, and
for Yukatek the children readily identified the aspectual contexts for the split. For
ergative marking, Brown et al. (2013) report that errors were likely to be of omis-
sion; errors of commission (i.e., overgeneralising, double marking and person
confusion) were infrequent.
One of the findings from comparing across the languages was that pre-conso-
nant ergative forms were acquired later than absolutive subject forms for Tzeltal,
Yukatek and K’iche’, but for Tzotzil the reverse was found. A difference between
languages was also reported for the development in use of the absolutive marker;
it was slower for K’iche’ and Tzeltal than the other two languages, suggesting that
suffixes were more salient than prefixes, based on the prosody. That is, as argued
by the authors, the variable position of the absolutive morphemes in Tzotzil seems
to affect the rate at which the children acquired them. Another finding was that
pre-consonant ergative markers forms, which were acquired later than the pre-
verbal ergative markers, took longer to acquire for children acquiring K’iche’ and
Tzeltal. An explanation for the Tzeltal data (p. 290) is that many of the ergative
affixes are non-syllabic and so less perceptually salient. For Yukatek, Tzeltal and
Tzotzil the aspect markers developed along with the pre-consonant ergative mark-
ers (p. 299), and so the single phoneme ergative forms formed a syllable with the
aspect marker. (See Brown et al. 2013, and Pye et al. 2007 for more details of the
development across the four languages).
The data presented by Brown et al. (2013) illustrate the value of comparing
acquisition patterns of related languages. Such comparisons add to our under-
standing of the features of a language that facilitate acquisition. Even though there
are similarities across the languages, differences in morpheme ordering in the ver-
bal complex and consistency in ordering affected the course of acquisition, as did
perceptual salience.

Conclusion

In summary, young children identify the constructions in their language using the
perceptual and cognitive skills they bring to the task, and, as proposed by Bates
and MacWhinney (e.g. 1989), using the valid cues provided in the language input.
For Hindi, perfective aspect co-occurs consistently with the ergative case marker;
children acquiring the language need to link the various perfective markers with
the case marker. That is, verb morphology is one cue to ergative alignment. The
co-occurrence of specific light verbs also provide a cue. In Warlpiri, children need
to acquire knowledge of which verbs have an ergative case frame; the use of erga-
tive case for instrumental function with ergative case-frame verbs, and ergative
case marking on words associated with an A argument (whether the A nominal
 Features of some ergative languages that impact on acquisition 

is overt or not) provide cues to which verbs have ergative case frames. The verb
­morphology for each verb class and co-occurring ergative case marking on argu-
ments of particular verbs could also be of value. In the Mayan languages, the
ordering of morphemes in the verbal cluster seems to provide the strongest cue in
identifying the case system, as shown by acquisition patterns in related languages
when morpheme ordering in the verbal cluster differed; prosody was also found
to influence children’s developing case system consistent with the use of ergative
alignment in their language.
In conclusion, acquiring the constructions in a language depends on identify-
ing linguistic forms and their functions, and the contexts in which the forms are
used. In the current paper examples from three language areas illustrated that chil-
dren draw on the clues available in the input language to detect ergative alignment.
The data reported show that much can be gained from research on the acquisition
of typologically different languages, and also by comparing acquisition in closely
related languages. Children are quickly attuned to systematic features of the input
language and to the co-occurrence of linguistic elements; they provide the cues for
constructing a grammar.

References

Bates, Elizabeth & MacWhinney, Brian. 1989. Functionalism and the Competition Model. In
The Crosslinguistic Study of Sentence Processing, Brian MacWhinney & Elizabeth Bates
(eds), 3–76. Cambridge: CUP.
Bavin, Edith L. 2000. Ellipsis in Warlpiri children’s narratives: An analysis of Warlpiri frog
­stories. Linguistics 38: 569–589. doi: 10.1515/ling.38.3.569
Bavin, Edith L. 2004. Focusing on where. In Relating Events in Narrative: Typological and
­Contextual Perspectives, Vol. 2, Sven Stromqvist & Ludo Verhoeven (eds), 17–35. Mahwah
NJ: Lawrence Erlbaum Associates.
Bavin, Edith L. 2013. The acquisition of ergative case in Warlpiri. In Bavin & Stoll (eds), 107–131.
Bavin, Edith L. & Stoll, Sabine. 2013. The Acquisition of Ergativity [Trends in Language Acquisi-
tion Research 9]. Amsterdam: John Benjamins. doi: 10.1075/tilar.9
Brown, Penelope. 2007. Culture-specific influences on semantic development: Acquiring the
Tzeltal ‘benefactive’ construction. In Learning Indigenous Languages: Child Language
Acquisition in Mesoamerica, Barbara Blaha Pfeiler (ed.), 119–154. Berlin: Mouton de
­Gruyter. doi: 10.1515/9783110923148.119
Brown, Penelope. 2008. Verb specificity and argument realization in Tzeltal child language. In
Crosslinguistic Perspectives on Argument Structure: Implications for Language Acquisition,
Melissa Bowerman & Penelope Brown (eds), 167–189. Mahwah NJ: Lawrence Erlbaum
Associates.
Brown, Penelope., Pfeiler, Barbara, de Léon, Lourdes & Pye, Clifton. 2013. The acquisition of
ergativity in four Mayan languages. In Bavin & Stoll (eds), 271–305.
Bybee, Joan L. 2006. From usage to grammar: The mind’s response to repetition. Language
82(4): 711–733. doi: 10.1353/lan.2006.0186
 Edith L. Bavin

Bybee, Joan L. 2010. Language, Usage and Cognition. Cambridge: CUP.

doi: 10.1017/CBO9780511750526
Bybee, Joan. 2013. Usage-based theory and exemplar representation. In The Oxford Handbook of
Construction Grammar, Thomas Hoffman & Graeme Trousdale (eds), 49–69. Oxford: OUP.
Comrie, Bernard. 2013. Ergativity. Some recurrent themes. In Bavin & Stoll (eds), 15–33.
Gentner, D. & Namy, L. L. 2006. Analogical processes in language learning. Current Directions
in Psychological Science 15, 297–301.
Gertner, Yael, Fisher, Cynthia & Eisengart, Julie. 2006. Learning words and rules: Abstract
knowledge of word order in early sentence comprehension. Psychological Science 17:
684–691. doi: 10.1111/j.1467-9280.2006.01767.x
Goldberg, Adele. 2006. Constructions at Work: The Nature of Generalization in Language.
Oxford: OUP.
Hale, Kenneth L. 1982. Some essential features of Warlpiri verbal clauses. In Papers in Warl-
piri Grammar: In Memory of Lothar Jagst [Work Papers of SIL-AAB: Series A, 6], Stephen
Swartz (ed.). 217–315. Darwin: SIL.
Hoff, Erika. 2006. How social contexts support and shape language development. Developmental
Review 26: 55–88. doi: 10.1016/j.dr.2005.11.002
Laughren, Mary, Hoogenraad, Robert, Hale, Kenneth L. & Granites, Robin Japanangka. 1996.
A Learner’s Guide to Warlpiri. Alice Springs: IAD Press.
Naigles, Letitia R. 1990. Children use syntax to learn verb meaning. Journal of Child Language
17: 357–374. doi: 10.1017/S0305000900013817
Naigles, Letitia R., Bavin, Edith L. & Smith, Melissa A. 2005. Toddlers recognize verbs in novel
situations and sentences. Developmental Science 8: 424–31.
doi: 10.1111/j.1467-7687.2005.00431.x
Narasimhan, Bhuvana. 2005. Splitting the notion of “agent”. Case marking in early Hindi.
­Journal of Child Language 32: 787–803. doi: 10.1017/S0305000905007117
Narasimhan, Bhuvana, Budwig, Nancy & Murty, Lalita. 2005. Argument realization in Hindi
caregiver-child discourse. Journal of Pragmatics 37: 461–495.
doi: 10.1016/j.pragma.2004.01.005
Narasimhan, Bhuvana. 2013. Ergative case-marking in Hindi child-caregiver speech. In Bavin
& Stoll (eds), 209–237.
Nash, David. 1986 Topics in Warlpiri Grammar. New York NY: Garland.
Nash, David. 1982. Warlpiri verbs and preverbs. In Papers in Warlpiri Grammar: In Memory of
Lothar Jagst [Work Papers of SIL-AAB:Series A, 6], Stephen Swartz (ed.), 165–216. Darwin:
SIL.
O’Grady, William. 2015. Frequency effects and processing. Journal of Child Language 42:
294–297. doi: 10.1017/S0305000914000658
O’Shanassy, Carmel. 2009. Language variation and change in a North Australian indigenous
community. In Variation in Indigenous Minority Languages [Impact: Studies in Language
and Society 25], James N. Stanford & Dennis Preston (eds), 419–439. Amsterdam: John
Benjamins. doi: 10.1075/impact.25.21os
O’Shanassy, Carmel. 2016. Distributions of case allomorphy by multilingual children speaking
Warlpiri and Light Warlpiri. Linguistic Variation 16: 68–102.
Pye, Clifton. 2017. The Comparative Method of Language Acquisition Research. Chicago IL: Uni-
versity of Chicago Press.
 Features of some ergative languages that impact on acquisition 

Pye, Clifton, Pfeiler, Barbara, de León, Lourde, Brown, Penelope & Mateo Pedro, Pedro. 2007.
Roots or edges? Explaining variation in children’s early verb forms across five Mayan
­languages. In Learning Indigenous Languages: Child Language Acquisition in Mesoamerica,
Barbara Blaha Pfeiler (ed.), 15–46. Berlin: Mouton de Gruyter.
doi: 10.1515/9783110923148.15
Rumsey, Alan, San Roque, Lila & Schieffelin, Bambi B. 2013. The acquisition of ergative marking
in Kaluli, Ku Waru and Duna (Trans New Guinea). In Bavin & Stoll (eds), 135–182.
Thiessen, E. 2016. What’s statistical about learning? Insights from modelling statistical learning
as a set of memory processes. Philosophical Transactions Royal Society B 372: 20160056.
doi: 10.1098/rstb.2016.0056
Tomasello, Michael. 2003. Constructing a Language: A Usage-Based Theory of Language Acquisi-
tion. Cambridge MA: Harvard University Press.
Tomasello, Michael. 2015. The usage-based theory of language acquisition. In The Cambridge
Handbook of Child Language, 2nd edn, Edith L. Bavin & Letitia R. Naigles (eds). ­Cambridge:
CUP. doi: 10.1017/CBO9781316095829.005
 Constructional pressures on ‘sit’ in Modern
Greek

Soteria Svorou
San José State University

This study argues for a constructional account of the non-postural senses of

Modern Greek verb kaθοme ‘I sit’. Corpus data provide a detailed account of
the constructional pressures on kaθοme as V1 in conjunctive coordination of
verbs (V1 and V2) and show that the non-postural senses emerge in syntactically
compact versions of the string with shared subject and tense/aspect where
locative adverbials and posture and disposition adjectives, characteristic of the
postural interpretations, never occur. In such uses, the morphologically coded
aspectual frame further determines a deliberate action or an extended duration
interpretation framing V2 verbs, typically of mental activity or communication.
Pragmatic inferences tied to the construction provide further motivation for
considering the string a form-meaning composite.

1. Introduction

Syntactic constructions play a major role in delimiting meaning potential and

tiny fluctuations in parameters making up a construction have repercussions for
its eventual interpretation. Meaning does not reside with lexical items alone; it is
constructed in the process of understanding and interpreting language strings.
From the point of view of a language user interpreting a string, for example, “…
he believes there’s a chance she went and …,” might initially prompt a movement
frame for the verb went, until the continuation with “… got pregnant behind his
back just so she could have the baby she wanted,”1 which would reverse the initial
interpretation to a non-motion schematization of went. Similarly, from the point
of view of the analyst, a specific sense that may initially appear to pivot on a certain
lexical item may in fact be due to constructional pressures arising from the given
syntactic arrangement as augmented by semantic and morphological elements and

. Retrieved from <https://www.lovebscott.com/news/ed-hartwell-speaks-says-his-­marriage-

to-keshia-knight-pulliam-crumbled-after-just-two-months-really-isnt-sure-if-that-baby-is-
his-video>

doi 10.1075/slcs.192.02svo
© 2018 John Benjamins Publishing Company
 Soteria Svorou

discourse pragmatic environments. This view on the interaction of syntax, seman-

tics and pragmatics is conceived within a usage-based approach of constructions
and constructionalization (Bybee 2013; Traugott & Trousdale 2014; Bybee 2015).
In order to uncover the nature of constructional pressures, a usage-based approach
focused on specific syntactic strings is advocated, as it allows for the desired close
inspection of context. The present study looks at a specific syntactic arrangement
in Modern Greek, the nexus of two verbs joined by ke ‘and’ represented by the
abstract pattern V1 ke V2. This type of syntactic arrangement has been shown to
facilitate alternative schematizations of V1 semantics in several languages, includ-
ing Modern Greek, with verbs of motion ‘go’ and ‘come’ (Newman & Rice 2008;
Svorou in press) as well as verbs of posture (Kuteva 1999; Newman 2002; Newman
& Rice 2004; Lemmens 2005.) The focus will be on a type of this syntactic arrang-
ment in which V1 is the lemma kaθοme ‘I sit’. Among other Modern Greek posture
verbs, ksaplono ‘I lie’, stekome ‘I stand’ and kremome ‘I hang’, kaθοme ‘I sit’ is the
only one that is found in this syntactic pattern with a non-postural interpretation
and, thus, is well-suited to shed light on how contextual factors interact with syn-
tactic arrangements to give rise to complex interpretations that justify the emer-
gence of constructions. As I will argue, Vkaθοme ke V2 represents a construction
that is reinforced by exemplars that have particular morphosyntactic and semantic
characteristics and tend to occur in discourse contexts that communicate speaker
attitude, such as surprise, disbelief, or criticism. The same pattern with pijeno ‘I go’
and erxome ‘I come’ as V1 is also found with non-motion interpretations in simi-
lar discourse contexts and tends to convey similar speaker attitude but the lexical
semantics of V1 color the interpretation accordingly (Svorou in press).
Such findings support a proposal for an abstract constructional schema V1 ke
V2 with a shared subject based on a network of less abstract schemas ([Vkaθοme ke
V2], [ Vpijeno ke V2], [Verxome ke V2 ]) formed by the interpretation of specific exem-
plars in language use.
The construction under investigation is illustrated in examples (1–2).2
(1) mesa stin trela=mas kaθom-astan ke pistev-ame
in in.art madness=poss sit-ipfv.pst.1pl and believe-ipfv.pst.1pl
kaθe ðiataraɣmeno
every deranged
“In our madness/fury, we kept on believing (lit. were sitting and believing)
every deranged person.”  (CMG)

. Modern Greek displays case marking morphology and agreement of determiners and
adjectives with the head noun within noun phrases. Nominative, accusative, and genitive cases
are distinguished. Case marking is not glossed in these examples for the sake of brevity.
 Constructional pressures on ‘sit’ in Modern Greek 

(2) … i ðiorɣanotes kaθ-isan ke vri-kan oles tis

art organizers sit-pst.pfv.3pl and find-pst.pfv.3pl all art
aðinamies ke ta laθi ton perasmenon eton…
weaknesses and art mistakes art past years.gen
‘… the organizers, acting in a determined manner, found (lit. sat and found)
all the weaknesses and the mistakes of past years…’  (HNC)
Situations expressed by V2 in these examples, believing and finding, are not situa-
tions that are normally carried out either following or simultaneously to sitting and
hence they invite an interpretation of kaθome ‘I sit’ other than ‘entering/being in
a seated position’, the prototypical senses of this verb. Of the two conjoined verbs,
it is V2 that contributes information about ‘what happened’; V1 kaθome reflects
the speaker’s view on the V2 situation being semantically schematized into a rich
aspectual frame through which V2 is interpreted as having lasted for a period of
time longer than expected for the type of V2 situation with additional nuances of
persistence or tolerance attributed to the agent of V2.
Similar schematic uses of posture verbs have been documented for several
languages. The existing body of research discusses them as sources of grammatical
markers or auxiliaries of progressive/continuous aspect (Bybee, Perkins & Pagliuca
1994; Kuteva 1999; Heine & Kuteva 2002; Newman 2002; Bybee 2015.) Corpus
studies have further documented quantitatively the existence of correlating factors
and patterns that contribute to the emerging grammatical status of the Vsit/stand/lie
and V2 as a construction, characterizing it as ‘pseudo-coordination’ (Lødrup 2002;
De Vos 2005; Wiklund 2007; Hilpert & Koops 2009; Viberg 2013; Ross 2016) or
‘auxiliated cardinal posture verb construction’ (Newman & Rice 2004 – English;
Lemmens 2005 – Dutch).
Against this background, the present study attempts to provide an account of
the construction. No prior such account exists for Modern Greek in the linguistics
literature or in grammars. The study adopts methodology consistent with usage-
based theory by analyzing actual language strings from corpora. The goal is to cap-
ture empirically the interaction of a number of morphological, syntactic, semantic
and pragmatic parameters that lead to the ultimate construction of meaning and
to show how such interactions may account for alternative interpretations of con-
structions thus providing the fine adjustments that lead to the emergence and con-
ventionalization of constructions.

2. The semantics and morphology of kaθοme ‘I sit’

Before proceeding with an account of the posture verb in the coordinate construc-
tions, an analysis of kaθome itself is needed. The verb kaθοme ‘I sit’, a medieval
 Soteria Svorou

development from ancient kaθemai, is polysemous. Τριανταφυλλίδης (1998)

offers the following senses (translation mine), which can be considered to be an
account of the verb’s meaning potential (Croft & Cruise 2004: 109.)

1. a. I lean somewhere with my buttocks, keeping the back straight and the
legs bent or straight depending on the height of the support in relation to
the ground or the floor.
b. I exist in a place seated somewhere
2. a. I live, I stay
b. I pass a relatively long temporal period somewhere
3.a. I stay inactive, I do not engage with something; I do not take initiative for
something
b. 1. I engage with something; to stress the aspect of the following verb
2. I accept, I tolerate
3. I wait, I am patient
4. a. for something that sinks, that recedes and falls || for something that deflates
b. for a ship that is driven ashore or that sinks more than needed because of
a big load.
5. for reduced outcome (oral, sports)

Of interest in this exploration, because of their relevance to the syntactic arrange-

ment under investigation, are senses 1a–1b and 3a–3b3. Specific aspects of a word’s
meaning potential may surface when triggered by contextual factors. Among them,
adverbials (locative, temporal, manner) and predicate adjectives play an important
role, as is shown in Section 5.5. The semantic integrity of kaθome depends on, and
is sensitive to, the amount of information disclosed and how much of its valence is
made explicit. Occupying a space in a seated position, ‘sit’, can be further explicated
by expressing a specific posture of the entity, specific location and time, as well as
elements of the psychological condition of the entity. The more valence elements
remain unspecified the more the semantic integrity of the verb is eroded. Depend-
ing on the morphologically expressed aspectual frame, imperfective or perfective,
the verb kaθοme is interpreted as an activity (‘be in a seated position’) with the
former, or as an achievement (‘enter into a seated position’) with the latter. The
aspectual frame also affects the interpretation of the Vkaθοme ke V2, as seen in (3).
(3) a. kaθisa ke efaga  sequential states of affairs
sit.pst.pfv.1sg and eat.pst.pfv.1sg
‘I sat and ate’
b. kaθomun ke etroga  simultaneous activities
sit.pst.ipfv.1sg and eat.pst.ipfv.1sg
‘I was sitting and eating’
 Constructional pressures on ‘sit’ in Modern Greek 

Semantically, the verb kaθοme instantiates two different experiential frames with
different sets of entailments: a posture frame and a change of posture frame. These
entailments account for its polysemy to a certain extent. In the posture frame,
kaθοme has the following entailments (Numbers in parentheses correspond to the
dictionary senses above):

–– Being seated entails occupying space and, therefore, being or existing (1b)
–– Being in a seated position generally occupies a significant temporal period,
i.e., it lasts for a while. (2a, 2b)
–– The seated individual is likely to engage in actions that normally take place in
a seated position and do not involve much physical movement.
–– Being in a seated position involves less visible action which implies lack of
activity on the part of the seated individual. (3a)
–– Inactivity in the seated position is metaphorized to inactivity in terms of out-
come. (5)
–– Being in a seated position, generally inactive, for a period of time is associated
with waiting, being patient, tolerating, or submitting to some other situation.
(3b2, 3b3)

In the change of posture frame, kaθοme has a different set of entailments:

–– Entering a seated position implies moving downwards in the direction of

gravity (4a–b)
–– Entering a seated position implies (and presupposes) agency, persistence, and
purposeful action.

These entailments also figure in how sitting verbs are interpreted in the coordinate
construction.
In terms of morphology, kaθome exhibits great complexity, given the highly
inflectional nature of Modern Greek, reflecting person/number, tense/aspect
(non-past/present, past, perfective, imperfective), and mood (realis and irrealis)
combinations. The future is expressed periphrastically with a preposed particle
θa (e.g., θa kaθome ‘I will be sitting’) and the irrealis (subjunctive) similarly with
a preposed particle na (e.g., prepi na kaθome ‘I must be sitting’). Appendix 1 con-
tains the kaθοme paradigm.

3. Coordination and pseudo-coordination

Conjunctive coordination as an interclausal relation presupposes the indepen-

dence of the coordinated clauses, both syntactically and semantically, and as such
 Soteria Svorou

is distinct from subordination in which one clause is dependent on the other

(Van Valin 2007: 82). The coordinated clauses are independent to a certain degree
because, as Langacker (2012) explains, in order for events expressed in clauses to
be conjoined and thus conceptualized as a whole, they would have to be inter-
preted to be analogous in some ways, the essence of what coordination requires.
In interpreting conjunctive coordinate utterances, the addressee would have to
construct a complex frame bringing together the events encoded by the coordi-
nated clauses so that they can be interpreted as analogous. Possible frames would
have to incorporate (i) an experientially-grounded estimate about the likelihood
of two events being related and (ii) the distribution of events on a timeline (Croft
2001). In this timeline, events may be simultaneous or in a temporal sequence;
clauses expressing temporally related events would form simultaneous or con-
secutive chains, respectively. It is also possible that the analogical basis of events
is not temporal order but conceptual aggregation, with the clauses expressing
them forming unordered collections.
In the type of conjunctive coordination construction exemplified in (1–2),
the analogical basis of the conjoined events consists of a shared agent partici-
pant and a shared tense/aspect frame. But there is further integration of the
two events in that V1 kaθome does not express a separate event but offers an
additional frame to conceptualize V2. Similar constructions in other languages
(e.g., Swedish, Norwegian, Afrikaans, English) have been described as examples
of pseudo-coordination, with the character of the coordination construction
varying according to the type of V1 (‘go’, ‘come’, ‘sit’, ‘try’, etc.) (Lødrup 2002; De
Vos 2005; Wiklund 2007; Hilpert & Koops 2009; Ross 2016). The term ‘pseudo-
coordination’, stemming from a tradition of formal linguistics, was initially
introduced to cover any coordinate structure which related events in temporal
or aspectual terms, contrasting them with simple coordination where no such
inferences are drawn. But even in later studies there seems to be lack of sig-
nificant discussion of further semantic differentiation of the resulting interpre-
tations. Constructions do not always have well-defined properties, leading to
ambiguity in their interpretation. For example, the string kaθete ke ɣrafi ‘sits and
writes’ could be interpreted as making reference to two simultaneous states of
affairs in which the shared participant is writing while in a seated position, the
most common posture while writing. But in (4), it is more likely that it would
be interpreted as an example of a pseudo-coordination construction in which
kaθete forms a frame of deliberate and persistent engagement for V2. This inter-
pretation is accentuated, first, by the assertion that he wrote with a pseudonym,
which shifts the profile from the writing process per se to the use of the pseud-
onym as a writer, second, by the nature of what he wrote – a novel, which cannot
be completed in one sitting – and third, by the statement of Ellison’s emotional
state, eksorjismenos ke apelpismenos ‘furious and desperate,’ as it provides a
 Constructional pressures on ‘sit’ in Modern Greek 

r­ eason for his deliberate and persistent action. (The non-past imperfective forms
are interpreted as historical presents)
(4) eksorjismenos ke apelpismenos, o elison kaθ-ete
furious and desperate art E. sit-npst.ipfv.3sg
ke ɣraf-i me psevðonimo ena miθistorima
and write-npst.ipfv.3sg with pseudonym a novel
pu paro-ði to “politikos orθo” best seler mjas
that echo.npst.ipfv.3sg art politically correct best seller art.gen
afroamerikanis sinaðelfu=tu
Afro-American.gen colleague.gen=poss
‘Furious and desperate, Ellison deliberately writes (lit. sits and writes) with
a pseudonym a novel that echoes the “politically correct” best-seller of his
Afro-American colleague.’  (HNC)
Because of such ambiguities, there is a need to explore an expanded set of exam-
ples with the goal of providing a more nuanced description of the construction
and its competing interpretations and also to provide an account of the contextual
cues that determine such a non-prototypical interpretation; because it is context,
and context only, that affects how this verb in this construction is interpreted.
A preliminary investigation of examples of Vkaθome ke V2 led to distinguishing
five types in terms of how kaθome is interpreted, posturally or non-posturally.
The three postural types differ in terms of how the two events are distributed on a
timeline. The two non-postural types differ in terms of how the sitting experiential
frame is schematized according to the aspect of the schema that is profiled.

a. Postural in Unordered Collection: kaθome is among a set of verbs in a con-

junctive coordinate structure with a shared agent participant; the expressed
events are conceptualized within the same time/space but they are not in a
temporal sequence and their order could be reversed without loss of meaning.
The verb kaθome is interpreted as a posture verb.
(5) sto sxima pu sas ðix-no parusiaz-onde
in.art chart that iobj.2pl show-npst.ipfv.1sg appear-npst.ipfv.3pl
parastatika i ðiafores anamesa s enan aðinato kj enan
schematically art differences between in one slim and one
paxisarko (sto xrono pu kaθ-onde ke kin-unde
obese in.art time that sit-npst.ipfv.3pl and move-npst.ipfv.3pl
ke i andistixes apolies se θermiðes).
and art corresponding losses in calories
‘In the chart I’m showing you, there appear schematically the differences
between a slim and an obese individual (the time they sit and they move and
the corresponding loss in calories.)’  (CMG)
 Soteria Svorou

b. Postural Sequential: kaθome is V1 in a coordinate structure that conjoins

temporally sequenced events with a shared agent participant. It is interpreted
as a change of posture verb and precedes the V2 event. Their order could not
be reversed and there is an implication that the sitting posture may extend at
least for the period of the V2 event.
(6) kaθ-isame ke sinex-isame na pin-ume
sit-pst.pfv.1pl and continue-pst.pfv.1pl spart drink-ipfv.1pl
‘We sat and continued to drink.’  (HNC)

c. Postural Simultaneous: kaθome is V1 in a coordinate structure that conjoins

co-temporal events with a shared agent participant. It is interpreted as a sitting
verb. Attention is directed to the simultaneity of the two events, not to their
initiation, although there is an implication that the sitting event started prior
to that of V2, as the order of the two verbs could not be reversed.
(7) sto vaθos tu maɣazju, s ena strongilo siðerenjo trapezaki,
in.art depth art shop.gen, at a round iron table,
kaθ-otan ki epin-e ɣulja ɣulja to faskomilo=tis
sit-pst.ipfv.3sg and drink-pst.ipfv.3sg sip sip art sage.tea=poss
mja ɣria.
art old.woman
‘In the far end of the shop, on a round iron table, an old woman was sitting
and drinking her sage tea sip by sip.’  (HNC)

d. Deliberate: kaθome does not express posture but it attributes additional per-
sistence, purposefulness, and patience on the part of the agent participant to
carry out the V2 event.
(8) bravo telika stus poðosferistes=mas, an ondos ekat-san
bravo finally to.art footballers=poss.3pl if indeed sit-pst.pfv.3pl
ke exa-san apo ti larisa jati tus
and lose-pst.pfv.3pl from art Larisa because obj.3pl
parakal-usan i larisei!
beg-pst.ipfv.3pl art Lariseians
‘Bravo finally to our footballers, if indeed they deliberately lost from Larisa
because Lariseians were begging them!’  (HNC) (said sarcastically)

e. Extended: kaθome does not express posture; it attributes additional extension

to the duration of the V2 event.
(9) ores-ores eɣo kaθ-ome ke anarot-jeme
hours-hours I sit-npst.ipfv.1sg and wonder-npst.ipfv.1sg
‘Some times I keep wondering.’  (HNC)
 Constructional pressures on ‘sit’ in Modern Greek 

The types differ in terms of semantic integration, with the extended and deliber-
ate types being the most integrated. This semantic integration must be accom-
panied by syntactic integration. Syntactic integration is achieved when valence
requirements of conjuncts are relaxed, the order of the conjuncts becomes fixed,
and possibilities of intervening elements are reduced or eliminated. Syntactic and
semantic integration and word order fixing in some tokens of an arrangement but
not others are indications of the existence of a separate construction. Construc-
tions emerge through constructionalization, a diachronic process that consists
in spreading throughout the language (and from speaker to speaker) of a new
form-meaning composite that is the result of micro-changes in speakers’ utter-
ances (Noël 2007; Traugott & Trousdale 2014; Bybee 2015). The form-meaning
composite, a construction, is characterized by a certain syntactic schema whose
individual components are held tightly together semantically and syntactically,
exhibiting a lesser degree of compositionality. New constructions splinter away
from existing constructions by fixing certain tokens that fill syntactic slots of the
arrangement and by incorporating pragmatic inferences that may accompany
such tokens. Tracing the history of a construction and finding its sources presup-
poses detemining the existence of the construction at some synchronic point.
This is what the present study is about. My goal is to show that the syntactic string
consisting of lemma kaθome (V1) followed by the multi-functional coordinating
conjunction ke or its phonological alternates ki and kj, followed by a verb (V2)
and its object(s), with a subject shared by both verbs, is a construction, and to
analyze its polysemy accounting for the parameters and contextual cues that con-
tribute to its multiple interpretations.

4. Why corpus methodology?

This study is about establishing a syntactic construction and close exploration of

context is crucial. Many aspects of context, some suspected others discovered in
the process of study, contribute to the ultimate interpretation of a given example.
Corpus exploration, as contrasted to introspection, allows for the uncovering of
patterns of contextual factors and co-occurrences.
The initial exploration of the construction was carried out using the Hel-
lenic National Corpus (HNC) (46 million words), which yielded 286 relevant
concordances, but in order to expand the number of examples and for the
sake of comparison, I have also considered two other corpora that are read-
ily and freely available, Corpus of Modern Greek (CMG) (36 million words)
and Corpus of Greek Texts (CGT) (30 million words). I have used the same
query instructions in HNC and CMG corpora: lemma kaθοme ‘I sit’ followed
by lemma ke ‘and’, followed by a verb, with an allowance for up to five words to
 Soteria Svorou

intervene between kaθοme and ke and between ke and V2. For the third corpus,
CGT, the same query instructions were used but there was an allowance of one
word between kaθοme and ke and up to five words between ke and V2. Table 1
presents the tokens retrieved from each corpus. Because of the differences in
corpus tagging and interface, it is not possible to delimit in all three corpora
the part-of-speech of the form that follows the conjunction ke. As a result, of
the total of 953 strings retrieved, 253 were identified manually as irrelevant and
were excluded from the study.3

Table 1. Tokens retrieved by lemma kaθοme {0–4 words} ke {0–4 words} POS: VERB
HNC CGT CMG Total

Kaθome ke V2 286 159 255 700

Irrelevant data 44 43 166 253
Total 330 202 421 953

The 700 tokens of relevant data were each coded for the following parameters that
were deemed relevant to accounting for the interpretation of the construction:

–– Semantic type of conjunctive coordination

–– Intervening elements between V1 and ke
–– Intervening elements between ke and V2
–– Subject person/number
–– Tense (npst, pst)
–– V1 aspect (ipfv, pfv)
–– V2 aspect (ipfv, pfv)
–– V2 and its semantic frame(s)
–– Sentence type containing string (decl, interrog, imper)

5. Contextual cues

In this section, each of the coded parameters is discussed and correlations between
various parameters are brought forth as guided by descriptions of the corpus data.
For the issues discussed here and because of the general similarities among the three

. Among those were strings in which ke was not joining verbs or verb phrases, or cases in
which there was conjunction but the two verbs had different subjects, or there were two verbs
conjoined but they were not part of the same constituent.
 Constructional pressures on ‘sit’ in Modern Greek 

corpora, the data is reported in totals with no specific results given for each corpus,
with the exception of Table 2. (Details per corpus are available upon request.)

5.1 Semantic types of coordination

With the exception of data from CMG that showed a predominance of simul-
taneous type of coordination (63%), in the other two corpora the extended type
is in the majority (HNC: 36%; CGT 63%). If the extended is calculated together
with the deliberate type, as both involve a non-posture interpretation, the pattern
becomes even more obvious (HNC 58%; CGT 71%), at least in these two corpora.
A possible explanation for the lower percentage of non-posture data strings in the
CMG may be its nature as it is exclusively based on written texts, while the other
two are balanced between oral and written texts.
Another, perhaps surprising, finding is the extremely low occurrence (1%) of
the unordered collection type of coordination in all three corpora. The few tokens
all involve lists of verbs of which kaθome is the first.

Table 2. Types of coordination in the three corpora

HNC CGT CMG Total

Non-postural Extended 103 (36%) 100 (63%) 86 (34%) 289 (41%)

Deliberate 63 (22%) 13 (8%) 76 (11%)
Postural Simultaneous 43 (15%) 20 (13%) 161 (63%) 224 (32%)
Sequential 75 (26%) 24 (15%)   4 (1%) 104 (15%)
Unordered   2 (1%)   2 (1%)   4 (1%)   8 (1%)
Collection
TOTAL 286       159      255      700 (100%)

5.2 Tense and aspect

The Vkaθome ke V2 string type shows an overwhelming tendency for shared tense/
aspect between the two verbs. As Table 3 shows, only 3% (27) of the strings
in the three corpora showed lack of agreement in tense and aspect between
kaθοme and V2. In these 27 strings, kaθοme occurred in the perfective aspect
and V2 in the imperfective and all were interpreted as sequential events. But the
rest of 673 strings all showed agreement in tense and aspect between the two
verbs. Of those, 76% occurred in various imperfective forms, showing a pat-
tern in agreement with the activity reading of kaθome. The rest were in various
perfective forms.
 Soteria Svorou

Table 3. Distribution by tense and aspect and V1 and V2 agreement

Non-Past Imperfective 316
Past Imperfective 112
Future Imperfective   5
na V-imperfective 96
TOTAL IMPERF 529 (76%)
Tense/Aspect Agreement between V1 and V2
Past Perfective 87
Future Perfective   2
na V-Perfective 47
Imperative (NPast Perf)   8
TOTAL PERF 144 (21%)
Total AGR 673
V1 Past Perf 20
V2 Past Imperf
V1 Past Perf   1
No Tense/Aspect Agreement between V1 and V2
V2 Non-Past Imperf
V1 na Perf   6
V2 na Imperf
Total NoAGR 27 (3%)
TOTAL 700

5.3  Semantic types and tense/aspect correlation

In considering type of conjunctive coordination and aspect, there are very signifi-
cant correlations, as can be seen in Table 4.

Table 4. Distribution of coordination types in terms of aspectual character of V1 and V2

Coordination type Imperf Perf Perf-Imp

Non-postural Extended 289 (100%)

Deliberate   2 (3%) 75 (97%)
Postural Simultaneous 223 (100%)
Sequential   9 (10%) 70 (66%) 25 (24%)
Collection   6 (75%)   2 (25%)
SUBTOTALS 529 146 25
TOTAL 700

The extended type is exclusively expressed with imperfective aspectual forms in

main and subordinate clauses (as in 10) in which kaθome and V2 agree.
 Constructional pressures on ‘sit’ in Modern Greek 

(10) ke i politia, andi na ka-θete ke na

and art state, instead spart sit-npst.ipfv.3sg and spart
aku-i tus kinðinoloɣus, θa eprepe na
listen-npst.ipfv.3sg art scaremongering fpart should spart
kan-i oti bori ja na.spart
do-npst.ipfv.3sg whatever be.able.npst.ipfv.3sg in.order.to
li-si ta provlimata pu iparx-un sta
solve-npst.pfv.3sg art problems that exist-npst.ipfv.3pl in.art
nisja ke na proselki-sume episkeptes.
islands and spart attract-npst.pfv.1pl visitors.
‘And the state, instead of continuing to listen to the scaremongering
individuals, it should have done whatever it could to solve the problems that
exist in the islands and to attract visitors.’  (HNC)

(11) sto saloni kaθ-onde ke sizit-un o

in.art living.room sit-npst.ipfv.3pl and discuss-npst.ipfv.3pl art
pateras, i kori kj o ðavið
father art daughter and art D.
‘In the livingroom, the father, the daughter, and David are sitting and
discussing.  (CGT)

The same applies to the simultaneous type in all three corpora. The two types differ
in the interpretation of kaθome: in simultaneous, the subject is in a seated position
and simultaneously engaging in the event of V2 (as in 11); in the extended the sub-
ject’s involvement with the V2 event is viewed as protracted and extensive while
posture is not at issue, something very clear in (10) with a subject such as i politia
‘the state’, a collective noun that brings to mind no physical form.
The great majority of the deliberate type tokens are correlated with the past
tense and perfective aspect, although it does occur in subordinate clauses, as in
(12). In examples of this type, the subject is involved with the V2 event in a deter-
mined and deliberate fashion and again the posture sense of kaθome is missing.
In the two examples with the non-past imperfective, the forms are interpreted as
historical presents, which is tantamount to a past perfective but with the discourse
effect of immediacy, as can be seen in (13).

(12) anarotj-eme mipos itan laθos na kats-o

wonder-npst.ipfv.1sg INTEROG was mistake spart sit-npst.pfv.1sg
ke na mira-so ta vivlia=mu se tosa mutra
and spart distribute.npst.pfv.1sg art books=poss to so.many guys
‘I wonder whether it was a mistake to deliberately distribute my books to so
many guys.’  (HNC)
 Soteria Svorou

(13) kj erx-ete i stiɣmi pu kaθ-ete ke

and come-npst.ipfv.3sg art moment that sit-npst.ipfv.3sg and
ɣra-fi se ðisko afti ti musiki pu
write-npst.ipfv.3sg in disk this art music that
ɣraf-tike apo enan anθropo sto telos
write-pass.npst.ipfv.3sg from one man at.art end
tis zois=tu
art life=poss
‘And there comes a moment when he deliberately records in a disk the music
that was written by a man at the end of his life.’  (HNC)

The sequential type is also correlated with the perfective aspect, although here
there is more complexity as in 25% of the tokens kaθome does not agree with V2
in aspect. In these, kaθome is in perfective form, thus viewing a complete sitting
event in a holistic manner, and is followed by the imperfective form of V2, as in
(14). In another 10% of the sequential tokens, both verbs are in the imperfective
form but the interpretation is that of a habitual (15), which leads to a posture verb
interpretation of kaθome framed as completed.
(14) kaθ-isa ke etimaz-omun na val-o ena
sit-pst.pfv.1sg and prepare-pst.ipfv.1sg spart put-pfv.1sg one
flitzani kafe otan paratir-isa oti olo to patoma itan
cup coffee when notice-pst.pfv.1sg that all art floor was
ðiasparto apo skitsa
scattered from sketches
‘I sat and was getting ready to pour a cup of coffee when I noticed that the
floor was scattered with sketches.’
(15) i parapano times isxi-un mono ja aftus pu
art above prices hold-npst.ipfv.3pl only for those that
servir-onde orθii sto kilikio ke oxi ja osus
serve-pass.npst.ipfv.3pl standing at.art bar and not for those
kaθ-onde ke ðin-un parangelia se servitoro
sit-npst.ipfv.3pl and give-npst.ipfv.3pl order to.art server
‘The above prices are valid only for those who are served standing at the bar
and not for those who sit and give an order to a server.’

The remaining 66% of the sequential tokens show agreement of the two verbs,
which occur in the perfective aspectual forms. The perfective aspectual frame
clearly distinguishes two events, one of entering into a seated position fol-
lowed by another event, both of which are viewed perfectively. It is interesting
that half of the tokens are in the past tense and the other half appear in the na
 Constructional pressures on ‘sit’ in Modern Greek 

V-perf subordinate form in which the whole Vkaθome ke V2 string is in a subor-

dinate clause.

5.4 Syntactic compactness of the string

In conducting searches in the three corpora, allowance was made for the occur-
rence of up to five words between the conjunction ke and V2. Generally, the types
of syntactic items that can occur before verbs include particles (the future θa, the
subordinating na and the optative as), the perfect auxiliary exo ‘I have’, negation
particles min and ðen, and object clitics which, if present, obligatorily precede the
head verb. In ditransitive verbs with two object clitics both precede the verb with
the theme/direct object clitic being closer to the verb. In strings with the future
particle, θa precedes the object clitics. Negative particles tend to occur before
object clitics and the future particle (as in 16). With subordinating particles na and
as, the order of these elements is then na /as – neg – indirect object clitic – direct
object clitic – verb, (as in 17.)
(16) ðen θa tus ta ixame dosi
neg fut iobj.3pl dobj.3pl have.pst.1pl give.pfv
‘We would not have given them to them.’
(17) eprepe na min tus ta ixame dosi
should spart neg iobj.3pl dobj.3pl have.pst.1pl give.pfv
‘We should have not given them to them.’

Additionally, because the word order is relatively flexible, other syntactic catego-
ries may appear before a verb, such as adverbials but also subjects and objects.
In the CMG and HNC searches, there was also allowance for up to five words to
intervene between V1 and ke to capture potential examples with adverbials that
modify V1 that may appear in that position.
The results are presented in Table 5 in the form of structural traces found in the
corpora. A structural trace is an occurrence of the basic V1 ke V2 schema in actual
text.4 It represents an abstraction from specific exemplars encountered in language
in use. The schema may be augmented with syntactic units showing a variety of
dependencies. Structural traces are organized according to whether there are any
intervening dependent items or not, the periphrastic nature of V1 and V2 (particles
na, θa and as) and whether items occur between ke and V2 (ke __ V2) or between
V1 kaθome and ke (V1 __ ke).

. In using the term ‘structural trace’, I am making an analogy to the Hebbian structural
memory trace in the brain as the foundation of long term memory formation (Hebb 1949).
 Soteria Svorou

Table 5. Distribution of structural traces of Vkaθome __ ke __ V2 in three corpora

Structural traces Total

No intervening V1 ke V2 325 (46%)

items
Periphrastic θa V1 ke θa V2   4
verb forms
as V1 ke as V2   1
na V1 ke na V2 62
Subtotals 67 (10%)
ke __ V2 V1 ke___V2 V1 ke iobj_cl V2   1
dobj_cl
V1 ke dobj_cl V2 73
V1 ke dobj_NP V2   4
V1 ke neg V2   8
V1 ke temp V2   1
V1 ke mann V2   1
Subtotals 88
θa V1 ke θa θa V1 ke θa dobj_cl V2   2
__ V2
na V1 ke na na V1 ke na neg V2   1
__ V2
na V1 ke na dobj_cl V2   6
Subtotals   9
Subtotal 97 (14%)
V1 ___ ke V1 ___ ke V2 V1 loc ke V2 80
V1 mann ke V2   3
V1 temp ke V2   9
V1 temp loc ke V2   3
V1 loc mann ke V2   6
V1 predadj ke V2 15
V1 predadj ke pfvaux V2   3
V1subj ke V2   6
V1 V2 ke V3   4
Subtotals 129
θa V1 __ ke θa V1 loc ke neg θa V2   1
θa V2 obj_cl
θa V1 loc ke θa V2   1
 Constructional pressures on ‘sit’ in Modern Greek 

Table 5. (Continued)

Structural traces Total

na V1 ___ ke na V1 predadj ke na V2   4
na V2 na V1 predadj ke na neg V2   1
na V1 predadj ke na obj_cl V2   1
na V1 temp ke na V2   5
na V1 temp ke na neg V2   1
na V1 loc ke na V2 57
na V1 temp ke na neg V2   2
na V1 temp ke na obj_cl V2   2
na V1 predadj ke na V2   1
loc
na V1 temp ke na V2   2
loc
na V1 predadj ke na V2   2
na V1 mann ke na V2   2
Subtotals 82
Subtotal 211 (30%)
TOTAL 700

Of the 700 tokens from all corpora, 46% exhibit the structural trace V1 ke V2 with
no intervening dependent items.5 The V1 ke __ V2 type of structural trace occurred
in 14% of the overall number of tokens.6 The most common intervening items are
object clitics of V2 followed by negative particles and locative, temporal or manner
adverbials. The V1 __ ke V2 type of structural trace occurred overall in 30% of the
tokens.7 The most common intervening classes of items are locative adverbials,
predicate adjectives of disposition, temporal and manner adverbials, and subject
NPs. There are also a few tokens with another verb intervening between V1 __ ke
V2, expressing a series of events.

. The three corpora differ, however, in how frequenct this trace occurs with CGT showing
the highest, 88%, followed by HNC at 57% and CMG at 23%.
. This trace occurred in different percentages in each corpus: HNC 9%; CGT 26%;
CMG 11%.
. Of the three corpora, the CMG showed the greatest frequency (57%) followed by HNC
(22%). The complete lack of tokens of this type in CGT is due to the corpus interface, which
does not allow wildcards in the search string.
 Soteria Svorou

Not all intervening items are created equal, making it crucial to distinguish
between items, on the one hand, in the semantic valence of the verbs and, on
the other, adjuncts. Of the intervening items in the V1 ke __ V2 type, object clit-
ics and object NPs are core valence items, and indeed account for 86 of the 97
overall tokens (89%) with similar correspondences in each corpus, and of those
82 are clitics.
Among the V1 ___ ke V2 structural trace tokens, the occurrence of peripheral
locative adverbials (prepositional phrases and adverbs) in 145 out of 211 was the
highest (69%). Expression of location, although debatable whether it is part of the
semantic valence of kaθome, is generally expected in cases where kaθome func-
tions as a posture or change of posture verb. Of the rest of the intervening items,
predicate adjectives of posture (e.g., skiftos ‘bent’, akinitos ‘immobile’, angaljasmena
‘embraced’, ðiploɣonatos ‘cross-legged’, ksaplomeni ‘lying’) and disposition (e.g.,
melanxolikos ‘melancholic’, amerimnos ‘carefree’, amilitos ‘silent’, siopilos ‘silent’)
were the most frequent, all of which are justified as they relate to a general experi-
ential frame of posture and sitting in particular. These findings beg the question of
whether such tokens are interpreted as simultaneous or sequential semantic types,
which sanction a posture or change of posture interpretation, and, in fact, leads us
to the next section to the discussion of how the five semantic types fare in terms of
the structural traces that express them.

5.5 Semantic types and their structural traces

The hypothesis that guided this investigation is that each semantic type distin-
guished has characteristic patterns of structural traces that motivate the meaning
and bond the construction. In the discussion that follows, I will show that, for the
interpretation of a given Vkaθome ke V2 token, various co-occurring items in addi-
tion to the lexical semantics of V2 and of kaθome are relevant.

Extended
Table 6 presents the structural traces of the extended semantic type. The extended
semantic type is the most frequent altogether. By far the most common structural
trace is V1 ke V2, with no intervening items in main and subordinate structures
with particle na, accounting for 242 tokens. Of the other structural traces, object
clitics and pronouns are found to intervene between ke and V2 in 33 tokens.
There are 5 tokens with the shared subject of the conjoined verbs occurring
between kaθome and ke, which reflects particular discourse manipulations. As an
example, in (18), the shared subject i paljanθropi ‘the skunks (of people)’ is post-
posed after the first verb because of the fronting of ti ‘what’, an argument of lene
‘tell’. It could also appear after lene at the end of this sentence.
 Constructional pressures on ‘sit’ in Modern Greek 

Table 6. Structural traces of extended semantic type

Structural trace Total

V1 ke V2 220 242
na V1 ke na V2 22
V1 ke obj_cl V2 28 33
V1 ke obj_PRO V2   1
na V1 ke na obj_cl V2   1
V1 temp ke obj_cl V2   2
V1subj ke obj_cl V2   1
V1subj ke V2   4
V1 ke mann V2   1
V1 temp ke V2   7
na V1 temp ke na V2   1
V1 loc ke V2   1
Total 289 (41%)

(18) min aku-te, kutavakja, ti kaθ-unde i

neg listen-IMPER.2pl puppies what sit-npst.ipfv.3pl art
paljanθropi ke sas le-ne.
skunks and iobj.2pl tell-npst.ipfv.3pl
‘Don’t listen, you puppies, what the skunks keep telling you.’  (CMG)

Considering that subjects and object clitics support argument structure require-
ments and are closely related to V2, these findings further reveal the tightness and
integration of the construction.
Peripheral adverbial modifiers are also found in 11 tokens, 10 of which are
temporal. Among them are me tis ores ‘for many hours’, ja mja olokliri evðomaða
‘for an entire week’, pende ores ‘for five hours’ and epi okto oloklira xronja ‘for
eight full years’. The extended semantic type is very much compatible with tem-
poral durational adverbs, which enhance the protracted engagement meaning of
kaθome in the construction, and can also be found in other positions, as in (20).
Temporal adverbials are not only compatible but, more importantly, they deter-
mine the extended interpretation in cases of ambiguity without them. Compare
(19) and (20). In (20), the ambiguity posed in (19) disappears because of the two
durational adverbials that make a postural reading highly improbable.

(19) kaθ-ete ke ta paratir-i

sit-npst.ipfv.3sg and obj.3pl watch.npst.ipfv.3sg
‘She sits and observes them. / She keeps observing them.’
 Soteria Svorou

(20) epi 31 xronja kaθ-ete ke ta paratiri

for years sit-npst.ipfv.3sg and obj.3pl watch-npst.ipfv.3sg
ateljotes ores tin imera
non-ending hours art day
‘For 31 years, she keeps observing them, for non-ending hours everyday.
(HNC)

Deliberate

Table 7. Structural traces of deliberate semantic type

Structural trace Total

V1 ke V2 47
V1 ke obj_cl V2 14
na V1 ke na V2 12
na V1 adv ke na V2 3
Totals 76 (11%)

The 76 deliberate type tokens retrieved in the two corpora shown in Table 7 rep-
resent 11% of the total tokens in all corpora. The number of different structural
traces in this type is restricted to four, an indication of the tightness of the con-
struction. There are only three tokens with an intervening adverbial between V1
and ke and these adverbials, I argue, have scope over the entire conjoined struc-
ture, not just V1 kaθome. In (21), the temporal adverb tora ‘now’ shows immediacy
not of sitting but of deliberately saying. It could be placed after po ‘say’ with no
change in meaning. The other two adverbs in the structural trace tokens are mazi
‘together’ and liɣo ‘a little’, both of which show similar modification and behavior
to tora ‘now’.
(21) to na kaθ-iso tora ke na po oti
art spart sit-npst.pfv.1sg now and spart say.npst.pfv.1sg that
o bajevits ðen me stirik-se, ðen
art B. neg obj.1sg support-pst.pfv.3sg neg
ex-i simasia.
have.npst.3sg meaning
‘For me to deliberately say now that Bajevic did not support me doesn’t
make sense.’

The remaining structural traces, which are in the majority, exhibit no intervening
items other than object clitics and the subordinating particle na. This particle,
even in coordinate structures, must be repeated. But the juncture is nevertheless at
 Constructional pressures on ‘sit’ in Modern Greek 

the core level, as modal auxiliaries, such as prepi ‘must’, have scope over the whole
structure, as in (22).
(22) θa prepi na kats-ume ke na ðu-me
fpart must spart sit-npst.pfv.1pl and spart see-npst.pfv.1pl
mazi ti bor-ume na kan-ume ja
together what be.able-npst.ipfv.1pl spart do-npst.ipfv.1pl for
ola afta
all these
‘We must deliberately investigate together what we can do for all these.’
(HNC)

The structural traces of both the extended and deliberate semantic types show a
pattern of syntactic tightness with a rather low number of non-argument items
intervening. The occurrence of clitics, but not full NP objects before V2, is also
indicative of this tightness. It is not, however, the type of tightness that is charac-
teristic of auxiliary verbs. The perfect auxiliary exo, for example, appears closer to
the verb than the clitic, as in (23).
(23) ixa kaθ-isi ke tis ixa pi tin istoria
have.pst.1sg sit.pfv and dobj.3sg have.pst.1sg tell.pfv art story
afti poles fores
this many times
‘I had told her this story deliberately many times.’

Also, a phenomenon known as clitic doubling – where an object clitic appears, even
though a full object NP occurs in the structure, thus doubling up the object – shows
the clitic before the V2 and not before the entire V1 ke V2. Consider examples (24):
(24) a. Ti kaθ-ese ke tu ɣraf-is tu
why sit-npst.ipfv.2sg and iobj.3sg write-npst.ipfv.2sg art
anθropu, kopela=mu?
man.gen girl=poss
‘My girl, why do you keep writing to the man?  (HNC)
b. *Ti tu kaθ-ese ke ɣraf-is tu
why iobj.3sg sit-npst.ipfv.2sg and write-npst.ipfv2sg art
anθropu, kopela=mu?
man.gen girl=poss
‘My girl, why do you keep writing to the man?

In (24a), the indirect object clitic tu duplicates on the already existing indirect
object tu anθropu and appears before V2 ɣrafis. An attempt to place it before the
whole kaθese ke ɣrafis structure in (24b) yields unacceptable results. I consider
 Soteria Svorou

this to be a solid argument that this Vkaθome ke V2 construction does not involve
kaθome as an auxiliary verb.

Simultaneous
The simultaneous semantic type names the conjunction of the event of sitting with
another event. As is shown in Table 8, 222 simultaneous tokens were retrieved
from the three corpora, which represent 32% of the total number. The most com-
mon structural trace is V1 ke V2, with no intervening items in main and subor-
dinate structures with particle na, accounting for 131 tokens. Another 46 (21%)
tokens contain locative, temporal, and manner adverbials before ke with scope
over V1 only. There are also 18 tokens with predicate adjectives in that position
that express posture and disposition and further reinforce the posture interpreta-
tion of kaθome. In 14 tokens (14%), an object clitic occurs before V2 and, in nine
tokens (4%), a negative particle with scope over V2 only further supports a two-
event interpretation of the simultaneous semantic type.

Table 8. Structural traces of simultaneous type

Structural traces Subtotal Total

V1 ke V2 114 131 (59%)

V1 V2 ke V3   4
θa V1 ke θa V2   3
na V1 ke na V2 10
V1subj ke V2   6   6
V1 ke   obj_cl V2 11 13
na V1 ke na obj_cl V2   2
V1 ke   neg V2   8   8
na V1 loc ke na neg V2   1 46 (21%)
na V1 mann ke na obj_cl V2   1
V1 mann ke V2   4
na V1 temp ke na V2   1
na V1 temp loc ke na V2   1
na V1 loc mann ke na V2   1
na V1 loc ke na V2 19
V1 temp ke V2   9
V1 temp loc ke V2   3
V1 loc mann ke V2   6
V1 predadj ke V2 18 18 (8%)
Total 222 222 (100%)
 Constructional pressures on ‘sit’ in Modern Greek 

Sequential

Table 9. Structural traces of sequential type

Structural traces Total

V1 ke V2 47
na V1 ke na V2   4
θa V1 ke θa V2   1
V1 ke obj_cl V2   8
na V1 ke na obj_cl V2   2
V1 loc ke V2   9
na V1 loc ke na V2 28
na V1 temp ke na V2   1
na V1 mann ke na V2   1
V1 ke neg V2   1
V1 ke temp V2   1
Totals 104 (15%)

The 104 tokens that represent the sequential type of conjunction constitute 15% of
the total number of tokens in all three corpora. The majority of these come from
the HNC. A striking but not unexpected result is that, in 39 of the 104 tokens
(37%), an adverbial of location, primarily, but also adverbials of time and man-
ner, in two tokens, occur before the conjunction ke, modifying V1 kaθome only.
There are also two tokens in which a temporal adverbial and a negative particle
are found before V2 having scope over V2 only. Adverbial and negation scope are
indeed excellent indicators of a syntactic chunk, in this case a clause. Sequential
coordination provides a syntactic juncture at the clausal level. Semantically, loca-
tive and manner adverbs modifying kaθome motivate a posture/change of posture
interpretation.

Collection
There were only 8 tokens of the collection type retrieved from all three corpora.
Among those few, characteristic is the presence in two examples of negative par-
ticles having scope over V2, but not V1 kaθome (with structural traces (na) V1
ke (na) neg V2) which supports a juncture at the clausal level and, semantically,
the interpretation of two events, and not one. This is further supported by the
presence of locative and manner adverbials modifying V1 kaθome in three of the
examples (with structural traces (na) V1 loc/mann ke (na) V2) and reveals the
posture interpretation of kaθome in this type of conjunctive coordination.
 Soteria Svorou

5.6  Subject Person/Number

In all corpora, there are more singular than plural subjects and more third per-
son than first and second person, with third person singular being the most com-
mon. We know that grammatical person distributions are sensitive to genre. These
trends reflect genre differences as captured by the corpora. Of 5026 occurrences of
lemma kaθome in the HNC, for example, 2204 (44%) are in 3sg and 1069 (21%)
in 3pl.

Table 11. Distribution of forms by person and number in the three corpora
By person/ number By number By person

1sg 141 (20%) 470 (67%) 217 (31%)

Singular

1st
2sg 73 (10%)
3sg 256 (37%) 92 (13%)

2nd
1pl 76 (11%) 230 (33%)
Plural

2pl 19 (3%) 3rd 391 (56%)

3pl 135 (19%)
700

First and second person are obviously animate and human. Of the third person
subjects, singular and plural alike, all are animate and most are human with only
a couple of animal subjects (e.g., to kunupi ‘the mosquito’; i foraðitsa ‘the mare’.
There are also a few metonymic nouns (e.g., i karðja mu ‘my heart’; to kratos ‘the
state’; i xaraktires ‘the characters’; to PASOK ‘PASOK’ (a political party); i gugl
‘Google’.
The animacy and humanness of these subjects is not at all surprising, as sitting
is a very human posture. Of the other extensions that yield non-posture readings,
the deliberate type also involves an animate, if not human, agent necessarily in
order to engage in purposeful, deliberate activity. Although one could argue that
the extended type does not necessarily need an animate agent, it, however, does
not occur with inanimate agents in my data. This is perhaps an indication that
the sense of agency that kaθome entails is still present in the extended type as well.

5.7 V2 verbs and their frames

Each token was further coded for the V2 verb and the semantic frame that the verb
instantiates. A frame is “a schematic representation of a situation involving various
participants, props and other conceptual roles, each of which is a frame element”
(FrameNet Glossary). Based on the understanding of the meaning of the verb in
 Constructional pressures on ‘sit’ in Modern Greek 

the particular token, a frame was assigned using FrameNet’s labels and categori-
zation.8 This is a necessary step in order to distinguish between senses of the verb
and different argument structures. Compare, for example, the verb ɣrafo ‘I write’
in the examples below that instantiate different frames, of Text_creation (25) and
of Contacting (26), which includes in its structure an extra argument, the recipient
sti nekri mitera tis ‘to her dead mother’.
(25) jati, peði mu, ðen asxol-ise me kati pjo
why child 1.poss neg deal-npst.ipfv.2sg with something more
erotiko ke kaθ-ese ke ɣraf-is afta ta
erotic and sit-npst.ipfv.2sg and write-npst.ipfv.2sg these art
praɣmata ja mena?
things about me
‘Why, my child, do you keep writing these things about me and don’t deal
with something more erotic?’  (CMG)
(26) ala kaθ-ise ki eɣrap-se ena ɣrama sti nekri
but sit-pst.pfv.3sg and write-pst.pfv.3sg one letter to.art dead
mitera=tis ja tin aðelfi pu ɣnoriz-e
mother=poss for art sister that know-pst.ipfv.3sg
ke aɣap-use
and love-pst.ipfv.3sg
‘But she sat and wrote a letter to her dead mother for the sister she know
and loved.’  (HNC)

The resulting data was then organized by frame and the verbs that instantiated
each. According to such organization, the verb ɣrafo ‘I write’ appears in two differ-
ent frames and such is the case with other verbs, for example vlepo ‘I see’ and akuo
‘I hear’ in Perception_experience and ‘I watch’ and ‘I listen’ in Perception_active.
Appendix 2 contains the list of frames with verbs instantiating them, organized
by frequency of occurrence in the three corpora. For this part, data from all three
corpora was collapsed into one set. Frames with 3 or more tokens are discussed
here. The following lists the frames by frequency of occurrence:

. FrameNet (Ruppenhoffer et al. 2010) is a lexical database with examples of how words
are used in actual texts. Meanings are described on the basis of Fillmorian semantic frames,
a semantic frame being a descritpion of a type of event or relation and the participants in it.
There is an assumption here that FrameNet’s categories and frames can be used as a tool to
account for any language, and indeed it has been developed for a variety of languages other
than English but Greek is not among them yet.
 Soteria Svorou

Perception_active 96 Activity_prepare 9 Trust 3

Statement 74 Activity_start 8 Rest 3
Cogitation 45 Experiencer_obj 8 Remembering_experience 3
Text_creation 40 Work 8 Record 3
Waiting 33 Questioning 7 Locating 3
Ingestion 23 Intentionally_act 6 Inspecting 3
Perception_experience 21 Play 5 Grant_ permission 3
Make_noise 17 Manufacturing 5 Give_impression 3
Chatting 13 Request_entity 4 Emotions_of_mental_activity 3
Scrutiny 12 Manipulation 4 Emotion_directed 3
Contacting 12 Manipulate_into_shape 4 Closure 3
Calculation 11 Judgement_communication 4 Cause_harm 3
Resolve_problem 10 Coming_up_with 4 Body_movement 3
Reading_activity 10 Attempt 4 Arriving 3

With the exception of the collection type, the same V2 verbs in the same frame
can be found in any of the other four types, sequential, simultaneous, deliberate,
and extended. The only frames found in the collection type, which is represented
by only 8 tokens in all three corpora, involve movement: Move, Arriving, Self_
motion and Posture (other than sitting). An example of a verb that can be found
in all four types is sizito ‘I discuss’, which instantiates the Statement frame. In
the sequential (27) and simultaneous (28) types, the verb kaθome is interpreted
as a change of posture and posture verb respectively, motivated by the locative
expressions sto trapezi ‘at the table’ and sto vaθos tu maɣazju ‘at the far end
of the shop.’ Locative expressions generally do not occur with the deliberate
(29) and extended (30) types, which differ from each other in terms of aspect,
as the deliberate appears with perfective and the extended with imperfective
morphology.

(27) Sequential
kal-unde i kinoniki eteri- kratos, erɣoðotes ke
call-pass.npst.ipfv.3pl art social partners state employers and
erɣazomeni- na kaθ-isun sto trapezi ke na
workers spart sit-npst.pfv.3pl at.art table and spart
sizit-isun ðio krisima θemata
discuss-npst.pfv.3pl two crucial topics
‘The social partners, state, employers and workers, are being called to sit at
the table and to discuss two crucial topics.’  (HNC)

(28) Simultaneous
kj i xara tu itan meɣali kaθos andikri-se
and art joy 3poss was big as see-pst.pfv.3sg
 Constructional pressures on ‘sit’ in Modern Greek 

sto vaθos tu maɣazju, to vulefti na

in.art depth art shop.gen art parliament.man spart
kaθ-ete ke na sizit-a mena
sit-npst.ipfv3sg and spart discuss-npst.ipfv1pl with.one
alon kirio
other gentleman
‘And his joy was great as he saw in the far end of the shop the member of
parliament sitting and discussing with another gentleman.’
(29) Deliberate
…o proeðros tis sirias xafez asad protin-e
 art president art S. H. A. propose-pst.pfv.3sg
na
kaθ-isun i ðio xores ke na

spart sit-npst.pfv.3sg art two countries and spart
sizit-isun tixon metra kata ton kurðon
discuss-npst.pfv.3sg possible measures against art K.
tu PKK.
art P.
‘According to the Turkish private channel NTV, the president of Syria Hafez
Asad proposed that the two countries sit and discuss possible measures
against the Kurds of PKK.’
(30) Extended
kaθ-omaste ke sizit-ame ti
sit-npst.ipfv.1pl and discuss-npst.ipfv.1pl what
θa ɣin-i me to ipurɣio ke ðen
fpart happen-npst.pfv.3sg with art ministry and neg
sizit-ame an se teliki analisi ta ɣrafia tipu me
discuss-npst.ipfv.1pl if in final analysis art offices press.gen with
afto to nomosxeðio proxor-un sosta i oxi.
this art bill proceed-npst.ipfv.3pl correctly or not.
‘We keep discussing what’s going to happen with the ministry and we don’t
discuss if, in final analysis, the press offices proceed with this bill correctly
or not.’

The forty most common frames can be organized in four larger categories of
events/frames: Perceptual events, communication events, mental activities, and
physical activities.
This is a very rough categorization; frames interact with one another in mul-
tiple and complex ways and a more accurate representation would require orga-
nization in terms of a network showing hierarchies and inheritance. Such study
needs to be undertaken in the future.
 Soteria Svorou

Table 12. Categories of V2 frames

Communication
Perceptual events events Mental activities Physical activities

Perception_active Statement Cogitation Make noise

Perception_experience Chatting Text_creation Ingestion
Experiencer_obj Contacting Scrutiny Work
Inspecting Questioning Calculation Intentionally_act
Request_entity Reading_activity Play
Judgement_ Resolve_problem Manufacturing
communication
Grant_ permission Coming_up_with Manipulation
Give_impression Trust Manipulate_into_shape
Remembering_ Record
experience
Emotions_of_ Rest
mental_activity
Emotion_directed Closure
Waiting Locating
Attempt Body_movement
Activity_prepare Cause_harm
Activity_start Arriving

Not surprisingly, many of the V2 verbs name situations and events that typically
occur while humans are in a seated position like eating, reading, writing, rest-
ing, watching television, thinking, talking, knitting, working, etc., all of which are
activities characterized by duration, just like the posture sense of V1 verb kaθome.
Others, however, have nothing to do with the posture of the human body, such as
believing, wondering, worrying, seeing, wasting time, teasing, criticizing, or even
waiting. Such situations are inherently durational and, as such, are compatible
with the extended type of coordination, in which they are commonly found. Situa-
tions such as judging, giving permission, agreeing, firing, finding, and remember-
ing generally lack duration and are more compatible with the deliberate type of
coordination.
An interesting effect of the semantics of V2 on the interpretation of the
semantic type can be seen in a few tokens in which the event expressed by V2
is understood to be undesirable, burdensome or detrimental, physically or
socially, to the shared subject, for example vrexome ‘I get wet’, jinome kleftis ‘I
become a thief ’ (31), ðexome eksftelismo ‘I accept embarrassment’ (32), xano
 Constructional pressures on ‘sit’ in Modern Greek 

‘I lose’, and xasomero ‘I waste time’. In such cases, kaθome has a non-postural
interpretation retaining the durational aspects of a sitting event (and, as such,
it is a subtype of the ‘extended’ type,) but because of the burdensome nature
of the V2 event, the agent is understood as tolerating or submitting to the
event of V2.

(31) “to oti ise anθropos tis ðekaras, ðen θeli rotima”,
art that be.npst2.sg man art penny.gen neg needs asking
ton plirose ap tin kali o kitelis, “afu kaθ-ese ke
him paid from art good art Kitelis, when sit-ipfvnpst.2sg and
jin-ese kleftis ja ena kafe.”
become-ipfv.npst.2sg thief for one coffee
‘”That you are a penny pincher, is not a question”, Kitelis replied to him
pointedly, “since you submit to becoming (lit. sit and become) a thief for one
coffee.”  (HNC)

(32) andi na ex-ete eliniki viomixania na

instead PART have-ipfv.npst.2pl Hellenic industry PART
kataskeva-zi stoles ja na antimetop-isun i
manufacture-ipfv.npst.3sg uniforms for PART face-pfv.npst.3pl art
ðinamis tis kenurjes asimetres apiles, kaθ-este ke
forces art new asymmetrical threats sit-ipfv.npst.2pl and
ðex-este ton ekseftelismo na sas
accept-ipfv.npst.2pl art embarrassment PART iobj.2pl
ferni eksinda tris stoles o miler ke na
bring-ipfv.npst.3sg sixty three uniforms art Miller and PART
sas tis kan-i ðoro
iobj.2pl obj.3pl do-ipfv.npst.3sg present
‘Instead of having some Greek industry manufacture uniforms so that the
(armed) forces can face the new lopsided threats, you submit to accepting
(lit. sit and accept) the embarrassment of Miller bringing you sixty three
uniforms and giving them to you as presents.’  (HNC)

5.8 Pragmatic confluences

While in the majority of the extended type tokens the construction occurs in a
declarative sentence thus being asserted, in a number of tokens (10 (9%) in HNC;
11 (11%) in CGT; and 24 (27%) in CMG), the construction occurs within a ‘why’
interrogative question form that is interpreted as a rhetorical question. The import
of such rhetorical questions is criticism by the speaker on the actions reported
by the sentence’s predicate. The criticism extended may be further reinforced by
 Soteria Svorou

unflattering characterizations of the subject – as in (33), mistirii anθropi ‘mysteri-

ous people’ – or swearing expressions – as in (34), ðjaolo ‘the hell’. The resulting
interpretation in these examples accentuates the speaker’s estimation of the sub-
ject’s engagement with the V2 event as being protracted and extended beyond her
preference.
(33) ala ke afti i ɣermani mistirii anθropi, ti
but and these art Germans mysterious people why
kaθ-onde ke psaxn-un?
sit-npst.ipfv.3pl and search-npst.ipfv.3pl
‘But these Germans, mysterious people, why do they keep searching? (CMG)
(34) ti ðjaolo tora kaθ-omaste ke
why hell now sit-npst.ipfv.1pl and
milame j=afta?
talk.npst.ipfv.1pl about=these
‘Why the hell now do we keep talking about these?

It is important to point out that only tokens of the extended type have been
found in the three corpora in such sentential contexts, which leads us to
consider that speaker comment may be part of the pragmatic aspects of the
construction. If so, this would fit in well with the interpretation of (35), even
though the construction is found in a declarative sentence. Notice, however,
the andi ‘instead’ adverbial clause that modifies the main clause containing the
extended construction; it sets up an opposition that could invite speaker com-
ment or criticism.
(35) ke andi o proθipurɣos na tu pi
and instead art prime.minister spart obj.3sg tell.npst.pfv.3sg
na pai
spart go.npst.pfv.3sg
sto astinomiko tmima tis perioxis=tu ke
to.art police station art.gen area.gen=poss and
na to ðosi, kaθ-ete ke
spart obj.3sg give.npst.pfv.3sg sit-npst.ipfv.3sg and
sinalas-ete ke sizita-i epi ores ke
negotiate-npst.ipfv.3sg and discuss.npst.ipfv.3sg for hours and
eksefteliz-i olin tin ðiikisi.
embarrass-npst.ipfv.3sg all art administration
‘And, instead of telling him to go to the police station of his area and to give it, the
Prime Minister keeps negotiating and discussing for hours and embarrassing the
whole administration.’
 Constructional pressures on ‘sit’ in Modern Greek 

Whether, historically, the first instances of the extended type construction

occurred in the rhetorical question sentential environment cannot be answered
here as it requires a diachronic study. But such a hypothesis is quite plausible as the
same rhetorical question environment also occurs with Vpijeno ke V2 in which the
verb pijeno ‘I go’ is schematized as a frame of intentionality with some pragmatic
inference of criticism on the part of the speaker (as in 36) and the criticism can
also be sensed in declarative environments (as in 37).
(36) o nulis ke o sakis skeftikane mipos
art N. and art S. think-pst.pfv.3pl whether
ine voθrokaθaristis- ti pa-ne ke
be.npst.3sg septic.tank.cleaner what go-npst.ipfv.3pl and
skeft-onde afta ta ðiðima?
think-npst.ipfv.3pl these art twins
‘Nulis and Sakis thought that he was a septic tank cleaner – what do these
twins end up thinking (lit. go and think)?

(37) ke me ton tropo=mu tin agapus-a ti maria ke

and with art way=poss obj.3sg love-pst.IMP.1sg art Maria and
tora afti piγ-e ke emin-e engios me
now she go-pst.pfv.3sg and stay.pst.3sg pregnant with
kapjon alo,…
someone else
‘And in my own way I loved Mary and now she intentionally got pregnant
(lit. went and got pregnant) with someone else, …’

5.9 Summary
We have seen that there are multiple interactions that contribute to the resulting
interpretation of the Vkaθome ke V2 construction: the type of V2 and its semantic
frame, its inherent aspectual character or Aktionsart, its emotional connotations,
the morphologically imposed aspectual frame, and the presence or absence of loc-
ative and/or temporal adverbials, and even the illocutionary force of the sentence
in which the construction occurs.
The correlation of the simultaneous and the extended type with imperfective
aspect and of the sequential and the deliberate with perfective aspect justifies bifur-
cation of the relationships among the different types. As it is shown in Figure 1,
these types differ with respect to the degree of conceptual integration, with the
extended and deliberate exhibiting most integration in their respective aspectual
frames, while the unordered collection type shows the least integration, lacking any
correlation with aspect, as it is compatible with either.
 Soteria Svorou

MOST
Imperfective Perfective

Degree of conceptual integration

Extended Deliberate

Simultaneous Sequential

Unordered
Collection

LEAST

Figure 1. Degree of conceptual integration

6. Conclusion

This study explored a particular instantiation of the syntactic string V1 and V2 that
occurs in structures that have been traditionally termed “conjunction reduction”
and more recently “elliptic coordination” (Langacker 2012). Apart from exhibiting
lack of the subject of V2, which is the reason for the term “conjunction reduction”,
this construction shows patterns of reinterpretation of V1 justifying subtypes. The
current study dealt with the instantiation of V1 by the Modern Greek posture verb
kaθome ‘I sit’ and its reinterpretation in this construction. Taking a usage-based
approach, examples of the string Vkaθome ke V2 were retrieved from corpora of
written and oral language and examined for parameters contributing to the con-
struction of meaning and the interpretation of the examples. It was shown that fine
semantic distinctions can be made and that such distinctions can be attributed to
an intricate interaction of the l­exical semantics of kaθome and its experientially-
grounded entailments, the lexical semantics and the semantic frames that V2 verbs
instantiate, the aspectual distinctions marked morphologically on the two verbs,
and the presence or absence of locative, temporal, or manner adverbials, or pred-
icative adjectives within the boundaries between the two verbs and beyond. All
these interacting parameters are organized and supported by the syntactic frame
of the string Vkaθome ke V2 which functions as the catalyst for the integration of the
 Constructional pressures on ‘sit’ in Modern Greek 

two verbs in ways that are primarily determined by the nature of the first verb. As
shown in Svorou (in press), motion verbs piɣeno ‘I go’ and erxome ‘I come’ in the
V1 syntactic position also exhibit similar characteristics of the type of construction
discussed here but with a resulting interpretation analogous and attributable to
the lexical semantics of these two verbs. Such findings support the hypothesis that
syntactic strings and not just lexical items can be associated with specific mean-
ing, challenging a clear lexicon-grammar distinction. In assessing the function of
kaθome in the Vkaθome ke V2 string, it is clear that its prototypical lexical semantics
motivates its ultimate interpretation but its sense of extended duration and deliber-
ate action only arise in the context of the syntactic string and the morphologically-
coded aspectual enhancements and justifies the proposal for a subtype of the V1 ke
V2 construction. The discourse context also seems to contribute to certain prag-
matic inferences tied to the construction providing further motivation for consid-
ering the string a form-meaning composite.
Although this study does not include it, a diachronic component would be in
order to further explore the identified parameters in the depth of time. Construc-
tionalization happens over time but its beginnings can be captured by synchronic
studies like this one.

Abbreviations

1 first person mann manner adverbial

2 second person npst non-past
3 third person obj direct object clitic
adv adverb pass passive
art definite article pst past tense
spart subordinating particle pfv perfective aspect
fpart future particle pfvaux perfect auxiliary ‘have’
fut future pl plural
iobj indirect object clitic predadj predicate adjective
ipfv imperfective aspect sg singular
interrog interrogative temp temporal adverbial
loc locative adverbial

References

Corpus of Modern Greek, Russian Academy of Sciences. <http://web-corpora.net/Greek­

Corpus> (36 million tokens)
Corpus of Greek Texts. <http://www.sek.edu.gr> (30 million tokens)
Hellenic National Corpus, Institute of Language and Speech Processing. <http://hnc.ilsp.gr>
(1990-current) (47 million tokens)
 Soteria Svorou

Bybee, Joan L. 2015. Language Change. Cambridge: CUP.

Bybee, Joan L. 2013. Usage-based theory and exemplar representations of consrtuctions. In The
Oxford Handbook of Construction Grammar, Thomas Hoffman & Graeme Trousdale (eds,),
49–69. Oxford: OUP.
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar: Tense,
Aspect, and Modality in the Languages of the World. Chicago IL: University of Chicago
Press.
Croft, William. 2001. Radical Construction Grammar. Oxford: OUP.
doi: 10.1093/acprof:oso/9780198299554.001.0001
Croft, William & Cruise, D. Alan. 2004. Cognitive Linguistics. Cambridge: CUP.
doi: 10.1017/CBO9780511803864
De Vos, Mark Andrew. 2005. The Syntax of Verbal Pseudo-coordination in English and Afrikaans.
Utrecht: LOT.
FrameNet Glossary. 2016. Retrieved from <https://framenet.icsi.berkeley.edu/fndrupal/
glossary>
Hebb, Donald O. 2002[1949]. The Organization of Behavior: A Neuropsychological Theory.
­Mahwah NJ: Lawrence Erlbaum Associates.
Heine, Bernd & Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
doi: 10.1017/CBO9780511613463
Hilpert, Martin & Koops, Christian. 2009. A quantitative approach to the development of
­complex predicates: The case of Swedish pseudo-coordination with sitta ‘sit’. In Syntac-
tic Complexity: Diachrony, Acquisition, Neuro-cognition, Evolution [Typological Studies
in Languages 85], T. Givón & Masatyoshi Shibatani (eds), 145–162. Amsterdam: John
Benjamins. doi: 10.1075/tsl.85.06aqu
Kuteva, Tania A. 1999. On ‘sit’/’stand’/’lie’ auxiliation. Linguistics 37(2): 191–213.
doi: 10.1515/ling.37.2.191
Langacker, Ronald W. 2012. Elliptic coordination. Cognitive Linguistics 23(3): 555–599.
doi: 10.1515/cog-2012-0017
Lemmens, Maarten. 2005. Aspectual posture verb constructions in Dutch. Journal of Germanic
Linguistics 17(3): 183–217. doi: 10.1017/S1470542705000073
Lødrup, Helge. 2002. The syntactic structures of Norwegian pseudocoordinations. Studia
­Linguistica 56(2): 121–143. doi: 10.1111/1467-9582.00090
Newman, John (ed.). 2002. The Linguistics of Sitting, Standing, and Lying [Typological Studies in
Language 51]. Amsterdam: John Benjamins. doi: 10.1075/tsl.51
Newman, John & Rice, Sally. 2004. Pattern of usage of sit, stand, and lie: A cognitively inspired
exploration in corpus linguistics. Cognitive Linguistics 15(3): 351–396.
doi: 10.1515/cogl.2004.013
Newman, John & Rice, Sally. 2008. Asymmetry in English multi-verb sequences: A corpus-based
approach. In Asymmetric Events [Converging Evidence in Language and ­Communication
Research 11], Barbara Lewandowska-Tomaszczyk (ed.), 3–22. Amsterdam: John
Benjamins. doi: 10.1075/celcr.11.03new
Noël, Dirk. 2007. Diachronic construction grammar and grammaticalization theory. Functions
of Language 14(2): 177–202. doi: 10.1075/fol.14.2.04noe
Ross, Daniel. 2016. Between coordination and subordination: Typological, structural and dia-
chronic perspectives on pseudocoordination. In Coordination and Subordination: Form
and Meaning – Selected Papers from CSI Lisbon 2014, Fernanda Pratas, Sandra Pereira &
Clara Pinto (eds), 209–243. Newcastle upon Tyne: Cambridge Scholars.
 Constructional pressures on ‘sit’ in Modern Greek 

Ruppenhoffer, Josef, Ellsworth, Michael, Petruck, Miriam R. L., Johnson, Christopher R. &
Scheffczyk, Jan. 2010. FrameNet II: Extended Theory and Practice. <https://framenet2.icsi.
berkeley.edu/docs/r1.5/book.pdf>
Svorou, Soteria. In press. Motion verb integration and cosubordination in Modern Greek. In
Applying and Expanding Role and Reference Grammar [HPSL – NIHIN Studies], Rolf
­Kalluweit, Lisann Künkel & Eva Staudinger (eds),163–185. Freiburg: Freiburg University
Press.
Traugott, Elizabeth Closs & Trousdale, Graeme. 2014. Constructionalization and Constructional
Changes. Oxford: OUP.
Τριανταφυλλίδης, Μανόλης. 1998. Λεξικό της Κοινής Νεοελληνικής. Ινστιτούτο Νεοελληνικών
Σπουδών του Αριστοτελείου Πανεπιστημίου Θεσσαλονίκης. <http://www.greek-language.
gr/greekLang/modern_greek/tools/lexica/triantafyllides/index.html>
Van Valin, Robert D. Jr. 2007. Recent developments in the Role and Reference Grammar theory
of clause linkage. Language and Linguistics 8(1): 71–93.
Viberg, Åke. 2013. Posture verbs: A multilingual contrastive study. Languages in Contrast 13(2):
139–169. doi: 10.1075/lic.13.2.02vib
Wiklund, Anna-Lena. 2007. The Syntax of Tenselessness: Tense/mood/aspect-agreeing Infinitivals.
Berlin: Mouton de Gruyter.
 Soteria Svorou

Present indicative Future imperfective indicative Irrealis (present imperfective) subjunctive

Singular Plural Singular Plural Singular Plural

1st kaθome kaθomaste θa kaθome θa kaθomaste na kaθome na kaθomaste
2nd kaθese kaθosaste θa kaθese θa kaθosaste na kaθese na kaθosaste
3rd kaθete kaθonde θa kaθete θa kaθonde na kaθete na kaθonde
Past Imperfective Indicative Conditional Imperfective Irrealis (Past Imperfective) Subjunctive
Singular Plural Singular Plural Singular Plural
1st kaθomun kaθomastan θa kaθomun θa kaθomastan na kaθomun na kaθomastan
2nd kaθosun kaθosastan θa kaθosun θa kaθosastan na kaθosun na kaθosastan
3rd kaθotan kaθondan θa kaθotan θa kaθondan na kaθotan na kaθondan
Past Perfective Indicative Conditional Perfective Irrealis (Future Perfective) Subjunctive
Singular Plural Singular Plural Singular Plural,
1st kaθisa kaθisame θa kaθiso θa kaθisume na kaθiso na kaθisume
2nd kaθises kaθisate θa kaθisis θa kaθisete na kaθisis na kaθisete
3rd kaθise kaθisan θa kaθisi θa kaθisun na kaθisi na kaθisun
Alternative perfective forms in free variation occuring in informal registers
1st ekatsa katsame θa katso θa katsume na katso na katsume
2nd ekatses katsate θa katsis θa katsete na katsis na katsete
3rd ekatse ekatsan θa katsi θa katsun na katsi na katsun
 Constructional pressures on ‘sit’ in Modern Greek 

Appendix 2. V2 verbs and their frames

Table 13. Verbs and frames with more than 3 tokens

Frames Verbs #

Perception_active Kitazo ‘look’ 29

Parakoluθo ‘watch’ 22
Akuo ‘listen 15
Xazevo ‘gawk’ 10
Vlepo ‘watch’ 9
Kamarono ‘flaunt’ 3
Aɣnandevo ‘gaze’ 2
Paratiro ‘observe’ 1
Psilafo ‘feel’ 1
Xerome ‘enjoy’ 1
Mirome ‘smell’ 1
Exo stilomeno vlema ‘stare’ 1
Vasto matia ‘keep eyes on’ 1
Subtotal 96
Statement Leɣo ‘say, tell’ 34
Milo ‘talk, speak’ 16
Sizito ‘discuss’ 20
Platiazo ‘discuss’ 1
Kreno ‘discuss’ 1
Koutsaro ‘coach’ 1
Perno t aftia ‘talk for a long time’ 1
Subtotal 74
Cogitation skeftome ‘think’ 29
anarotjeme ‘wonder’ 2
ðialoɣizome ‘speculate’ 2
siloɣizome ‘meditate’ 2
analoɣizome ‘cogitate’ 2
remvazo ‘muse’ 5
prosefxome ‘pray’ 1
θeoro ‘consider’ 1
kano skepsis ‘have thoughts’ 1
Subtotal 45
(Continued)
 Soteria Svorou

Table 13. (Continued)

Frames Verbs #
Text_creation ɣrafo ‘write’ 36
ðiiɣume ‘narrate’ 3
afiɣume ‘narrate’ 1
Subtotal 40
Waiting perimeno ‘wait’ 33
Subtotal 33
Ingestion pino ‘drink’ 10
katevazo ‘down (drinks)’ 1
rufo ‘sip’ 1
troɣo ‘eat’ 10
siɣotroɣo ‘eat slowly’ 1
Subtotal 23
Perception_experience vlepo ‘see’ 16
akuo ‘hear’ 5
exo psevðesθisi ‘have illusion’ 1
ɣnorizo ‘experience’ 1
Subtotal 21
Make_noise kleo ‘cry’ 9
fonazo ‘scream’ 2
jelo ‘laugh’ 2
θrino ‘wail’ 1
klapsurizo ‘whimper’ 1
anastenazo ‘moan’ 1
ɣurɣurizo ‘purr’ 1
Subtotal 17
Chatting kuvendjazo ‘chat’ 8
grinjazo ‘grumble’ 3
pjano kuvenda ‘chat’ 1
kano kutsoboljo ‘gossip’ 1
Subtotal 13
Scrutiny analio ‘analyze’ 4
meleto ‘study’ 3
psaxno ‘search’ 3
leptoloɣo ‘scrutinize’ 1
perierɣazome ‘pry’ 1
Subtotal 12
 Constructional pressures on ‘sit’ in Modern Greek 

Table 13. (Continued)

Frames Verbs #
Contacting ɣrafo ‘write’ 10
tilefono ‘all’ 1
afiɣume ‘narrate’ 1
Subtotal 12
Calculation Metro ‘count, measure’ 5
Arθrizo ‘sum up’ 1
Ksanametro ‘recount’ 1
Ipoloɣizo ‘calculate’ 2
Loɣarjazo ‘tally’ 2
Subtotal 11
Reading_activity ðiavazo ‘read’ 9
ksanaðiavazo ‘reread’ 1
Subtotal 10
Resolve_problem asxolume ‘engage’ 6
Lino ‘solve’ 1
Vrisko lisi ‘find solution’ 1
Ðiorθono ‘fix’ 1
Palevo ‘struggle’ 1
Subtotal 10
Activity_start Arxizo ‘begin’ 8
Subtotal 8
Activity_prepare Etimazo ‘prepare’ 4
Etimazome ‘get ready’ 3
Ekpono ‘prepare’ 2
Subtotal 9
Experiencer_obj Anaðevo ‘recall’ 1
Maɣevome ‘be enchanted’ 1
Enoxlo ‘bother’ 1
Skotizome ‘care’ 1
Ðiaskeðazo ‘gallivant’ 1
Efisixazo ‘placate’ 1
Skao ‘worry’ 1
Zalizo ‘dizzy’ 1
Subtotal 8
(Continued)
 Soteria Svorou

Table 13. (Continued)

Frames Verbs #
Work Ðulevo ‘work’ 4
Erɣazome ‘work’ 1
Peðevome ‘struggle’ 2
Tiranjeme ‘toil’ 1
Subtotal 8
Questioning Roto ‘ask’ 5
Psiloroto ‘ask discreetly’ 2
Subtotal 7
Intentionally_act Kano ‘do’ 6
Subtotal 6
Play Pezo ‘play’ 5
Request_entity Parangelno ‘order’ 3
Ðino parangelia ‘give an order’ 1
Subtotal 4
Manufacturing Ftjaxno ‘make’ 3
Kano ‘make’ 1
Epekserɣazome ‘process’ 1
Subtotal 5
Manipulation Xaiðevo ‘caress’ 2
Krato ‘hold’ 1
Pjano ‘grab’ 1
Subtotal 4
Manipulate_into_shape Pleko ‘knit’ 1
Strivo ‘twist’ 1
Mantaro ‘darn’ 1
Ifeno ‘weave’ 1
Subtotal 4
Judgement_communication Katiɣoro ‘accuse’ 1
Katiɣorume ‘be accused’ 1
Kano kritiki ‘criticize 1
Kano atakes ‘attack’ 1
Subtotal 4
 Constructional pressures on ‘sit’ in Modern Greek 

Table 13. (Continued)

Frames Verbs #
Coming_up_with Sxeðiazo ‘design’ 3
Epinoo ‘devise’ 1
Subtotal 4
Attempt Prospaθo ‘try’ 3
Zitjanevo ‘beg’ 1
Subtotal 4

Table 14. Verbs and frames with 3 tokens

Frame Verb # tokens

Trust pistevo ‘believe’ 3

Rest ksekurazome ‘rest’ 1
anapavome ‘rest’ 1
isixazo ‘relax’ 1
Remembering_ θimame ‘remember’ 2
experience
anapolo ‘recollect’ 1
Record maɣnitofono ‘tape record’ 1
ixoɣrafo ‘sound record’ 1
kano ixoɣrafisi ‘sound record’ 1
Locating vrisko ‘find’ 3
Inspecting eksetazo ‘examine’ 2
perierɣazome ‘peruze’ 1
Grant_permission afino ‘let’ 2
ðexome ‘accept’ 1
Give_impression fenome ‘appear’ 1
exo ifos ‘have expression’ 1
ðixno ‘appear’ 1
Emotions_of_ terpome ‘enjoy’ 1
mental_activity
ksefandono ‘regale’ 1
apolamvano ‘enjoy’ 1
Emotion_directed stenoxorjeme ‘get distressed’ 1
varjeme ‘get bored’ 1
zalizome ‘get dizzy’ 1
(Continued)
 Soteria Svorou

Table 14. (Continued)

Frame Verb # tokens

Closure aniɣo ‘open’ 3
Cause_harm ðerno ‘beat’ 1
troɣo ksilo ‘get spanked’ 1
petrovolo ‘stone’ 1
Body_movement kuno ‘wave’ 2
xazmurjeme ‘yawn’ 1
Arriving erxome ‘come, return’ 1
epistrefo ‘return’ 1
ɣirizo ‘return, come back’ 1
 know and understand in ASL
A usage-based study of grammaticalized topic
constructions

Terry Janzen
University of Manitoba

Topic constructions in ASL are understood to be composed of information

identifiable to the addressee that serves as the reference point from which
to view the comment or comments immediately following it. This study
compares the instances of use in a corpus of conversational ASL of KNOW and
UNDERSTAND as lexical verbs and their grammaticalized form and function
when they appear as topic markers, and with KNOW as a discourse marker. Topic
phrases introduced by KNOW tend to have a wider domain as an identifiable
reference point rather than a specific entity or event within the signer’s and
addressee’s shared knowledge. Topic phrases introduced by UNDERSTAND tend
to introduce an idea that the signer wishes to set as a reference point.

1. Introduction1

Topic constructions in American Sign Language (ASL) are understood to be

composed of information identifiable to the addressee that serves as the ground
or reference point from which to view the comment or comments that imme-
diately follow it (Janzen 1998a, 1999), functioning similarly to topic-comment
structure described in other languages (e.g., Haiman 1978; Li & Thompson 1976).
Topics in ASL are marked with raised eyebrows and typically a backward head
tilt that remain for the duration of the topic phrase. Topic phrase content is vari-
able because anything understood to be known or shared information can be
subjectively chosen by the signer as a grammatical topic – noun phrases and full
clauses are common. It is of interest then that two lexical verbs are regularly seen

. This study has benefitted greatly from discussion with Barbara Shaffer, Erin Wilkinson,
Jila Ghomeshi, Kevin Russell, two anonymous reviewers, and importantly, the generosity of
members of the deaf community in Winnipeg who contributed to the corpus. As always, any
errors remain my own.

doi 10.1075/slcs.192.03jan
© 2018 John Benjamins Publishing Company
 Terry Janzen

to begin certain topic phrases. These are KNOW (Janzen 2007) and UNDER-
STAND (­Janzen 2010).
Russell, Wilkinson and Janzen’s (2011) corpus study of undershoot in ASL
looked at signs that have a canonical contact point on or near the signer’s head but
where instances of use in conversational data did not reach their target location.
Two of these signs are the verbs of cognition KNOW ‘to know’ and UNDER-
STAND ‘to understand’, both with a target contact point on the signer’s forehead.
For KNOW, the actual contact clusters around two areas, the forehead, as expected,
but also the lower cheek. For UNDERSTAND Russell et al. also found a significant
number of instances of use that did not reach the forehead, but a cluster around
an alternate location did not take place, and some instantiations were located at
the most extreme distances from the target contact point of all the undershot signs
noted in the study. Russell et al. speculate that the two cluster locations for KNOW
might indicate an emerging second meaning or function for the second cluster
at the lower cheek, which may be consistent with Bybee and Scheibman’s (1999)
description of new constructions being formed when constituent properties begin
to differ phonetically and semantically from a related form, resulting in differenti-
ated cognitive representations, such as with I don’t know and I dunno in English.
The present study examines in greater detail what might be taking place with
the instances of use of both KNOW and UNDERSTAND. Both items appear in
topic constructions to introduce the topic phrase and when they do, they tend not
to reach the forehead contact point. That is, KNOW more often contacts the cheek
in these constructions, which may account for the cluster observed in Russell et al.
UNDERSTAND in topic constructions also greatly undershoots the forehead, but
does not appear to cluster elsewhere. In these cases KNOW and UNDERSTAND
do not carry lexical meaning, they are no longer accompanied by the same mor-
phology that the lexical verbs are, and they appear to function as overt markers
of particular topic constructions, which indicates the emergence of new construc-
tions (Bybee 2006). This study compares the instances of use in a corpus of con-
versational ASL (Russell et al. 2011 examined a portion of this corpus) between
KNOW and UNDERSTAND as lexical verbs and their grammaticalized form and
function when they appear as topic markers. A third category that emerges for
KNOW is its use as a discourse marker. This use is less central to the focus of this
paper, but its analysis is included because it contributes to the overall understand-
ing of how KNOW functions lexically and grammatically in ASL.
In Section 2 below, I summarize the role that grammaticalization has
played in the evolution of signed languages as background for the discussion
of KNOW and UNDERSTAND in the sections that follow. Section 3 looks in
particular at the grammaticalization of topic marking and topic constructions
in ASL. This serves as a backdrop for KNOW and UNDERSTAND as topic
 know and understand in ASL 

­ arkers in certain topic-­comment usages. In Section 4, I outline the corpus

m
data that this study is based upon. Sections 5 and 6 examine in detail the gram-
maticalization of the lexical verbs KNOW and UNDERSTAND and their emer-
gence in topic constructions. In Sections 7 and 8, I conclude by discussing the
overall significance of this aspect of language evolution.

2. Grammaticalization in signed languages

In contrast to the literature on grammaticalization in spoken languages, that for

signed languages is yet relatively small. Studies do exist, however, at least for some
signed language grammatical categories. Part of the problem is a distinct lack of
historical signed data. Signed languages do not have written forms, thus detailed
recording has only been possible since the availability of visual recording devices.
The earliest known recordings for ASL are a small set of ASL oratory texts filmed
around 1913 and widely available today on DVD.2 The films were compiled by
the National Association for the Deaf in the United States for the purpose of pre-
serving the signed language at the time. The signed language used in the films is
sometimes referred to as Old ASL (OASL) in contrast to modern, or currently-
used, ASL. Because this data source is limited, and because the preservation nature
of the films means that the signing style is formal, careful, and more than likely
rehearsed, their usefulness in grammaticalization studies is also somewhat lim-
ited. Nonetheless we are able to see some clear examples of constructions already
undergoing grammaticalization (see, for example, Janzen 1998a, 1999; Janzen &
Shaffer 2002; Shaffer 2000).
Other valuable sources are historic dictionaries of signed languages, for exam-
ple, Long (1918) for ASL, Brouland (1855) and others (see Renard & Delaporte
2004 for a collection of Old French Sign Language (Old LSF) dictionaries), given
that ASL has many historical roots in Old LSF. We now have a good understanding
of many lexical sources that are frequently found in grammaticalized construc-
tions, for example verbs of desire (e.g., want) and of motion (e.g., go) as source
words for grammaticalized markers of future (Bybee, Perkins & Pagliuca 1994;
Heine & Kuteva 2002), modal constructions, and the like, such that we can at
times look at dictionary entries of potential “candidate” words to see relation-
ships that might exist with modern grammatical constructions. These dictionaries
sometimes include descriptions of usage, which may suggest a variety of m ­ eanings

. The Preservation of American Sign Language. Sign Media, Inc., 4020 Blackburn Lane,
­Burtonsville, MD 20866–1167.
 Terry Janzen

and usages, in some cases pointing to an already grammaticalizing item. In Long’s

(1918) dictionary, as one example, possessive HAVE is described as “Cross the
open hands in front, palms to self and then draw them up against the breast”
(Long 1918: 26), iconically representing bringing the possessed thing to the sign-
er’s body. In modern ASL HAVE, signed with the fingertips of both hands moving
toward and making contact with the upper chest, but not crossed, can have either
possessive or existential meaning, and when existential, the signer and the signer’s
body are not significantly iconic (e.g., the room has curtains). In casual signing
in modern ASL only one hand is frequently used, which is one type of phonetic
reduction.
However, because our knowledge of grammaticalization has become expan-
sive, and many general principles have been identified (e.g., Hopper 1991), syn-
chronic data itself can often reliably reveal grammaticalization that has taken, or is
in the process of taking, place. In the work on signed languages, sources for gram-
maticalizing items have frequently been shown to be gestural, which Heine and
Kuteva (2007) acknowledge signifies an important distinction between grammati-
calization in spoken and signed language. Janzen and Shaffer (2002) and Wilcox
(2007) give evidence of two routes of grammaticalization, one in which there is a
grammaticalization path from a gestural source through a lexical item to a gram-
matical construction, and one where there is no intervening lexical item, in other
words the grammaticalized construction evolves directly from a gestural source.
Much of the work on language change in signed languages has been on histor-
ical lexical development and lexicalization (Frishberg 1975; Janzen 2012a; Leeson
& Grehan 2004). But given that mechanisms of change, including those that result
in grammaticalization, appear to be universal (Bybee 2001), we might expect that
mechanisms of both lexical and grammatical change might not be affected by
language modality, and this is borne out in the studies of signed language gram-
maticalization undertaken to date. One of the first such ASL studies was Janzen
(1995) on the verb FINISH grammaticalizing into both perfective and completive
markers. Much of the grammaticalization research on ASL has focused on modals
(Shaffer 2000, 2002, 2004; Shaffer & Janzen 2016; Wilcox & Wilcox 1995), dis-
course markers (Wilcox 1998), and topic marking and topic-comment construc-
tions (Janzen 1998a, 1999, 2007; Janzen & Shaffer 2002).

3. Topic marking and topic constructions in ASL

In topic-comment constructions, the topic phrase consists of shared or identifiable

information to the addressee and the comment that follows is some new informa-
tion (see, e.g., Givón 1984; Li & Thompson 1976), The topic phrase acts as a ground
 know and understand in ASL 

or reference point from which to view the comment, and the syntactic connec-
tion between the two is often fairly loose. In topic-comment constructions in ASL,
Janzen (1998a, 1999) found that topics are frequently clausal, full noun phrases
(but rarely pronominal), temporal phrases, and the grounding information may be
found either as earlier mentions in the preceding discourse or within the pragmatic
domain, that is, information assumed by the signer to be known by the addressee.
In addition, Janzen (1998a, 1999) shows that topic marking is not a topic mainte-
nance device, but rather a strategy in which successively marked topics consistently
move the discourse in a slightly different direction, and therefore signal topic shift.
Items that maintain a high degree of topicality in a sequence of utterances, espe-
cially when nominal, tend to be characterized by less coding, frequently going to
zero expression. This combination of marked topic shift and reduced coding for
highly topical items has two results. First, marked topic phrases tend to have more
fully specified information (e.g., full clauses or noun phrases) and second, the most
topical elements in utterances may not appear in topic phrases, but are found within
the comment phrase. A simple example is found in (1) from Janzen (2007: 184):
(1) [TOMORROW NIGHT]-top WORK
‘Tomorrow evening I am working.’

In (1) the grounding or reference point information is the temporal phrase

TOMORROW NIGHT, which functions as the “topic” by virtue of its accessibility
as a known and reasonable frame for an event to take place. However, an element
that is higher in topicality in this construction is first person reference, marked as
zero, understood as an element closely associated with the verb WORK, and there-
fore not a part of the topic phrase. Note that ASL also allows for an overt pronoun
in the comment, such as a 1s pronoun either immediately preceding or following
the verb. Note also that in terms of topic and comment function, TOMORROW
NIGHT is the (shared) reference point whereas the new information, and there-
fore the utterance focus, is that the signer will be working. Thus, (1) might be the
answer to a question such as what are you doing tomorrow evening? but not when
are you next working?
Topic marking has been described as raised eyebrows and a slight backward
head-tilt that continue throughout the articulation of the topic phrase. This may
be followed by a brief pause, along with the eyebrows lowering and the head
returning to a neutral position between the topic and the comment. Although
earlier writing on ASL considered these “nonmanual” (facial and head) markers
as obligatory for topics (e.g., Baker & Cokely 1980), Janzen (2007) questions the
obligatoriness of them, citing examples of clear topic-comment utterance struc-
ture that have no perceivable facial and head marking. One example is given in (2)
from Janzen (2007: 187):
 Terry Janzen

(2) WINDOW LEAN.TOWARD(with elbow)

topic comment
‘(Mom) leaned out the window…’ (more literally: as for the window, (mom)
leaned out (it))
Janzen (2007) speculates that one possibility is that this represents a continu-
ing grammaticalization process, in that the structure of the utterance itself is
sufficient to identify it as topic-comment, and the facial markers represent a
level of redundancy that is increasingly unnecessary. This, however, has not
been studied in detail, although it is of interest in the present study, as some
corpus examples of KNOW- and UNDERSTAND-topics appear to show the
same characteristics.
Janzen (1998a, 1999) suggests that the grammaticalization pathway leading to
grammaticalized topics is:
generalized questioning gesture > yes/no question marking > topic marking
This pathway, then, follows the route discussed in Janzen and Shaffer (2002) and
Wilcox (2007) as not passing through a stage where a lexical word has devel-
oped from the source gesture, but rather the gesture has evolved more directly
into a grammatical marker. A generalized questioning gesture involving raised
eyebrows, often with an accompanying forward head-tilt, continues to be used
not only in ASL-signing communities, but more generally whether or not a yes/
no question is actually vocalized (think of a bartender making this gesture while
pointing to a patron’s empty glass in a noisy bar). In ASL this facial and head ges-
ture combination is the only marker of a yes/no question; word order between
a statement and a yes/no question is not altered. But in ASL topic marking,
while the eyebrow raise remains as a marker, the head-tilt is typically slightly
back rather than forward. The shift from a forward to a backward head-tilt may
signal that whereas a yes/no question is meant to be interactive – to invite a
response – the backward tilt suggests that the construction is more of a presen-
tation rather than a question requiring a response. As well, yes/no questions
frequently ­contain a great deal of shared information – the addressee must be
able to identify the question content in order to respond to it. Therefore even
though a topic-marked construction may appear strikingly like a yes/no ques-
tion in form, its function has changed in that it has grammaticalized to have an
information and discourse organizing function, losing the interactive function
that yes/no questions have.
In what follows, we look at instances where the ASL topic phrase includes
either KNOW or UNDERSTAND, such that the lexical verbs indicating a men-
tal state have grammaticalized as an additional topic marker in certain topic
constructions. On the surface, it may seem that topics that include KNOW or
 know and understand in ASL 

UNDERSTAND, most often positioned at the beginning of the topic phrase, make
the function of the construction ambiguous between a true topic and a yes/no
question. Indeed, there are times when the topic chosen by the signer, which is a
subjective discourse choice, does not clearly indicate identifiable information for
the addressee, and if the signer suspects as much, she may hedge on the choice of
topic, or negotiate the topic in terms of its status as identifiable information, and
in these cases, the construction may function somewhere between a topic and a
yes/no question. But there are many times when this is clearly not the case, when
the topic is not being negotiated. The present study is a step in the direction of
identifying instances of use in ASL conversational discourse, and looking further
into their form and their function.

4. Data

The data for this study were taken from a corpus of conversational ASL recorded
on video in 2000 at the University of Manitoba. The corpus consists of ten con-
versations, each between a study participant and a research assistant (RA). All
the participants and the RA were deaf members of the Winnipeg, Canada, deaf
community. All considered ASL to be their first language, had used it either
from birth or from an early age, reported that they used it every day as adults,
and were considered to be members of the deaf community by other members.
The RA was instructed to draw narrative stories from the study participants and
to try to let the participant dominate the conversation. In all cases the RA was
previously known to the participant socially. Each recording lasts for nearly one
hour, with a total of nine hours and nine minutes of conversational discourse in
the corpus.
The conversations were recorded simultaneously by three cameras, one
focusing on the study participant, one on the RA, and one camera capturing
both interlocutors in the frame. The recordings were made on videotape, and
were subsequently converted to digital files. These files were then annotated
using ELAN by ASL-fluent RAs. Critical for the present analysis were two lexical
tiers (which does not imply lexical or grammatical function in the annotations),
one for the right hand, which was the signer’s dominant hand in each case (all
participants were right-handed), and one for the left hand. To varying degrees,
all participants “switched” dominance from time to time to articulate signs or
phrases on the normally non-dominant hand, but this does not affect the present
study. Once annotated, ELAN has a searchable function so that each instance of
the strings KNOW and UNDERSTAND can be listed, time-coded, and analyzed
within its signed construction.
 Terry Janzen

5. Lexical know and know-topic constructions

KNOW ‘to know’ is a mental state verb in ASL articulated with the fingertips of
the dominant hand approaching and contacting the mid-forehead slightly to the
ipsilateral side of the midsagittal line on the body. That is, for a right-handed per-
son, the canonical contact point is just right of center, mid-­forehead. Verbs like
KNOW (and UNDERSTAND) making contact with the forehead is not likely very
arbitrary; when the processes of thinking and attaining knowledge are profiled in
ASL discourse, the movement of the sign toward the signer and the point of con-
tact for signs are closely associated with the upper half of the head.
However, Russell et al. (2011) show, based on a subset of the same corpus
used in the present study, that undershoot is a very regular phonological process
in signs that have the head and face as their canonical contact location: signs fre-
quently undershoot their targets. In the case of KNOW, this means that the forehead
is not always reached as the canonical contact point. Russell et al. (2011) analyzed
3075 such tokens of 229 sign types in six out of the ten conversational ASL record-
ings, concluding that while undershoot was significant, it could not convincingly
be attributed solely to ease of articulation. Ease of articulation has been claimed
in previous work on this phenomenon, especially when rapid and casual discourse
was being examined (e.g., Mauk 2003; Mauk and Tyrone 2008) but it appears that
signers do not reduce their reach up to higher canonical locations just because of the
excess effort involved. In rapid, casual signing, sign lowering often does take place,
but Russell et al. found many examples where a lower location, such as the cheek or
chin, appeared to be a deliberate target location. When this was the case, the sign was
stressed by a tense movement toward the location and contact was made by tapping
twice or more. Thus lowering in these instances did not represent a reduced target
approximation. Because of this, the tokens of KNOW and UNDERSTAND merit
a closer look. In this section, we examine KNOW tokens in the corpus as a whole.
In Figure 1, from Russell et al. (2011), KNOW tokens are mapped on a graph
that shows both those with actual contact (black circles) and those with move-
ment toward the head but without contact (clear circles). The intersection of the
dark lines at zero both vertically and horizontally represent the canonical con-
tact point on the signer’s forehead across all instantiations for six of the signers.
Directly below the intersection of these lines lies a space where no tokens show
contact, which indicates where the right eye is. The graph shows a cluster of con-
tact points near, but somewhat to the side of, the forehead, along with a trail that
extends down across the cheek and chin. The trail is loosely clustered on the cheek.
This graph begs the question: If location is part of the phonetic makeup of the
sign, does a contact point away from the canonical contact point correspond to a
­meaning or functional difference for the sign? This question is examined as part of
 know and understand in ASL 

the present study. First, however, we examine the distribution, categorization, and
functions of KNOW in the corpus.
KNOW

-50

-100

-60 -50 -40 -20 0 20 40

Figure 1. Scatter graph for know locations (Russell, Wilkinson & Janzen 2011).3

5.1 KNOW tokens across the ASL conversational corpus

Table 1 shows that there are 218 tokens of KNOW in the corpus altogether. The
largest category of tokens is lexical (125) but there are a large number (58) that
function as a discourse marker much the same as the phrase you know in Eng-
lish (Holmes 1986) where the signer is not talking about someone having specific
knowledge about something, but is using it as an interactional element or a filler.
There are also a total of 35 instances of KNOW appearing in a topic phrase, with

. All graphs from Russell et al. (2011) used with kind permission from John B
­ enjamins
Publishing Co., Laboratory Phonology (2:2).
 Terry Janzen

all but 2 occurring at the beginning of the phrase. In five of these cases, there is no
facial topic-marking, but the structure of the construction is consistent with topic-
comment structure and KNOW introduces the topic rather than having lexical or
discourse marker function. An additional 10 tokens of KNOW were annotated
but in these cases the function of KNOW is indeterminate, due either to there not
being enough clausal contextual information to make a determination of category,
or because its articulation is not clear. These cases are left out of the analysis, leav-
ing the total number of tokens analyzed at 218. Across all the remaining categories,
contact is made at or near the canonical forehead location in 112 cases, whereas
in 106 instances, the location is somewhere other than the forehead, primarily on
the upper, mid, or lower cheek, the chin, or in some cases out from the side of the
head and without actual contact.
What follows is first a discussion of the distribution of KNOW as a lexical
word, as a discourse marker, and in a KNOW-topic construction in the corpus and
second, an examination of the overall locational distribution of tokens.

Table 1. KNOW tokens in the conversational corpus (percentages in parentheses)

Forehead Non-forehead Totals

Lexical 86 39 125
(68.8) (31.2) (57.3)
Discourse Marker 11 47 58
(19.0) (81.0) (26.6)
KNOW-topic Marked 13 17 30* 35
(43.3) (56.7) (13.8) (16.1)
Non-marked 2 3 5
(40.0) (60.0) (2.3)
112 106 218
(51.4) (48.6)
* Numbers in this column are totals of marked and non-marked KNOW-topics; percentages in the
column combine to equal 16.1% of the total number of KNOW tokens in the corpus, shown in the last
column.

5.2 Lexical KNOW

KNOW is considered lexical if it is understood that it refers to someone knowing
something. In many cases the determination of KNOW as lexical is helped by
the inclusion of a subject pronoun either preceding the verb or following it, but
not always, since argument NPs and pronouns are not obligatory in ASL (Slobin
2006). In this corpus 125 out of 218 tokens (57.3%) are lexical, by far the largest
percentage of KNOW tokens. Examples (3–4) below show KNOW as lexical verbs.
 know and understand in ASL 

(3) PRO.3 KNOW BETTER

‘He/she knows better.’

(4) PRO.3 [KNOW FINISH]-nod

‘He/she knew (that).’

As these examples show, lexical KNOW identifies the mental state of someone
being discussed in the conversation. Figure 2 shows the prototypical contact point
location, mid-forehead and often just ipsilateral to the midsagittal line, that is, for
a right-handed signer, just to the right of the line that divides the right and left
halves of the body vertically.

a. b.

Figure 2. Lexical KNOW: (a) Long (1918) dictionary; (b) present corpus, dominant hand at
canonical contact point on the forehead (left hand shows a simultaneous pronominal point).

Figures (2a) and (2b) show that articulation of KNOW has remained stable
over the almost one-hundred year span from 1918 to the 2000s, although it
might be noted that in Long’s (1918) dictionary there is just one photograph
and there is no discussion or indication of locational variation in Long’s writ-
ten entry for KNOW.
As mentioned, if subject and object NPs or pronouns are overt, they may help
to identify the token as lexical, but they are not obligatory. When they are present,
they may precede or sometimes follow the verb. Figure (2b) shows that addition-
ally, because KNOW is a one-handed sign, a subject pronoun may be articulated
simultaneously alongside the verb. Such simultaneity occurs frequently in signed
languages (Vermeerbergen, Leeson & Crasborn 2007), and while it presents issues
 Terry Janzen

for the study of word order, it does not interfere with the function or form of the
verb itself.4

5.3  KNOW as a discourse marker

Phrases such as I know and you know frequently function as discourse markers
in English. Bybee (2010: 223, f3) notes that in the Switchboard corpus data she
analyzes, fifteen percent of all finite verbs used are think and know, and these func-
tion as discourse markers. Know can also function as a hedge (Lakoff 1975), but
Holmes (1986) suggests that this is not always the case, that you know can function
to give intratextual coherence to someone’s discourse in English, as a verbal filler,
or reflecting the speaker’s “relevant knowledge of the addressee in the context of
utterance” (1986: 16).
In the ASL corpus, KNOW as a discourse marker occurs 58 times out of
218 tokens, or 26.6% of the time. This then is the second largest type category.
In most cases the discourse marker KNOW occurred without any nominals,
that is, neither a subject nor an object, but seemed to equate to the commonly
used you know in English. This is not too surprising given Slobin’s (2006) anal-
ysis of ASL as a head-marked language, meaning that nominal or agreement
markers would appear on the verb as head, such that overt nominals in the
clause are not required unless there is some pragmatic reason to include them.
However, some verbs including KNOW may be classed as “plain verbs” (Pad-
den 1988) that do not take agreement marking, yet pattern in the same way.
Table 2 shows that this is the case for 65.5% of KNOW tokens as discourse
markers. Ten (17.2%) tokens occurred with a subject or object nominal while
another 17.2% were followed by a closely connected clausal complement. In
all these cases KNOW did not appear to function as a fully mental state verb
in a proposition or assertion that someone knows something, but rather as a
marker in the discourse to draw attention to something or to check in with the
addressee to make sure they were following.

. Simultaneity may be a factor in some instances of grammaticalization, however. Janzen

(2012b) gives the example of an existential instance of HAVE in ASL (grammaticalized from
possessive HAVE) that has reduced in phonetic form from a two-handed to a one-handed
sign, which then leaves the other hand available for simultaneous articulation of other ma-
terial. In Janzen’s example, the signer signed HEARING, DEAF with her right hand while
simultaneously signing a very reduced form of HAVE with her left hand, meaning ‘there are
both hearing and deaf people who…’
 know and understand in ASL 

Table 2. KNOW as discourse marker tokens (percentages in parentheses)

subject/object nominal complement clause without nominals Total

10 10 38 58
(17.2) (17.2) (65.5)

KNOW as discourse markers are shown in (5) and (6).

(5) CL:small.amounta CL:large.amountb // KNOW
‘You put a little money in one and more money in the other, you know?’
(6) KNOW WANT MORE TRAVEL
‘You know, you want to travel more.’
KNOW having a discourse marker function has clearly grammaticalized from the
lexical verb KNOW and occurs as one-quarter of all tokens of KNOW in the con-
versational corpus this study is based upon, but it will not be discussed further
here. Rather, the main focus of this study concerns topic constructions, and it is
to this we turn next.

5.4 KNOW as a topic marker

KNOW as a topic marker in a topic phrase constitutes 16.1% of all KNOW tokens.
In this corpus there are a few tokens of KNOW that occur in yes/no questions
directed at the addressee, and one as part of a topic construction that functions as
a mental state verb. These were categorized as lexical rather than as KNOW-topics.
The single topic phrase containing lexical KNOW is given in (7).
(7) BUT [KNOW RULE]-top NOT.YET
‘But does he know the rules? Not yet/but as for knowing the rules, he
doesn’t yet.’
In (7) the signer is making an assumption that the idea of knowing the rules is
given information, or a given state that he subsequently evaluates for the referent
they are talking about. In any case, this example does not qualify as a KNOW-topic
with grammatical function because it functions as a lexical mental state verb.
For most topic constructions in connected discourse in ASL, the markers
of the phrase as grammatical topic, as discussed above, are raised eyebrows and
potentially a slight backward head tilt that is maintained throughout the sign-
er’s articulation of the topic phrase. This has been found to be the case in both
monologic discourse (Janzen 1998a, 1999) and in conversation as shown in the
present study. As Table 1 above shows, 16.1% of KNOW tokens appeared not as
lexical verbs, but as the first element in a topic phrase. KNOW-topics do not seem
to appear frequently in ASL discourse, and the addition of KNOW in the topic
 Terry Janzen

phrase indicates a particular additional function, as shown in Table 3. Mainly, the

shared or identifiable information in KNOW topics tends to be a generality, a type
of frame, or something generic, thus labeled in Table 3 as KNOW-topics about
some domain. Twenty-four out of 35 (68.6%) of KNOW-topics in the corpus are
of this type. Examples of this type of KNOW-topics are given below, but first we
outline the other KNOW-topic categories in Table 3.

Table 3. KNOW-topics (percentages in parentheses)5

domain negotiation topic/discourse marker5 Total

24 8 3 35
(68.6) (22.8) (8.6)

A second function appears to be topic negotiation (8 out of 35 or 22.8%), at least in

part. Some of the evidence that Janzen (1998a, 1999) and his colleagues (e.g., Jan-
zen & Shaffer 2002; Janzen, Shaffer & Wilcox 1999) discuss for topic constructions
to have grammaticalized from yes/no question constructions is that yes/no ques-
tions and topics have similar formational characteristics. Yes/no questions are also
marked by an eyebrow raise, for example. One difference between the two, however,
is while yes/no questions tend to have a forward head tilt, which is a gesture inviting
a response, topics tend to have a backward head tilt suggesting that such interaction
is not expected or warranted. In other words, the topic has become regularized as a
sentence element. A second piece of evidence might be that yes/no question content
is itself frequently highly topical – in order for an addressee to evaluate the question
content (and respond yes or no), that content must be familiar or identifiable.
But what if the questioner is unsure whether the topic content is in fact identi-
fiable? An obvious discourse strategy would be to check with the addressee. When
this takes place in ASL, there appears to be some movement back toward the inter-
active sense of a yes/no question and frequently this takes the form of a question
that asks do you know x? So, there are two essential differences between domain
content and negotiating KNOW-topics. First, the sense of negotiation function is
higher in those KNOW-topics categorized as negotiating and lower for domain
content topics. Second, negotiating KNOW-topics tend to include a nominal that
is individuated and referential whereas those in the domain content category
are generic and non-referential. For a negotiated topic, the signer assumes the
addressee has knowledge concerning a specific entity as in (8).

. In three cases it was unclear whether KNOW was functioning as part of a topic phrase or
as a discourse marker. These instances are not included in the discussion below.
 know and understand in ASL 

(8) [KNOW CL:wide.tube SMOKE]-top CL:wide.tube.falling

‘You know that chimney? It fell down.’
In (8) the signer is telling her interlocutor, who she knows well, about a chimney
on an outdoor brick grill at a place that they have both visited, and here is checking
to make sure that they both have the same chimney in mind.
Negotiation seems to have somewhat less importance in domain content
KNOW-topics. The structure of the topic phrase is quite similar, however, so it may
be difficult to say that a negotiating function has disappeared entirely. The main
thrust of these topics seems to be to present a more global domain that is assumed to
be identifiable because of more general experience with that thing, so the question
in these may be less like do you know x? and more like you know what (an) x is like.
Examples are (9–12).
(9) [KNOW WIND CL:5(both hands)wind.blowing]-top BEAR FREQUENT
SMELL
‘Given what wind is like, a bear can frequently smell (something).’
(10) [KNOW TALK DREAM]-top BUT ACTION
‘You know how dreams are often big talk, but it’s actions (that are important).’
(11) [KNOW A-P-R-I-L]-top pause NOTHING WHITE BEAR NOTHING
‘Given what April [i.e., spring time] is like, it’s easy for polar bears, easy.’
(12) [KNOW R-O-A-D HIGHWAY]-top off.to.sidert [NEG on.side.road]-neg.
nod CL:3(vehicle).move.overrt FINISH pause
‘Given that you know what highways are like, (we) moved off to the side,
not onto a side road but just to the side of the road.’
Each of topic phrases in these examples evokes an experiential domain that the
signer can assume the addressee would know. Here the term “domain” is chosen
to align with Langacker’s (2008) sense of generality, as opposed to Filmore’s (1982)
“frames” which may evoke specific events and their characteristics. It may turn
out, however, that frames are applicable here, but for the purposes of this study, we
will remain consistent with Langacker’s term.6
In (9) the domain is the idea of wind, and within this domain, we might gen-
erally know that the wind blows frequently, that being upwind or downwind can
mean very different things, for example with smoke from a fire pit, and impor-
tantly here, how the wind carries scents. In (10) the idea of dreams (as in wish-
ing for something in the future) and talking about dreams evokes the common
experience of a lot of talk (and no action), and here once again it is not a specific
dream being referred to, but the domain more generally. The topic in (11) evokes

. I thank Jila Ghomeshi for critical discussion on frames and domains in the context of
this study.
 Terry Janzen

the domain of springtime with the reference to the month of April and not just
the month itself, and not to a specific month in a particular year. And in (12) the
road is taken to indicate physical characteristics of highways along with relational
information – that highways have lanes in two directions and shoulders and cross-
roads; here the topic functions as a basic domain with which to understand what
the action described in the comment signifies.
All in all these topic tokens constitute a particular category of topics that min-
imally signal some degree of negotiation but tend to evoke a domain rather than
something specific or referential. Three cases (8.6%) are evidently more of a hybrid
of these two characteristics. As an example of a KNOW-topic that appears to have
more of a negotiation function, consider (13):
(13) [WHITE S-H-E-L-L KNOW CL:5(claw.location)+++]-top DETEST7
‘In the Whiteshell you know how there are so many cottages, I hate that.’

In (13) the signer has just previously said that she likes a cottage area that is more
sparsely populated, and here compares an area called the Whiteshell Provincial
Park where the population is denser. This location is topic-marked as ostensibly
recognizable information pertaining mostly to the characteristic of density – she
is checking with her interlocutor as to whether she is familiar with how dense the
area is. This topic phrase too points to characteristics of a specific location rather
than a more abstract domain.
A final point to mention regarding the categorization of KNOW-topics is, as is
common for language categories generally, that they must be considered having both
prototypical tokens as well as tokens that are positioned more toward the peripheries
of their categories, that there is overlap between categories, and that the categories
themselves have fuzzy edges (Givón 1986; see also Bybee & Eddington 2006). This
is, in essence, another way of saying that with actual interactive discourse as data, a
great deal of variation is evident. Pragmatic effects such as topic negotiation are sca-
lar, reflecting the intersubjective nature of discourse, and when applied across actual
instances of use, in saying that something has “domain” content we must also bear
in mind that some instances may much better fit what we think of as a domain than
others. A parallel case, understood along these lines, is that of transitivity (Hopper &
Thompson 1980; Thompson & Hopper 2001), where it is argued that transitivity is
not only a property of the verb, but is distributed over numerous clause elements, the

. The fact that KNOW is not in an initial position here bears mention but when and why a
marker such as KNOW is topic medial (or in a small number of cases, in a final position) is left
for future analysis. In the case of (13) the topic phrase may actually consist of two intonational
phrases, first the name of the area as shared knowledge of a specific location and second, the
negotiation of knowledge of the specific density of housing there.
 know and understand in ASL 

result of which positions instances of use on a low-high transitivity continuum. In

the same way it is not just the presence of a single element such as KNOW, or of topic
markers themselves, that determines category fit, but a combination of structural,
semantic and pragmatic variables that help determine category fit.

5.5 Location variation in KNOW tokens

Having looked at the category distribution of KNOW in Section 5.4 above, we next
focus on the variation that exists among KNOW tokens in terms of the location of
their articulation. In Russell et al.’s (2011) study on undershoot in ASL sign con-
tact and location points as canonical targets, it is evident that for many tokens of
KNOW, the canonical contact target of mid-forehead is not reached. As suggested
at the beginning of Section 5 above, here we are interested in examining whether
a contact point or location away from the canonical mid-forehead location corre-
sponds to a change in meaning or function for KNOW. This question is prompted
in part because the graph in Figure 1 shows that while there is a distinct cluster of
tokens located at and near the mid-forehead location, there is also a trail of lowered
tokens that form a loose cluster around the cheek and chin areas. Table 1 shows that
when KNOW is lexical, 86 out of 125 (68.8%) instances of use contact the forehead
area while 39 (31.2%) are located elsewhere, mostly, as Figure 1 shows, at a low-
ered location. In contrast to this, 11 out of 58 (19.0%) of discourse marker KNOW
tokens are located at the forehead while 47 (81.0%) are not. For KNOW-topics, 15
out of 35 (combined marked and non-marked instances; 42.9%) have a forehead
location and 20 (57.1%) do not. Figure 3 shows the non-canonical chin-level loca-
tion point for one such token taken from the category of discourse marker.

Figure 3. KNOW as discourse marker articulated in a non-canonical location, in this case

articulated with the left hand.
 Terry Janzen

Clearly there is a great deal of variation in where tokens of KNOW do make

contact, but these numbers suggest that the differences are not random. Overall,
KNOW tokens show an almost equal distribution at, or away from, the forehead
(51.4% and 48.6% respectively), but when categorized as lexical, discourse marker,
and KNOW-topic, some distinct differences can be seen. Because lexical KNOW
is a mental state verb, an iconicity principle may be at work regarding location
choice: the sign is more likely to iconically index the brain – where knowledge is
held. Sixty-eight percent of verbal KNOW instantiations have contact with the
forehead as opposed to 31.2% which do not, which contrasts with KNOW as a
discourse marker, where 19% of usages contact the forehead and 81% do not,
and as a topic-phrase element where 41.7% contact the forehead and 58.4% do
not. As use becomes more grammatical, both in the context of discourse marker
and topic marker function, the lexical sense of knowledge loses significance, such
that the iconic tie to the forehead is loosened, and lowering is more likely to
take place. This appears to be more evident for discourse marker KNOW than
it is for KNOW-topics, which might indicate that KNOW-topics are still in an
intermediate position and are more likely to evoke some notion of knowledge.
Here it is not assumed that there is a grammaticalization pathway from lexical
KNOW to topic-marking KNOW to discourse marker KNOW, but rather that
lexical KNOW has grammaticalized in two directions. This is complicated by the
fact, however, that discourse markers generally can and often do occur with topic
marking, which Janzen (1998a, 1999, 2007) suggests indicates that they function
as shared discourse structure or textual knowledge, following Traugott (1989),
and thus represent a high degree of grammaticalization.
It is not known whether the looser scatter of tokens around the cheek and chin
area is indicative of some process either in terms of location lowering or grammat-
icalization, but scalarity and gradience, especially regarding location in ASL, have
been noted elsewhere (e.g., Slobin 2008). Perhaps here the mental state semantics
of lexical KNOW are more precisely definable as is the physical location of that
knowledge, whereas as this diminishes in more grammaticalized constructions in
which a looser connection to a specific location represents a kind of emerging
iconicity of a different sort (see Wilcox 2004 for similar discussion regarding lexi-
calization and usage).

6. Lexical understand and understand-topic constructions

UNDERSTAND occurs in the corpus in some constructions that are similar to

those for KNOW, but not in all ways. Table 4 shows the distribution of UNDER-
STAND according to function (lexical or topic), and location (forehead or
 know and understand in ASL 

e­ lsewhere). There are 61 tokens of UNDERSTAND altogether, with 20 (32.8%)

occurring as lexical verbs and 41 (67.2%) in topic constructions. Three additional
tokens were annotated but their status is not clear so they are left out of the analysis
here. No UNDERSTAND tokens in this corpus functioned as a discourse marker.

Table 4. UNDERSTAND (percentages in parentheses)

Forehead Non-forehead Totals

Lexical 5 15 20
(25.0%) (75.0) (32.8)
UNDERSTAND- phrasal 8 24 32* 41
topic (25.0) (75.0) (52.5) (67.2)
alone 4 5 9
(44.4) (55.6) (14.8)
17 44 61
(27.9) (72.1)
* Phrasal and alone topics combine to give a total of 41; percentages equal 67.2% of the total number of
tokens of UNDERSTAND in the corpus. Out of the 41 topic constructions, 32 (78%) are phrasal and in 9
(22%) UNDERSTAND occurs alone with topic marking.

Table 4 reveals that UNDERSTAND occurs at a substantially higher rater (67.2%)

in a grammatical context than as a lexical verb (32.8%), and that in topic construc-
tions it occurs most of the time (78.0%) as an element in a longer topic phrase and
substantially less as a topic-marked element alone (22.0%). In terms of location
and contact of the hand with the forehead, Stokoe, Casterline and Croneberg’s
(1965) dictionary of ASL lists UNDERSTAND as having contact with the fore-
head with an optional movement away from the head as the index finger extends.
Janzen (1998b) suggests that the change-of-state nature of UNDERSTAND is
iconically represented in its phonological form. The canonical sign has no inher-
ent movement but rather an opening aperture of the hand (Brentari 1998), chang-
ing from a closed fist with the palm facing the signer’s forehead to an extended
index finger. The closed handshape metaphorically represents the initial state of
not yet understanding, and the extended index finger metaphorically represents
the final state of an emerging (understood) idea, which corresponds to Wilcox’s
(2000) notion of the extended finger as an idea metaphor.
However, in this corpus only one lexical UNDERSTAND token made clear
contact with the signer’s forehead, and as shown in Table 4 the majority of tokens
(72.1%) were not articulated at the forehead. Those that were at the forehead were
recorded as such because the forehead is the location closest to where these signs
occurred and yet they occurred not with contact, but out in front of the fore-
head. Figure (4a) is UNDERSTAND taken from Long (1918) and 4(b) shows
 Terry Janzen

­ NDERSTAND as sole lexical verb in this corpus with canonical contact at the
U
forehead location (before the index finger is extended). However, as mentioned,
no other token made contact with the forehead, and many instantiations were
articulated nowhere near the signer’s head. Figure 5 shows two such tokens. In Fig-
ure (5a) we see a topic construction UNDERSTAND token articulated at shoulder
height, and the most extreme articulation in this sense is in 5(b), which is a lexical
verb in the construction meaning ‘I understand’, but with no overt pronominal.

a. b.

  
Figure 4. (a) UNDERSTAND from Long (1918) – an arrow up along the forehead indicates
the direction of the index finger extension; (b) canonical forehead location and contact.

a. b.

Figure 5. two non-forehead instantiations of UNDERTAND, (a) in a topic construction and

(b) as a lexical verb.

These examples show much greater locational variation than KNOW has, which
means that, if the iconicity analysis for KNOW above is correct, the same iconic
alignment to the semantics of UNDERSTAND does not hold. Table 4 indicates
that whether or not UNDERSTAND has lexical or grammatical (topic) meaning,
it is much more likely to be articulated away from the forehead, and as Figure (5b)
 know and understand in ASL 

illustrates, some of the most extreme examples of undershoot are lexical. Why
these two mental state verbs in ASL behave so differently with respect to iconicity,
however, is left to further study (but see Section 7 below).

6.1 Lexical UNDERSTAND

As mentioned, lexical UNDERSTAND is a change-of-state type of mental verb.
Lexical examples of UNDERSTAND are given in (14) and (15). In (14) there is
no overt subject nominal, but there is an explicit complement verb. The subject
pronoun in (15) is expressed overtly.
(14) [FOOD]-top [UNDERSTAND]-nod/emph TRUE
‘(The dog) fully understands feeding time.’
(15) PRO.3 NEG UNDERSTAND
‘He doesn’t understand.’

6.2 UNDERSTAND-topics
The vast majority of UNDERSTAND tokens in the corpus appear in topic-marked
constructions of which there are two sorts. First, most UNDERSTAND-topics are
phrasal (78.0% of all UNDERSTAND-topics), with UNDERSTAND in the ini-
tial position. The function of UNDERSTAND-topics is not quite as clear as that
of KNOW-topics, but in this corpus it appears that they tend to introduce some
information that may not be readily identifiable to the addressee but which the
signer wishes the addressee to accept as a given state of affairs, or as an agreed
upon reference point from which to view the comment. In this sense they are a
type of topic negotiation marker along with, but with a different negotiating func-
tion, KNOW-topics. Examples are given in (16–18).
UNDERSTAND SWIM]-top gesture:raise.hand pause NEG PERMIT
(16)
[
gesture:two.hands.on.ball
‘Given when you’re swimming, you are not permitted to have both hands on
the ball.’
(17) [UNDERSTAND PRO.1 SIGN]-top NEG TALK
‘It’s given that I would be signing, not speaking.’
(18) [UNDERSTAND 20 DOLLAR]-top PRO.1 DON’T.MIND PAY PRO.3 20
DOLLAR
‘Given the idea of paying twenty dollars, I don’t mind paying him the
twenty dollars.
In each of these cases, the signer introduces an idea or state of affairs in the marked
topic phrase that might be partially identifiable from the general topic of discourse
 Terry Janzen

or which the signer might expect to be acceptable as an identifiable ground or ref-

erence point from which to understand the significance of the comment. Similar
to KNOW-topics, these UNDERSTAND-topics are often more abstract or generic
instead of referential. Additionally, they are not identifiable from previous refer-
ence in the discourse event, and nor can they be fully recoverable from pragmatic
knowledge or experience, but the signer does not appear to wish to negotiate the
topic. Rather, the topic is presented as if, even if not readily identifiable to the
addressee, as a given.
The second type of UNDERSTAND-topic is where UNDERSTAND alone is
topic-marked. Examples are given in (19) and (20).
(19) [UNDERSTAND]-top EVERYDAY CLEAN SCOOP.OUT BATHROOM
‘Every day you have to clean out the litterbox.’
(20) [UNDERSTAND]-top NEG PERMIT CAMERA NEG PERMIT
‘(They) don’t permit photos, they don’t.’

There is no analogous construction in English, so the translations of (19) and (20)

do not reflect how the UNDERSTAND-topic might be operating. Here it appears
that topic-marked UNDERSTAND may function much like other connectives in
ASL such as BUT and FINISH (meaning ‘and then’) that Janzen (1998a, 1999,
2007) notes also can be topic-marked, and which primarily have textual identifi-
ability, acting as a sort of pivot connecting something previous to something com-
ing up in a textually predictable way. All topic phrases have the textual function
of pointing ahead: the focus of the utterance is not the topic, but is information
contained in the comment. The topic, in this sense, is a grammaticalized means to
get to the comment information – it is backdrop, or grounding information that
situates, and in some cases contextualizes, the comment. When connectives do
this, they are not informational in any way, but are textual “sign posts” that give the
addressee a direction as to how to read the comment that is coming up. As such,
topic-marked UNDERSTAND is an intersubjective grammatical mechanism that
points ahead. However, at this point we are best able to suggest a preliminary anal-
ysis for tokens of topic-marked UNDERSTAND, as a larger scale study is needed
to gain a clearer picture of how they function overall.

7. The emergence of categories in ASL grammar

In this study we have seen that both lexical verbs KNOW and UNDERSTAND
have grammaticalized as topic marking elements for certain types of topics in ASL.
As each of the two verbs has grammaticalized, such that new categories of mean-
ing and usage have emerged, there remains a substantial degree of variation both
 know and understand in ASL 

in form and function. But these are not isolated cases, so it might be helpful to
consider how the situation for these items fits within a larger picture of usage and
variation.
Variation in scatter density around a canonical location is attested not only
between the items KNOW and UNDERSTAND. To compare, Russell et al.
(2011) report on the ASL preposition FOR and the wh-question word WHY.
That these two items are not in the same category as are the verbs KNOW
and UNDERSTAND does not bear on the present discussion; what is critical,
rather, is the difference in location variance between the two. FOR and WHY
also have a canonical location at the forehead, ipsilateral to the dominant hand,
slightly to the right of the eye for a right-handed signer, identical to the loca-
tion for KNOW and UNDERSTAND. As can be seen in Figure (6a), the scatter
of instantiations of FOR is fairly tight, although with some instances of use
trailing below, with a cluster on the upper cheek. This is somewhat similar to
what we see with KNOW, but the details of FOR in terms of usage and meaning
have not yet been worked out. WHY, shown in 6(b), on the other hand, has no
similar clustering, although there appears to be a slight concentration at what
corresponds to the very center of the face.
These facts illustrate several important theoretical claims. First, these four
words have the pronunciation feature of mid-ipsilateral forehead as canonical
location, but the variation in instantiation location among the four in this dis-
course corpus shows that it is item-based (Bybee 2013): change in location across
instantiations of forehead-based words does not take place similarly for all words.
Second, the variation in actual location as these words are used in ASL discourse
cannot be accounted for simply by an ease of articulation leading to sign lowering
explanation. If this were the case, then we might again expect there to be more
similarity in the overall result. As well, as Russell et al. (2011) claim, such an expla-
nation cannot account for those instantiations that are raised instead of lowered.
Finally, in the case of KNOW and FOR, but not for UNDERSTAND and WHY,
some lower clustering does take place. For KNOW at least, there is evidence that
this lower cluster represents the emergence of a new category with a distinct gram-
matical, rather than lexical, function, and along with it, the loss of mental verbal
meaning.
But why should there be such a difference in the scatter of instantiation loca-
tions between KNOW and UNDERSTAND? One possibility is that while these two
words share the same location feature, other aspects of their articulation differ sig-
nificantly. Canonical KNOW is articulated with a movement toward the forehead
such that its final segment constitutes a resting position at, and in contact with, the
forehead. Figure 1 above in Section 5 for KNOW shows that in the canonical loca-
tion contact is nearly one-hundred percent, represented by black dots as opposed
 Terry Janzen

a. FOR

-50

-100

-80 -60 -40 -20 0 20 40

b. WHY

-50

-100

-80 -60 -40 -20 0 20 40

Figure 6. Scatter graphs for (a) FOR and (b) WHY from Russell et al. (2011).

to clear dots which indicate a lack of contact. Even when an alternate location
from the forehead is the case for KNOW, and when a lack of contact with the face
is possible, the tendency for contact is still very high. UNDERSTAND, however,
has no movement toward or away from the contact point. Instead it has a local
movement (Brentari 1998) from a beginning ‘S’ (closed fist) handshape to a final
 know and understand in ASL 

‘1’ (extended index finger) handshape. Thus the movement in UNDERSTAND is

not dependent on movement toward or away from a location, which may make it
more conducive to articulation in a wider spatial sphere.

8. Conclusion

The present study examines KNOW- and UNDERSTAND-topics from a gram-

maticalization perspective in an attempt to better understand more about their
function and their form. Location variation in undershot location as opposed to
canonical forehead location appears to be a factor, but it must be considered a
tendency, as instances of undershoot in grammaticalized constructions are widely
apparent, especially for KNOW. However, widespread undershoot occurs with
lexical tokens as well, and some instances of both grammaticalized KNOW and
UNDERSTAND maintained a forehead location.
Several questions remain. For one, topic constructions in ASL overall have not
been studied in a corpus-based conversational context, and while the present study
makes some claims regarding domain-type and individuated and referential con-
tent in these topic constructions, we do not yet know how these categories compare
with topics that do not contain KNOW or UNDERSTAND within the topic con-
struction, and it is presumed that KNOW- and UNDERSTAND-topics are not the
most frequently occurring categories of topics. As well, the study of topic type and
topic frequency across various genres in ASL is far from complete, and frequency
may well be a factor in how these categories behave. In this corpus study, KNOW
occurs more frequently (218 tokens) than UNDERSTAND (61 tokens).
What is evident in the present study is that a significant number of KNOW
and UNDERSTAND tokens have grammatical, rather than lexical, function. The
largest number of KNOW tokens in this corpus are lexical, but 26.6% function
as discourse markers and 16.1% are KNOW-topics. On the other hand, there
are no instances of use of UNDERSTAND as discourse markers, and 67.2% are
UNDERSTAND-topics. Topics introduced by both KNOW and UNDERSTAND
both can serve the discourse purpose of topic negotiation, but more so, signers
use them to go beyond topic-marking specific and referential nominal mentions.
KNOW-topics tend to set a domain as a reference point, such as knowledge of the
effects of wind or the characteristics of spring, as examples in Section 5.4 above
show. UNDERSTAND-topic function is somewhat less clear, although a larger
study may reveal more. Nonetheless, in this study we can see that in these topics
the information is not necessarily shared, but rather the signer introduces an idea
or a possible state of affairs that she wishes to set as a ground or reference point,
even if the topic-marked statement is hypothetical.
 Terry Janzen

With the inclusion of either KNOW or UNDERSTAND to introduce the

topic phrase, it seems clear that the signer is signaling that the topic is somewhat
out of the ordinary; the majority of topic phrases in ASL are not marked in this
way. While this study adds to our understanding of topic and topic-marking in
ASL, it also points to the fact that topic is not at all a simple category, but one
where structural, semantic and pragmatic factors in each instance of use combine
intersubjectively to produce shared meaning among discourse participants.

Transcription key

Upper case word glosses indicate ASL signs. Words separated by a period (e.g., MOVE.OVER)
indicate that more than one English word is used to denote a single ASL sign. Letters separated
by hyphens (e.g., C-A-R) represent fingerspelling. Plus signs (++) denote repeated movement.
Square brackets indicate that a facial gesture is maintained throughout the phrase enclosed.
emph indicates emphasis. […]-top represents topic marking. Subscript letters represent spatial
locations associated with entities positioned in the space around the signer, and are labeled ‘a’,
‘b’, etc., arbitrarily. Subscript text indicates a ‘path’ of locations. Subscript 1 represents a location
near the signer. PRO.1, PRO.2, and PRO.3 are first, second and third person singular pronouns.
Classifier constructions are CL: plus a handshape label and/or a description which is italicized
and in parentheses. Additional italicized words in the transcriptions indicate pauses and ges-
tures between signed phrases. The translation line is an English approximate that does not nec-
essarily represent equivalent grammatical features or lexical categories to those found in ASL.

References

Baker, Charlotte & Cokely, Dennis. 1980. American Sign Language: A Teacher’s Resource Text on
Grammar and Culture. Silver Spring MD: T.J. Publishers.
Brentari, Diane. 1998. A Prosodic Model of Sign Language Phonology. Cambridge MA: The MIT
Press.
Brouland, Joséphine. 1855. Spécimen d`un Dictionaire des Signes, suivi d’Explication du Tableau
Spécimen d`un dictionaire des signes du langage mimique, mettant toute personne en état de
l’apprendre seule. Paris: Boucquin, Imprimerie de l’Institution Nationale des Sourds-Muets
de Paris.
Bybee, Joan. 2001. Phonology and Language Use. Cambridge: CUP. doi: 10.1017/CBO9780511612886
Bybee, Joan. 2006. From usage to grammar: The mind’s response to repetition. Language 82(4):
711–733. doi: 10.1353/lan.2006.0186
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan L. 2013. Usage-based theory and exemplar representation. In The Oxford Handbook
of Construction Grammar, Thomas Hoffman & Graeme Trousdale (eds), 49–69. Oxford:
OUP.
 know and understand in ASL 

Bybee, Joan & Eddington, David. 2006. A usage-based approach to Spanish verbs of ‘becoming’.
Language 82(2): 323–355. doi: 10.1353/lan.2006.0081
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar: Tense,
Aspect and Modality in the Languages of the World. Chicago IL: University of Chicago Press.
Bybee, Joan & Scheibman, Joanne. 1999. The effect of usage on degrees of constituency: The
reduction of don’t in English. Linguistics 37(4): 575–596. doi: 10.1515/ling.37.4.575
Filmore, Charles J. 1982. Frame semantics. In Linguistics in the Morning Calm, Linguistic Society
of Korea (ed.), 111–137. Seoul: Hanshin.
Frishberg, Nancy. 1975. Arbitrariness and iconicity: Historical change in American Sign Lan-
guage. Language 51: 696–719. doi: 10.2307/412894
Givón, T. 1984. Syntax: A Functional-Typological Introduction, Vol. 1. Amsterdam: John
Benjamins. doi: 10.1075/z.17
Givón, T. 1986. Prototypes: Between Plato and Wittgenstein. In Noun Classes and Categoriza-
tion [Typological Studies in Language 7], Colette Craig (ed.), 77–102. Amsterdam: John
Benjamins. doi: 10.1075/tsl.7.07giv
Haiman, John. 1978. Conditionals are topics. Language 59: 781–819. doi: 10.2307/413373
Heine, Bernd & Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
doi: 10.1017/CBO9780511613463
Heine, Bernd & Kuteva, Tania. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP.
Holmes, Janet. 1986. Functions of you know in women’s and men’s speech. Language and Society
15(1): 1–21. doi: 10.1017/S0047404500011623
Hopper, Paul. 1991. On some principles of grammaticization. In Approaches to Grammatical-
ization, Vol. I: Focus on Theoretical and Methodological Issues, Elizabeth Closs Traugott &
Bernd Heine (eds), 149–187. Amsterdam: John Benjamins. doi: 10.1075/tsl.19.1.04hop
Hopper, Paul J. & Thompson, Sandra A. 1980. Transitivity in grammar and discourse. Language
56: 251–299. doi: 10.1353/lan.1980.0017
Janzen, Terry. 1995. The Polygrammaticalization of FINISH in ASL. MA thesis, University of
Manitoba, Winnipeg.
Janzen, Terry. 1998a. Topicality in ASL: Information Ordering, Constituent Structure, and the
Function of Topic Marking. PhD dissertation, University of New Mexico, Albuquerque
NM.
Janzen, Terry. 1998b. Multi-level iconic relationships in American Sign Language grammar. In
Proceedings of the First High Desert Linguistics Society Conference, April 3–4, 1998, Vol. 1,
Catie Berkenfield, Dawn Nordquist & Angus Grieve-Smith (eds), 159–172. Albuquerque
NM: High Desert Linguistics Society.
Janzen, Terry. 1999. The grammaticization of topics in American Sign Language. Studies in
­Language 23(2): 271–306. doi: 10.1075/sl.23.2.03jan
Janzen, Terry. 2007. The expression of grammatical categories in signed languages. In Verbal
and Signed Languages: Comparing Structures, Constructs and Methodologies, Elena Pizzuto,
Paola Pietrandrea & Raffaele Simone (eds), 171–197. Berlin: Mouton de Gruyter.
Janzen, Terry. 2010. Pragmatics as start point; Discourse as end point. Keynote Presentation.
9th Conference of the High Desert Linguistic Society (HDLS), University of New Mexico,
Albuquerque NM.
Janzen, Terry. 2012a. Lexicalization and grammaticalization. In Sign Language: An International
Handbook [Handbooks of Linguistics and Communication Sciences (HSK) series], Markus
Steinbach, Roland Pfau & Bencie Woll (eds), 816–841. Berlin: Mouton de Gruyter.
doi: 10.1515/9783110261325.816
 Terry Janzen

Janzen, Terry. 2012b. From embodied perspective to grammar in American Sign Language.
Keynote Presentation, Conceptual Structure, Discourse and Language (CSDL) 11. Univer-
sity of British Columbia, Vancouver, Canada, 17–20 May.
Janzen, Terry & Shaffer, Barbara. 2002. Gesture as the substrate in the process of ASL gram-
maticization. In Modality and Structure in Signed and Spoken Languages, Richard P. Meier,
Kearsy Cormier & David Quinto-Pozos (eds), 199–223. Cambridge: CUP.
doi: 10.1017/CBO9780511486777.010
Janzen, Terry, Shaffer, Barbara & Wilcox, Sherman. 1999. Signed language pragmatics. In Hand-
book of Pragmatics, Installment 1999, Jef Verschueren, Jan-Ola Östman, Jan Blommaert &
Chris Bulcaen (eds), 1–20. Amsterdam: John Benjamins.
Lakoff, Robin. 1975. Language and Women’s Place. New York NY: Harper Colophone.
Langacker, Ronald W. 2008. Cognitive Grammar: A Basic Introduction. Oxford: OUP.
doi: 10.1093/acprof:oso/9780195331967.001.0001
Leeson, Lorraine & Grehan, Carmel. 2004. To the lexicon and beyond: The effect of gender on
variation in Irish Sign Language. In To the Lexicon and Beyond: Sociolinguistics in European
Deaf Communities, Mieke Van Herreweghe & Myriam Vermeerbergen (eds), 39–73. Wash-
ington DC: Gallaudet University Press.
Li, Charles N. & Thompson, Sandra A. 1976. Subject and topic: A new typology of language. In
Subject and Topic, Charles N. Li (ed.), 457–490. New York NY: Academic Press.
Long, J. Schuyler. 1918. The Sign Language: A Manual of Signs. Omaha NE: Dorothy Long
Thompson.
Mauk, Claude E. 2003. Undershoot in Two Modalities: Evidence from Fast Speech and Fast
Signing. PhD dissertation University of Texas, Austin TX.
Mauk, Claude E. & Tyrone, Martha. 2008. Sign lowering as phonetic reduction in American Sign
Language. In Proceedings of the 8th International Seminar on Speech Production, Rudolph
Sock, Suzanne Fuchs & Yves Laprie (eds), 185–188. Le Chesnay, France: INRIA.
Padden, Carol. 1988. Interaction of Morphology and Syntax in American Sign Language. New
York NY: Garland.
Renard, Marc & Delaporte, Yves. 2004. Aux Origines de la Langue des Signes Française: Brouland,
Pélissier, Lambert, les premiers illustrateurs 1855–1865. Paris: Langue des Signes Éditions
Publications.
Russell, Kevin, Wilkinson, Erin & Janzen, Terry. 2011. ASL sign lowering as undershoot: A cor-
pus study. Laboratory Phonology 2(2): 403–422. doi: 10.1515/labphon.2011.015
Shaffer, Barbara. 2000. A Syntactic, Pragmatic analysis of the Expression of Necessity and
­Possibility in American Sign Language. PhD dissertation, University of New Mexico,
­Albuquerque NM.
Shaffer, Barbara. 2002. CAN’T: The negation of modal notions in ASL. Sign Language Studies
3(1): 34–53. doi: 10.1353/sls.2002.0026
Shaffer, Barbara. 2004. Information ordering and speaker subjectivity: Modality in ASL. Cogni-
tive Linguistics 15(2) 175–195. doi: 10.1515/cogl.2004.007
Shaffer, Barbara & Janzen, Terry. 2016. Modality and mood in American Sign Language. In
The Oxford Handbook of Mood and Modality, Jan Nuyts & Johan van der Auwera (eds),
448–469. Oxford: OUP.
Slobin, Dan I. 2006. Issues of linguistic typology in the study of sign language development of
deaf children. In Advances in the Sign Language Development by Deaf Children, Brenda
Schick, Mark Marschark & Patricia E. Spencer (eds), 20–45. Oxford: OUP.
 know and understand in ASL 

Slobin, Dan I. 2008. Breaking the molds: Signed languages and the nature of human languages.
Sign Language Studies 8(2): 114–130. doi: 10.1353/sls.2008.0004
Stokoe, William C., Casterline, Dorothy C. & Croneberg, Carl G. 1965. A Dictionary of ­American
Sign Language on Linguistic Principles. Washington DC: Gallaudet College Press.
Thompson, Sandra A. & Hopper, Paul J. 2001. Transitivity, clause structure, and argument struc-
ture: Evidence from conversation. In Frequency and the Emergence of Linguistic Structure
[Typological Studies in Language 45], 27–60. Amsterdam: John Benjamins.
doi: 10.1075/tsl.45.03tho
Traugott, Elizabeth Closs. 1989. On the rise of epistemic meanings in English: An example of
subjectification in semantic change. Language 65(1): 31–55. doi: 10.2307/414841
Vermeerbergen, Myriam, Leeson, Lorraine & Crasborn, Onno (eds). 2007. Simultaneity in Signed
Languages: Form and Function [Current Issues in Linguistic Theory 281]. ­Amsterdam:
John Benjamins. doi: 10.1075/cilt.281
Wilcox, Phyllis. 1998. GIVE: Acts of giving in American Sign Language. In The Linguistics of
Giving [Typological Studies in Language 36], John Newmann (ed.), 175–207. Amsterdam:
John Benjamins. doi: 10.1075/tsl.36.07wil
Wilcox, Phyllis Perrin. 2000. Metaphor in American Sign Language. Washington DC: Gallaudet
University Press.
Wilcox, Sherman. 2004. Cognitive iconicity: Conceptual spaces, meaning, and gesture in signed
language. Cognitive Linguistics 15(2): 119–147. doi: 10.1515/cogl.2004.005
Wilcox, Sherman. 2007. Routes from gesture to language. In Verbal and Signed Languages:
­Comparing Structures, Constructs and Methodologies, Elena Pizzuto, Paola Pietrandrea &
Raffaele Simone (eds), 107–131. Berlin: Mouton de Gruyter.
Wilcox, Sherman & Wilcox, Phyllis. 1995. The gestural expression of modality in ASL. In Modal-
ity in Grammar and Discourse [Typological Studies in Language 32], Joan Bybee & Suzanne
Fleischman (eds), 135–162. Amsterdam: John Benjamins. doi: 10.1075/tsl.32.07wil
 Traces of demonstrative grammaticalization in
Spanish variable subject expression
ella ‘she’ vs. él ‘he’

Rena Torres Cacoullos

Penn State University

Third person subject pronouns are widely hypothesized to arise from the
grammaticalization of demonstratives. Analysis of variation between pronominal
and unexpressed subjects in 13th–16th century Spanish texts (N = 1,947) reveals
that subjects referring to women favored pronominal expression and were more
sensitive to the syntactic role of the previous mention. The data distribution
shows that feminine referents were less topical, or more deictically distant: they
occurred less frequently than masculine ones as subjects, and their previous
mention was more likely to have been in non-subject role – a particularly
favorable environment for pronoun ella ‘she’. Thus, as a vestige of demonstrative
origins, variable subject pronoun use could express topicality. An intermediate
stage of demonstrative > pronoun grammaticalization may be use of the pronoun
for unexpected subjects.

1. Variation as a window into grammaticalization

Grammaticalization helps us understand variation between linguistic forms

with the same grammatical function. An example of a linguistic variable is Span-
ish subject expression: choice of grammatical means of mentioning an acces-
sible subject whose referent is human and specific, be it an interlocutor in the
discourse event (first and second person) or a third person. The variants are
pronominal subjects and unexpressed, or null, subjects. Variation in third per-
son singular (3sg) subject pronoun expression is illustrated in (1), from a 13th
century text. In (a), the 3sg subject is expressed, in the form of the feminine
subject pronoun ella. In (b), which immediately follows in the text, the subject
pronoun is left unexpressed. Unexpressed subjects are indicated by a zero before
the verb in the Spanish original and parentheses around the subject pronoun in
the English translation.

doi 10.1075/slcs.192.04tor
© 2018 John Benjamins Publishing Company
 Rena Torres Cacoullos

(1) Variation in Spanish 3sg subject expression

a. Pronominal subject (subject pronoun)
 Et en esto despertó su marido, et dixo a la muger: – Dame mi ferramienta
toda, ca me quiero ir de mañana a un noble omne. Et ella non le dio
sinon la navaja.
‘And just then her husband awoke, and said to the woman: -Give me all
my tools, because I want to go in the morning to a noble man. And she
gave him but the knife’  [Calila, p. 141]
b. Unexpressed subject (null)
Et él díxole: – Dame mi ferramienta. Et Ø diole de cabo la navaja.
‘And he said to her: – Give me my tools. And (she) again gave him
the knife’  [Calila, p. 141]

Variable subject expression in Spanish (as in Italian and Portuguese) could be

said to lie somewhere between Latin, which had no personal pronouns of the
third person (Greenough, Kittredge, Howard & D’ooge 1903: 176 (§ 295c)), and
French, in which subject pronouns are no longer optional (Lamiroy & De Mulder
2011: 303–304).
Third person subject expression is interesting because in some languages the
difference between the grammatical persons is explicitly grammaticalized. For
example, in languages with switch reference systems, where “as a device for refer-
ential tracking” same reference and disjoint reference are marked morphologically
on the verb, the marking is sometimes limited to third person verbs (Haiman &
Munro 1983: ix,xi). Similarly, split ergative marking is in some cases applied only
to third person (Delancey 1981). In present-day Spanish, the difference between
3sg and 1sg takes the form of differing sensitivity to the probabilistic constraint
of accessibility on subject expression, measured as distance – the number of inter-
vening clauses – between the target subject and the previous mention of the same
referent in subject position (Torres Cacoullos & Travis 2018). Once we set aside
lexical subjects, 3sg subjects are more likely to cluster close together, that is, to
occur with no intervening clauses or at a distance of fewer than five clauses from
their previous mention. Beyond this difference in contextual distribution, 3sg
(unexpressed and pronominal) subject referents are less impacted by accessibility
at low degrees of distance. This results in a greater ratio of unexpressed to pro-
nominal subjects for 3sg than for 1sg.
Cross-linguistically, one source for third person personal pronouns is demon-
strative pronouns (Heine & Kuteva 2002: 112–113; Heine & Song 2011). This is
the case for 3sg pronouns in Spanish. According to Allen and Greenough’s New
Latin Grammar for Schools and Colleges (1903, § 142) “The personal pronouns of
the third person – he, she, it, they – are wanting in Latin, a demonstrative being
sometimes used instead” (see also § 295c). The Spanish pronouns derive from
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

Latin distal demonstrative pronouns ille ‘that one, masc.sg.nom’, illa ‘that one, fem.
sg.nom’ (Penny 2002: 133). The demonstrative source of 3sg subject pronouns has
not been controversial, but so far text-based evidence for the grammaticalization
path has been lacking, because in the earliest Spanish texts we already have 3sg
subject pronouns (as in example (1)). In this study, such evidence for an inter-
mediate stage along a gradual route – a twilight zone – is discovered in variation
patterns.1
According to the grammaticalization hypothesis of retention (persistence) of
source meaning (e.g., Bybee & Pagliuca 1987; Hopper 1991), grammaticalizing
constructions or morphemes have semantic content deriving from the meaning of
their source construction. It is reasonable that such retention of source meaning
should be more evident in earlier than in later stages. The question then becomes:
Do we find differences between early and present-day Spanish in variation pat-
terns of 3sg pronominal and unexpressed subjects?
Demonstrative pronouns (‘this’, ‘that’) distinguish referents according to
spatial location and distance. Demonstratives are a classic instance of deic-
tic elements in that their reference is determined in relation to the here and
now of the speaker. We can thus further refine our question to: Do we find
­vestiges of deictic distal meaning (pointing to ‘that’) in early Spanish 3sg sub-
ject ­pronouns? The trick will be to operationalize deictic distal meaning to
be tested as part of the configuration of probabilistic constraints on variable
subject expression.

2. Data

Tokens were extracted from four texts, beginning with the 13th century Calila
et Dimna (1251), a collection of moral tales translated from Arabic to Spanish
(Castilian) under the supervision of king Alfonso X. The name of the text is
taken from the two jackals narrating most of the stories. Mentions of animals in
subject position are counted because the animals here are anthropomorphized
and are referred to with personal pronouns (‘he’, ‘she’). The Conde Lucanor
(1335) is also a set of didactic short stories, including ones based on Aesop and
Arabic folktales. La Celestina (1499) is a well-known tragicomedy written in
dialogue. It can be viewed as a feminist text in being the only one of the four in
which feminine subjects are more frequent than masculine ones. Finally, in the

. I thank a reviewer for suggesting “twilight zone” for an intermediate stage between end-
points of grammaticalization.
 Rena Torres Cacoullos

picaresque novel Lazarillo (1554), the title character narrates the woes of his
service under mean masters. (Complete titles are listed under Corpus, before
References.)
The envelope of variation is reference to an accessible 3sg specific human
subject via a personal pronoun and/or agreement marking on the verb. Non-
specific referents are outside the purview of variable subject expression
because they are virtually never realized as 3sg personal pronouns él or ella.
We set aside lexical (full NP) subjects because compared with unexpressed and
pronominal subjects their referents tend to be less accessible, with a distant
previous mention (10 or more intervening clauses) (Torres Cacoullos & Travis
2018). Also set aside are postverbal subject pronouns because the condition-
ing of subject position differs from that of subject expression (Benevento &
Dietrich 2015). Subject-headed relatives are excluded because the subject is
expressed via a relative pronoun and wh- interrogatives are excluded because
subjects are either unexpressed or positioned post-­verbally. Finally, we do not
count references to religious entities (God, the devil) and metalinguistic or
fossilized contexts (e.g., proverbs). With approximately 500 tokens of eligible
3sg verbs extracted from each of the four texts, the total number of obser-
vations, or tokens, is N = 1,947. The overall rate of subject pronouns in this
dataset is 10%.
The evolution of demonstratives to pronouns involves loss of deictic mean-
ing. With semantic bleaching we would expect generalization of the use of the
forms to more contexts, which should be reflected in their increased frequency.
This is not borne out, however. The four texts do not show a neat increase in
overall rate (Calila et Dimna 7%, N = 501; Lucanor 14%, N = 547; Celestina
10%, N = 428; Lazarillo 10%, N = 471). We know that overall rates of subject
pronouns are equivocal, being susceptible to the preponderance or dearth in
a data set of some very propitious context, which may be fortuitous or due to
extra-grammatical situational considerations, such as topic or genre (Travis
2007: 129–131).
Figure 1 now compares the aggregate rate of 3sg subject pronouns in our
early texts with those reported for present-day spoken varieties. Though we
do find that the early rate of 10% is lower than that of present-day Spanish in
Madrid (10% vs. 14%, p = 0.0002, by Fisher’s exact test), there are rate differ-
ences of the same magnitude within present-day varieties, from 14% in Madrid
to 23% in Mexico City and, within Puerto Rico, from 29% in Castañer to 39% in
San Juan. The comparison, then, does not provide the grounds to claim that the
rate difference, though statistically significant, is linguistically meaningful as a
gauge of change. The conclusion is that if there was a notable increase in the use
of (what eventually became) 3sg subject pronouns since Latin, it likely predates
the earliest Spanish texts.
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

50 Unexpressed Ø
40 Pronoun

30
(%)
20

10

0
Early SPAN Madrid, Spain Mexico, D.F. Castañer, P.R. San Juan, P.R.

Figure 1. Rate of 3sg subject expression (pronominal vs. Ø) in early Spanish texts compared
with present-day varieties
**Madrid 14%, N = 2501 (Enríquez 1984: 348); Mexico, D.F. 27%, N = 450 (Lastra & Martín
Butragueño 2015: 43); Castañer, P.R. 29%, N = 603 (Holmquist 2012: 211); San Juan, P.R. 39%,
N = 443 (Cameron 1992: 233).

Dialect differences in overall expression rates have been interpreted as a reflex of

change in progress, in particular for Caribbean varieties, toward obligatory sub-
ject pronouns. But in a comparison of San Juan, Puerto Rico, and Madrid, Spain,
Cameron (1993) found identical probabilistic constraints, evidence against a dif-
ferent grammar of subject expression. And even if a rate difference is taken as
evidence for a change, it fails to speak to the direction of that change. An overall
rate difference alone, as Cameron (1995: 24, n. 26) puts it, “does not permit us to
say if speakers from San Juan are changing in the direction of a greater frequency
of pronominal expression or if the speakers from Madrid are so doing in the direc-
tion of lesser frequency of pronominal expression.”
So, since we cannot rely on rate increases in order to characterize change, we
look to the linguistic conditioning of variation between expressed él (masculine)
or ella (feminine) vs. unexpressed 3sg subjects. To discover the linguistic condi-
tioning, we operationalize hypotheses about speaker choices by coding each token
according to features of the linguistic context, which in turn define the factors for
the statistical analysis.

3. O
 perationalizing deictic function and other motivations for the use of
subject pronouns

We begin with accessibility. Givón’s 1983 influential cross-linguistic volume on

topic continuity established that less continuous referents (or referents that are
lower on the accessibility scale) tend to be associated with more linguistic cod-
ing. More linguistic coding applies to stressed vs. unstressed pronouns, full Noun
Phrases vs. pronouns, or, as is the case here, subject pronouns vs. agreement mor-
phology on the verb (what we refer to as unexpressed subjects).
 Rena Torres Cacoullos

Accessibility is operationalized as distance from the previous mention of the

referent of the target subject. This is configured here as an opposition between
a distance of zero, when the previous mention is in the immediately preceding
clause in any syntactic role, and a distance of one or more intervening clauses from
the previous mention.
The examples in (2) illustrate the coding. In (2a) the target 3sg verb is bolded
in ‘(he) had promised her’ and the previous mention of the subject referent,
underlined, is in the immediately preceding clause, ‘(he) fulfilled for her’. In (2b),
the target 3sg is in ‘(he) wouldn’t have left’ and the previous mention of the refer-
ent is ‘blind man’. Here more than one clause intervenes (‘they hadn’t come’ and
‘I think’).
(2)
a. 
Distance from previous mention: 0 intervening clauses (more
accessible)
fasta quel’ Ø cumpliesse lo quel’ Ø avía prometido
‘until (he) fulfilled for her what (he) had promised her’ 
 [Lucanor, cuento L]
b. Distance from previous mention: 1+ intervening clauses (less accessible)
 Fue tal el coraje del perverso ciego que, si al ruido no acudieran, pienso Ø
no me dejara con la vida
‘The fury of the perverse blind man was such that if they hadn’t come
thanks to the noise, I think (he) wouldn’t have left me alive’ 
 [Lazarillo, Tratado Primero]
To operationalize deictic function, we reason as follows. Spatial deixis – “proxim-
ity, distance, visibility” – may extend to temporal deixis “involving proximity…
associated with the prior mention of the referent in discourse” (Givón 1984: 354).
Our hypothesis is that, because they derive from the Latin distal demonstrative,
3sg subject pronouns point to more distant, or less topical, referents than unex-
pressed subjects do. The intuitive notion of topicality may be operationalized
through anaphoric and cataphoric measures such as counting distance from, and/
or number of, previous and subsequent mentions (e.g., Myhill 2005). Since we
are interested in whether the subject pronoun points back, we implement here an
anaphoric measure, looking to the previous mention.
We reason that more deictically ‘distant’ – less ‘topical’ – referents are those
previously mentioned in a syntactic role other than that of subject, given the
association between subject and topic (e.g., Comrie 1989; Givón 1979: 209,
1983: 22). For example, Schwenter & Torres Cacoullos (2014: 529) show for 3sg
direct object clitics that referents previously mentioned as subject of the imme-
diately preceding clause are more likely to exhibit topic persistence (with two or
more mentions in the subsequent 10 clauses) than those whose previous men-
tion was as direct object.
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

Thus, while for accessibility we code the distance from the previous mention,
for deixis-topicality we code the syntactic role of the previous mention. The exam-
ples in (3) illustrate. In both, the previous mention (underlined) is in the immedi-
ately preceding clause, but in (3a) the previous mention occurs as subject, while in
(3b) it appears in a non-subject syntactic role (here, as indirect object).
(3) a.
S yntactic role of previous mention: subject (less deictically distant/
more topical)
 por el amor et grant fiança que en mí Ø avía, que Ø me quería dexar
toda su tierra
‘for the friendship and great confidence that (he) had in me, that (he)
wanted to leave me his lands’  [Lucanor, I]
b. 
Syntactic role of previous mention: non-subject (more deictically
distant/less topical)
 Saladín le dixo que reçelava quel’ Ø pidría que non le Ø fablasse más en
aquel fecho. Et ella díxol’ que non le demandaría esso
‘Saladin told her that he feared that (she) would ask him that (he) not
speak to her about that. And she told him that she would not demand
this’  [Lucanor, L]

The third constraint tested is priming, or the tendency for speakers to repeat the
variant they most recently used. Priming is a domain-general mechanism applying
outside of language, for example to human motor control (van der Wel, Flecken-
stein, Jax & Rosenbaum 2007). For subject expression, the general pattern is that
pronouns lead to pronouns (Cameron 1994). To test this in our early Spanish data,
we code the realization, or form of the previous mention. We distinguish previous
mentions as pronouns (as opposed to previous unexpressed mentions), again in
any syntactic role. In (4), the previous mention of the target ‘he had given’ is in the
form of the subject pronoun él ‘he’ in ‘he asked’.
(4) 
Form of previous mention: pronominal (pronoun to pronoun priming)
 Et él preguntól’ si tenía todas las cosas que él le diera por escripto
‘And he asked him if he had all the things that he had given him in writing’
 [Lucanor, XX]

Finally, we also coded for the gender of the referent. A gender effect has not been
reported for subject pronoun expression in present-day Spanish (as far as I know).2

. A grammatical variable that is sensitive to referent gender is the form of object pronoun
clitics in peninsular Spanish. When their referents are masculine animates, direct objects tend
to be marked by le, the etymologically dative (indirect object) form, rather than lo or la, the
accusative forms, a tendency known as leísmo (Klein Andreu 1992: 171).
 Rena Torres Cacoullos

4. Linguistic conditioning of subject pronouns in early Spanish texts

To ascertain the linguistic conditioning of early Spanish 3sg subject expression we

are interested in direction of effect: Do the contextual features, or factors, favor or
disfavor subject pronouns as opposed to the unexpressed alternative?
Table 1 shows the results of a Variable rule analysis using Goldvarb X (Sankoff
1988; Sankoff, Tagliamonte & Smith 2012), a statistical program designed for lin-
guistic variation in natural language use, where data distributions are not controlled
and factors operate simultaneously.3 The first column in Table 1 gives the probabil-
ity value for a pronominal subject corresponding to each factor, such that the closer
to 1, the more a pronoun is favored in the context represented by the factor. (The
closer the probability value is to 0, the stronger the disfavoring effect, or, conversely,
the favoring of the unexpressed variant.) The columns that follow give the subject
pronoun rate (%) and the number of subject pronouns and total number of tokens
(pronominal and unexpressed variants) in that context.
As we see in Table 1, accessibility is important. A pronominal subject is
favored when the previous mention of the referent is at a distance of one or more
intervening clauses. At the same time there is a priming effect, such that previous
mention in the form of a pronoun again favors a pronoun. Feminine referents
favor the pronominal variant more than masculine referents do. Finally, a subject
pronoun is favored when the syntactic role of the previous mention was other than
that of subject, in support of the hypothesis that in these earlier Spanish texts 3sg
subject pronouns point to more distant, or less ‘topical’, referents, consistent with
their distal demonstrative origins.
What is the relationship between distance and syntactic role of the previous
mention? Figure 2 depicts the rate of subject pronouns by distance from the previous
mention and syntactic role of the previous mention. The lowest rate (4%, 39/1,101) is
for previous mentions in the immediately preceding clause (a distance of 0 interven-
ing clauses) in the syntactic role of subject. In other words, subject continuity highly
favors unexpressed subjects (Silva-Corvalán 1994: 157). In contrast, the highest rate
obtains for previous mentions at a distance of one or more (1+) intervening clauses
in non-subject role (23%, 29/125). For the remaining two configurations, the rate
is similar for previous mentions at 0 distance as non-subject (15%, 33/223) and for
previous mentions at 1 distances as subject (17%, 64/367).

. Factor groups (predictors) as coded are entered into a statistical model, and the outcome
variable (in our case, 3sg subject expression) is ‘predicted’ based on these groups. Variable-rule
analysis determines the predictors that together account for the largest amount of variation,
in terms of stepwise increase of log likelihood, such that the addition of any of the remaining
predictors does not significantly increase the fit to the model (Sankoff 1988).
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

Table 1. Logistic regression analysis of factors contributing to the choice of pronominal

vs. Ø 3sg human specific subjects in early Spanish texts (Variable rule analysis)4
N = 1,947; Input: .08; Overall rate: 10%

Prob. % Pron N

Accessibility: Distance from the previous mention

1 or more clauses .73 20% 122/623
0 clauses .39 5% 72/1,324
Priming: Form of the previous mention*
Pronominal .64 19% 67/353
Unexpressed (Ø) .47 7% 70/1,070
Gender of referent
Feminine .61 14% 74/527
Masculine .46 9% 120/1,420
Deixis-topicality: Syntactic role of the previous mention
Other than subject .59 18% 62/348
Subject .48 7% 103/1,468
Also included in the analysis is syncretism in verb morphology (syncretic .57 – 12% – 88/714 vs. non-
syncretic .45 – 8% – 93/1,105); presence of a reflexive marker (non-significant).4
* Form of previous mention excludes previous mentions as full NP and previous mentions beyond 5
clauses.

. A popular hypothesis is that subject pronouns serve to resolve or compensate for poten-
tial ambiguity of reference. This is at least partially the legacy of formal syntactic approaches
appealing to a null subject parameter whereby “the subject can be dropped” in languages such
as Spanish because their verbs carry person-number agreement (Chomsky 1981 [1993]: 241)).
Here tokens were coded according to person syncretism in verb morphology (where 3sg and
1sg is identical, namely Imperfect and Subjunctive forms). The effect is limited to less acces-
sible subjects (1+ intervening clauses from previous mention) (cf. Cameron 1993: 317).
 Rena Torres Cacoullos

25
Previous non-subject
(N=348)
Previous subject
20
(N=1468)

15

10

0
0 intervening clauses 1+ intervening clauses
(N=1324) (492)

Figure 2. Rate of subject pronouns by distance and syntactic role of the previous mention

But Figure 2 also tells us that the effects of accessibility (operationalized as dis-
tance from the previous mention) and deixis-topicality (syntactic role of the pre-
vious mention) are independent probabilistic constraints. For both syntactic roles
of previous mention – as subjects (solid line) and as non-subjects (dotted line) –
the rate of pronominal subjects rises from left to right: the subject pronoun rate
is higher when the previous mention is at a distance of one or more intervening
clauses, 1+, than at a distance of 0 intervening clauses, that is, when it is in the
immediately preceding clause. And, at the same time, for both distances of 0 or 1 ,
the rate is higher for previous mentions as non-subjects than for previous men-
tions in the syntactic role of subject.
Someone might counter that both distance and syntactic role of previous
mention have a common denominator.5 Whether we call this accessibility or
topic continuity, or think of the former as more cognitive and the latter as more
discourse-based, the data indicate that there are two dimensions. Furthermore,
distance and syntactic role do not contribute equally to selection of a pronominal
over an unexpressed subject. Distance was already more important than syntactic
role in the early texts, seen in the steeper slope of the solid line than the dotted
line in Figure 2. The importance of syntactic role may have (further) declined over
time: whereas in the early Spanish texts the subject pronoun rate is about four

. I thank a reviewer for commenting that a common denominator of both lines in Figure 2
is the use of subject pronouns in situations of a topic shift.
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

times lower for subject mentions than for non-subject mentions in the immedi-
ately preceding clause (“0 intervening clauses”, 4% vs. 15%, Figure 2), in present-
day data the difference is smaller.6

5. Th
 e intersection of referent gender and syntactic role of previous
mention

But why are feminine subjects more likely than masculine subjects to be pronomi-
nal rather than unexpressed in our early Spanish texts? To interpret the gender
effect, we look at cross-tabulations of gender and syntactic role of the previous
mention. We are interested in two pieces of evidence from the cross-tabulations:

1. the difference between the genders in the effect of syntactic role of the previ-
ous mention, and
2. the difference between them in their contextual distributions as concerns the
syntactic role of the previous mention.

First is the rate of subject pronouns according to gender of the referent and syntac-
tic role of the previous mention, shown in Figure 3. On the one hand, the differ-
ence in the pronoun rate between masculine, to the left, and feminine, to the right,
is more pronounced for previous mentions in non-subject role. This is seen in the
dotted line, which has the steeper slope (for previous mentions as non-subject,
the subject pronoun rate is 13%, 27/209, masculine vs. 25%, 35/139, feminine; for
previous mentions as subject, it is 6%, 69/1096, vs. 9%, 34/372). And on the other
hand, the difference in pronoun rate between subjects whose referent was previ-
ously mentioned as a subject and those with a non-subject previous mention is
more pronounced for feminine than for masculine referents. This is seen in the
gap between the two lines, which is greater for feminine (for feminine referents,
the subject pronoun rate is 25% previous non-subject vs. 9% previous subject; for
masculine referents it is 13% vs. 6%). This cross-tabulation, then, indicates a dif-
ference between the genders in the effect of syntactic role of the previous men-
tion: it affects subject expression with feminine referents more than it does subject
expression with masculine ones. That is, feminine subject expression is more sen-
sitive to the syntactic role of the previous mention.

. In one present-day dataset, the pronoun rate in the context of a previous mention in the
immediately preceding clause as the subject vs. as an object or oblique is 29%, 245/858 vs. 34%,
60/149 for 1sg and 3sg combined (Torres Cacoullos & Travis, 2018), and in another dataset,
20% (176/873) vs. 35% (60/170) for “all types of subjects” (Silva-Corvalán, 1994: 157).
 Rena Torres Cacoullos

Second, consider token counts according to gender and proportions of tokens

according to the syntactic role of the previous mention, shown in Figure 4. On
the vertical axis are 3sg subject token counts (unlike the preceding figures, which
so far have depicted subject pronoun rate). On the one hand, in these data, 3sg
subjects referring to women are notably less frequent than those referring to men,
with fewer than half the occurrences, as seen in the relative height of the bars. This
means that, in the aggregate, (pronominal and unexpressed) subjects referring to
women are less expected than subjects referring to men.

30
Previous non-subject (N=348)
Previous Subject (N=1468)
25

20

15

10

0
Masculine (N = 1305) Feminine (N = 511)

Figure 3. Rate of subject pronouns by gender and syntactic role of the previous mention

1400 Previous non-subject (N=348)

1200 Previous Subject (N=1468)

1000

800

600

400

200

0
Masculine (N = 1305) Feminine (N = 511)

Figure 4. Token frequencies and contextual distributions (proportions of tokens) of 3sg

subjects according to gender of the referent and syntactic role of the previous mention

On the other hand, the especially revealing result of the cross-tabulation of gender
and syntactic role of the previous mention are the data distributions. Here we are
interested in the proportions of data according to the syntactic role of the ­previous
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

mention, shown by shading within each bar. These indicate a difference in the
contextual distributions of feminine and masculine 3sg subject referents. The pro-
portion of previous mentions as subject (the darker shade in each bar) is 84%
(1096/1305) for masculine but 73% (372/511) for feminine (p < 0.0001, Fisher’s
exact test). Thus, in these early Spanish texts 3sg subjects with feminine referents
are not only less frequent than subjects with masculine referents but, when they
do occur, the referent is less likely to have a previous mention in subject role. The
data distributions at the intersection of referent gender and syntactic role of the
previous mention allow the following conclusion: feminine referents appearing as
3sg subjects tended to be more deictically distant than masculine ones in the early
Spanish texts.
In terms of topicality, this pair of results – 3sg subjects with feminine referents
are less frequent and are less likely to have a previous mention as subjects – sug-
gests that feminine referents are less topical than masculine referents in these texts
overall. This claim is grounded in the association between the syntactic role of
subject and the discourse status of topic, where topicality has been measured by
topic persistence, or mentions in subsequent discourse (e.g., Givón 1979: 209, 298;
Schwenter & Torres Cacoullos 2014).
The difference in topicality between 3sg subjects referring to men and
women is intriguing when we pause to think of grammatical gender. Cross-
linguistically, it is proposed that gender systems that “fulfill the function of
providing a means for referent tracking […] tend to be sex-based, since male
and female humans are equally discourse-prominent and topic-worthy entities”
(Luraghi 2011: 459). Apparently male and female humans are not so equal in
the early Spanish texts. At the same time, feminine may tend to be the more
specified gender. For example, the distribution of personal pronouns in spoken
Dutch has come to depend on semantic properties of the nominal referent, in
particular its degree of individuation, such that neuter corresponds to a low
degree of individuation, masculine to a high degree of individuation, and the
feminine is restricted to “female persons and a few feminine animals” (Audring
2006: 113). Spanish gender assignment is also consistent with feminine being
the more specified class. Feminine word endings display lesser schematicity
than masculine ones, where schematicity “refers to the degree of dissimilar-
ity of the members of a class” (Bybee 2010: 67). Thus, 95% of all nouns in a
database of nearly 2,500 are covered by the rule that the endings -a, -d, -ción,
-sión, -tis, -sis, and -iz are feminine while any others are masculine (Eddington
2002: 62–66).
Returning to our variably expressed subject pronouns in early Spanish texts,
we can say that feminine ella may have more specific meaning than masculine él:
the lesser topicality of feminine referents makes them less likely, or unexpected,
subjects.
 Rena Torres Cacoullos

6. Conclusion

Usually highlighted in the study of grammaticalization is the loss of specific fea-

tures of meaning, as (combinations of) words or partially schematic constructions
generalize to use in more contexts en route to becoming grammatical expressions,
a process known as semantic depletion (Givón 1975: 94), semantic ­bleaching
(Lehman 1982/2002: 43) or desementicization (Heine & Reh 1984: 36). Joan
Bybee has often stressed the converse observation, namely retention in grammati-
calization. The hypothesis of retention is directly relevant to the empirical study of
linguistic variation and change. According to the retention hypothesis, the seman-
tic content of grammatical expressions includes not only their current grammati-
cal functions but also retains features or nuances of meaning derived from their
source constructions (Bybee, Perkins & Pagliuca 1994: 138, 148).
An oft-cited cross-linguistic grammaticalization path is that whereby pro-
nouns originate in demonstratives, which involves loss of deictic meaning. But
tangible traces of those origins have been hard to identify. The study of the evolu-
tion of Romance languages from Latin faces the same problem. Nobody doubts
that the distal demonstratives of Latin evolved into the 3sg pronouns of Spanish.
But since the forms apparently already functioned as pronouns in the earliest texts
written in Spanish, it has not been possible to track the process, the details of
‘how’. Here we tackled this problem by looking at quantitative patterns of variation
between pronominal and unexpressed 3sg subjects in early Spanish texts.
We asked whether probabilistic constraints in variably expressed subject pro-
nouns in earlier texts as compared with Modern Spanish could be attributed to
traces – retention – of semantic nuances from their distal demonstrative origins.
We operationalized distal deictic meaning by means of the syntactic role of the
previous mention of the subject referent. This contextual feature does correlate
with 3sg pronominal subjects: previous mention of the referent in non-subject
role favors use of a pronominal over an unexpressed subject more than previ-
ous mention as a subject does. This result alone, though, could be interpreted as
merely a cognitive effect of accessibility, since referents in subject position are “the
most easily retrieved from memory” (Bock & Warren 1985: 61).
However we also saw that referent gender correlates with pronominal sub-
jects. Feminine referents favor use of ella ‘she’ more than masculine referents favor
expression of él ‘he’. Why? The gender effect is tied to a contextual (data) distri-
bution difference – 3sg feminine subjects have a lower proportion of previous
mentions of their referent in the syntactic role of subject – as well as to a genuine
difference in the effect of the syntactic role of the previous mention – 3sg feminine
subject expression is more sensitive to the syntactic role of the previous mention.
From these quantitative facts we can infer that ella designates less topical and thus
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

more deictically distant referents than él does. We can then interpret the correla-
tion of referent gender with subject pronoun rate as reflecting the demonstrative
origins of subject pronouns, a manifestation of retention.
In sum, together the differences in context of occurrence and probabilistic
effect between feminine and masculine 3sg referents suggest that subject expres-
sion in earlier Spanish was sensitive not only to accessibility (measured by distance
from the previous mention of the subject referent), but also to socio-pragmatic
considerations of topicality, marking less likely or unexpected – more deictically
distant – subjects, as measured by the intersection between the syntactic role of
the previous mention and the gender of the referent.
The quantitative analysis reveals the contribution of demonstrative origins to
the variation patterns in support of the retention hypothesis. The variation pat-
terns, in turn, suggest the evolution in (5):
(5) distal demonstrative >
subject pronoun constrained by topicality
(pronoun favored for less topical referents, or unexpected subjects) >
subject pronoun constrained by accessibility
According to this proposed elaboration, an intermediate stage along the demon-
strative > pronoun grammaticalization path is the use of subject pronouns for ref-
erents that are less likely to be subjects in the discourse. This hypothesis can be
tested in studies of variable 3sg subject expression in other historical datasets of
Spanish and in other languages for which the demonstrative > pronoun grammati-
calization path has been claimed.

Acknowledgements

This paper is the outcome of the joint work of Grant Berry, Chris Champi, Lauren Perrotti and
Miguel Ramos for SPAN 508 Hispano-Romance Linguistics (Language Change) at Penn State
in Spring 2014, inspired by Language Change (Bybee 2016) a draft of which Joan kindly made
available to us at the time. Co-authored versions were presented at New Ways of Analyzing
Variation (NWAV) 43, Chicago (2014) (Torres Cacoullos et al. 2014) and at the Congreso Inter-
nacional De Historia De La Lengua Española (CIHLE) X, Universidad de Zaragoza (2015).

Corpus

[Calila], Anonymous, 1984, Calila et Dimna. Juan Manuel Cacho Blecua & María Jesús Lacarra
(eds). Madrid: Castalia.
[Lucanor], Don Juan Manuel, 1969, El conde Lucanor o Libro de los enxiemplos del conde Lucanor
et de Petronio. J. M. Blecua (ed.), Madrid: Castalia.
 Rena Torres Cacoullos

[Celestina], Fernando de Rojas, 1993, La Tragicomedia de Calisto y Melibea. Dorothy S. Severin

(ed.), Madrid: Cátedra.
[Lazarillo], Anonymous, 1995, La vida de Lazarillo de Tormes y de sus fortunas y adversidades.
The Project Gutenberg Etext, prepared by Jesús Joglar.

References

Audring, Jenny. 2006. Pronominal gender in spoken Dutch. Journal of Germanic Linguistics
18(2): 85–116. doi: 10.1017/S1470542706000043
Benevento, Nicole, & Dietrich, Amelia. 2015. I think, therefore digo yo: Variable position of
the 1sg subject pronoun in New Mexican Spanish-English code-switching. International
Journal of Bilingualism 19(4): 407–422. doi: 10.1177/1367006913516038
Bock, J. Kathryn & Warren, Richard K. 1985. Conceptual accessibility and syntactic structure in
sentence formulation. Cognition 21: 47–67.
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan. 2016. Language Change. Cambridge: CUP.
Bybee, Joan & Pagliuca, William. 1987: The evolution of future meaning. In Papers from the sev-
enth International Conference on Historical Linguistics [Current Issues in Linguistic Theory
48], Anna Giacalone Ramat, Onofrio Carruba & Giuliano Bernini (eds), 109–122. Amster-
dam: John Benjamins. doi: 10.1075/cilt.48.09byb
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar: Tense,
Aspect, and Modality in the Languages of the World. Chicago IL: Chicago University Press.
Cameron, Richard. 1992. Pronominal and Null Subject Variation in Spanish: Constraints,
­Dialects, and Functional Compensation. PhD dissertation, University of Pennsylvania,
Philadelphia PA.
Cameron, Richard. 1993. Ambiguous agreement, functional compensation, and nonspecific
tú in the Spanish of San Juan, Puerto Rico, and Madrid, Spain. Language Variation and
Change 5(3): 305–334. doi: 10.1017/S0954394500001526
Cameron, Richard. 1994. Switch reference, verb class and priming in a variable syntax. In Papers
from the 30th Regional Meeting of the Chicago Linguistic Society: Parasession on Variation
in Linguistic Theory, Vol. 2, Katherine Beals, Jeannette Denton, Robert Knippen, Lynette
Melnar & Hisami Suzuki (eds), 27–45. Chicago IL: CLS.
Cameron, Richard. 1995. The scope and limits of switch reference as a constraint on pronominal
subject expression. Hispanic Linguistics 6–7: 1–27.
Chomsky, Noam. 1981[1993]. Lectures on Government and Binding: The Pisa Lectures, 7th edn.
Dordrecht: Foris.
Comrie, Bernard. 1989. Language Universals and Linguistic Typology, 2nd edn. Chicago IL:
­University of Chicago Press.
Delancey, Scott. 1981. An interpretation of split ergativity and related patterns. Language 57(3):
626–657. doi: 10.2307/414343
Eddington, David. 2002. Spanish gender assignment in an analogical framework. Journal of
Quantitative Linguistics 9: 49–75. doi: 10.1076/jqul.9.1.49.8482
Enríquez, Emilia V. 1984. El pronombre personal sujeto en la lengua española hablada en
Madrid. Madrid: Consejo Superior de Investigaciones Científicas, Instituto Miguel de
Cervantes.
 Traces of demonstrative grammaticalization in Spanish variable subject expression 

Givón, Talmy. 1975. Serial verbs and syntactic change: Niger-Congo. In Word Order and Word
Order Change, Charles N. Li (ed.), 47–112. Austin TX: University of Texas Press.
Givón, Talmy. 1979. On Understanding Grammar. New York NY: Academic Press.
Givón, Talmy. 1983. Topic continuity in discourse: An introduction. In Topic Continuity in
Discourse: A Quantitative Cross-linguistic Study, T. Givón (ed.), 1–41. Amsterdam: John
Benjamins. doi: 10.1075/tsl.3.01giv
Givón, Talmy. 1984. Syntax. A functional-typological introduction, Vol I. Amsterdam: John
Benjamins.
Greenough, J. B., Kittredge, G. L., Howard, A. A. & D’ooge, Benj. L. 1903. Allen and Greenough’s
New Latin Grammar for Schools and Colleges. Boston MA: The Athenaeum Press, Ginn &
Company.
Haiman, John & Munro, Pamela. 1983. Introduction. In Switch Reference and Universal Gram-
mar [Typological Studies in Language 2]. Amsterdam: John Benjamins. doi: 10.1075/tsl.2
Heine, Bernd & Reh, Mechtild. 1984. Grammaticalization and Reanalysis in African Languages.
Hamburg: Helmut Buske.
Heine, Bernd & Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
doi: 10.1017/CBO9780511613463
Heine, Bernd & Song, Kyung-An. 2011. On the grammaticalization of personal pronouns.
­Journal of Linguistics 47(3): 587–630. doi: 10.1017/S0022226711000016
Holmquist, Jonathan. 2012. Frequency rates and constraints on subject personal pronoun
expression: Findings from the Puerto Rican highlands. Language Variation and Change
24(2): 203–220. doi: 10.1017/S0954394512000117
Hopper, Paul J. 1991. On some principles of grammaticization. In Approaches to grammatical-
ization, Vol. 1, Elizabeth Closs Traugott & Bernd Heine (eds.), 17–35. Amsterdam: John
Benjamins.
Klein-Andreu, Flora. 1992. Understanding standards. In Explanation in Historical Linguistics,
G. W. Davis and G. K. Iverson (eds.) [Current Issues in Linguistic Theory 84], 167–178.
Amsterdam: John Benjamins.
doi: 10.1075/cilt.84.11kle
Lamiroy, Béatrice & De Mulder, Walter. 2011. Degrees of grammaticalization across languages. In
The Oxford Handbook of Grammaticalization, Heiko Narrog & Bernd Heine (eds), 302–317.
Oxford: OUP.
Lastra, Yolanda & Butragueño, Pedro Martín. 2015. Subject pronoun expression in oral Mexican
Spanish. In Subject Pronoun Expression in Spanish: A Cross-dialectal Perspective, Ana M.
Carvalho, Rafael Orozco & Naomi Lapidus Shin (eds), 39–57. Washington DC: George-
town University Press.
Lehmann, Christian. 2002[1982]. Thoughts on Grammaticalization, 2nd rev. edn. Erfurt: Semi-
nar für Sprachwissenschaft der Universität. <https://www2.uni-erfurt.de/sprachwissen-
schaft/ASSidUE/ASSidUE09.pdf>
Luraghi, Silvia. 2011. The origin of the Proto-Indo-European gender system: Typological con-
siderations. Folia Linguistica 45(2): 435–464. doi: 10.1515/flin.2011.016
Myhill, John. 2005. Quantitative methods of discourse analysis. In Quantitive Linguistik: Ein
internationales Handbuch, Reinhard Köhler, Gabriel Altmann & Raimond Gerikhovich
Piotrowski (eds), 471–497. Berlin: Walter de Gruyter.
Penny, Ralph. 2002. A history of the Spanish language. Cambridge: Cambridge University Press.
Sankoff, David. 1988. Variable rules. In Sociolinguistics: An International Handbook of the Sci-
ence of Language and Society, Vol. 2, Ulrich Ammon, Norbert Dittmar & Klaus J. Mattheier
(eds), 984–997. Berlin: Walter de Gruyter.
 Rena Torres Cacoullos

Sankoff, David, Tagliamonte, Sali & Smith, Eric. 2012. Goldvarb LION: A variable rule appli-
cation for Macintosh, University of Toronto. <http://individual.utoronto.ca/tagliamonte/
goldvarb.htm>
Schwenter, Scott A. & Torres Cacoullos, Rena. 2014. Competing constraints on the variable
placement of direct object clitics in Mexico City Spanish. Revista Española de Lingüística
Aplicada/Spanish Journal of Applied Linguistics 27(2): 514–536. doi: 10.1075/resla.27.2.13sch
Silva-Corvalán, Carmen. 1994. Language Contact and Change: Spanish in Los Angeles. Oxford:
Clarendon Press.
Torres Cacoullos, Rena & Travis, Catherine E. Forthcoming, 2018. Bilingualism in the Commu-
nity: Code Switching and Grammars in Contact. Cambridge: CUP.
Torres Cacoullos, Rena, Berry, Grant, Champi, Chris, Perotti, Lauren & Ramos, Miguel. 2014.
Early conditioning of Spanish variable subject expression. Paper presented at New Ways of
Analyzing Variation (NWAV) 43, Chicago, 23–26 October.
Travis, Catherine E. 2007. Genre effects on subject expression in Spanish: Priming in narrative
and conversation. Language Variation and Change 19(2):101–135.
doi: 10.1017/S0954394507070081
Travis, Catherine E. & Lindstrom, Amy M. 2016. Different registers, different grammars? Sub-
ject expression in English conversation and narrative. Language Variation and Change
28(1):103–128.
van der Wel, Robrecht P. R. D., Fleckenstein, Robin M., Jax, Steven A. & Rosenbaum, David A.
2007. Hand path priming in manual obstacle avoidance: Evidence for abstract spatiotem-
poral forms in human motor control. Journal of Experimental Psychology: Human Percep-
tion and Performance 33(5):1117–1126.
 The company that word-boundary sounds keep
The effect of contextual ratio frequency on word-final
/s/ in a sample of Mexican Spanish

Earl K. Brown
Brigham Young University

This paper analyzes the frequency with which words occur in phonological
contexts favorable to reduction, referred to here as “contextual ratio frequency.”
Duration and center of gravity of word-final /s/ are measured in the speech of ten
speakers of Mexican Spanish living in Salinas, California. The 1,028 tokens are
subjected to mixed effects linear regression, with speaker and lexical item entered
as random effects. The results suggest that contextual ratio frequency is a better
predictor of word-boundary sound reduction than lexical frequency, and support
the idea that words that occur frequently in phonological contexts conducive
to reduction are reduced more often because words are represented mentally as
malleable cognitive entities that respond to usage-based factors.

Introduction

Usage-based models of language

Usage-based models of language are based on the assumption that language usage
affects the mental representation of language in memory, that is, its grammar.
Bybee (2006: 730) asserts that “usage feeds into the creation of grammar just as
much as grammar determines the shape of usage” and that “grammar is the cogni-
tive organization of experience with language.” This perspective has been applied
to phonology (e.g. Bybee 1998; Bybee 2001), morphology (e.g. Bybee 1985)
and syntax (e.g. Du Bois 1985; Givón 1979; Hopper 1998) among other areas of
­language study.
An important aspect of language usage is frequency. Words occur at differ-
ent rates of usage in spontaneous speech; some words occur frequently or very
frequently, while others occur less frequently or occur only once, or not at all in
a given conversation. Usage-based models of language propose to track these
fluctuations in frequency and how frequency might impact future use (e.g. Bybee
2007; Bybee & Hopper 2001). Authors have argued that frequent words are more

doi 10.1075/slcs.192.05bro
© 2018 John Benjamins Publishing Company
 Earl K. Brown

subject to articulatory reduction than are infrequent words. More than a century
ago, Schuchardt (1885: 26, translated by Vennemann & Wilbur 1972: 58) noted
that, with respect to the “phonetic transformation” of words, “Rarely-used words
drag behind; very frequently used ones hurry ahead.” Hooper (1976) points out
that higher-frequency memory and nursery are more likely than lower-frequency
armory and cursory to experience post-tonic schwa deletion, and therefore be
pronounced with two syllables rather than three. File-Muriel (2009) finds that
frequency is the most influential factor in the reduction of word-medial, syllable-
final /s/ in a reading task in the Spanish of Barranquilla, Colombia. Many other
studies find that frequency plays a significant role in the reduction of sounds in
speech (cf. Bybee 2001; Labov 1994; Moonwomon 1992; Phillips 2006). Likewise,
significant influence on the reduction of sounds has been found for the frequency
of multiword strings (cf. Bybee & Scheibman 1999; Scheibman 2000) and the fre-
quency of multiword phrases in relation to the frequency of the individual words
that comprise them (cf. Bush 2001; Jurafsky et al. 2001).
A corollary of lexical frequency and string frequency is contextual ratio fre-
quency, that is, the ratio at which words occur in contexts that lead to the reduc-
tion of their word-boundary sounds. Some words occur frequently in these
contexts, and thus experience articulatory reduction of one or both of their word-
boundary sounds more often than words that occur infrequently in these contexts.
For example, word-final Spanish /s/ is reduced more often when followed by a
consonant than when followed by a vowel or a pause. Hence, words with final /s/
in Spanish that occur frequently before a consonant experience reduction of that
/s/ more often than words that occur infrequently in this context. Usage-based
models of language claim that contextual ratio frequency is also stored in mem-
ory and therefore conditions future production of the word-boundary sounds in
those words. With reference to the deletion of /t/, /d/ in American English, Bybee
(2002: 261) proposes that: “Words that occur more often in the context for change
change more rapidly than those that occur less often in that context.” A growing
body of research provides empirical support for this assertion. Brown (2004) and
Raymond and Brown (2012) find that word-initial /s/ in the colonial Spanish of
northern New Mexico and southern Colorado is reduced more often in words that
occur more frequently in the context that most favors reduction in that variety,
even when phonological context is controlled for. Brown (2009a) reports similar
results for the influence of contextual ratio frequency with word-final /s/ in Cali,
Colombia, but only when that sound is analyzed in the phonological context most
favorable to reduction. Brown and Raymond (2012) analyze the effect of contex-
tual ratio frequency on the variable outcome in Spanish of word-initial Latin /f/,
as some words have retained /f/ while others have lost it, for example, Span. favor
‘favor’ < Lat. favor versus Span. hablar ‘to talk’ < Lat. fabulare. Their analysis of
 The company that word-boundary sounds keep 

the Medieval Spanish play La Celestina (1499) provides support for the influence
of contextual ratio frequency by revealing that words that occurred more often in
the context for reduction are more likely to be realized as [Ø], with the grapheme
“h”, in Modern Spanish than are words that occurred infrequently in that context.
Brown (2015) also finds evidence in support of the significant influence of con-
textual ratio frequency from an analysis of word-initial /d/ in the speech of early
Spanish-English bilinguals in New Mexico. She proposes that different articula-
tions of /d/ in cognates and non-cognates in English and Spanish can be attributed
to variable rates of contextual ratio frequency.

Spanish /s/
Owing to the frequent occurrence of /s/ in comparison to other consonants as
well as the large variability of pronunciation of this sound in the Spanish-­speaking
world, “the aspiration and deletion of /s/ in dialects of Spanish may be the most
extensively treated of all sound changes being investigated from an empirical, vari-
ationist perspective” (Ferguson 1990: 64).1 Many other factors have been shown
to condition the realization of this sound, including linguistic, social, and usage-
based factors. The preceding and following phonological contexts have been
shown to exert a strong influence, perhaps the strongest, on the articulation of /s/,
with more reduction when /s/ is preceded by a non-high vowel as well as when it is
followed by a consonant (cf. Brown & Torres Cacoullos 2003; Brown 2004; Brown
2009a; Brown & Brown 2012). Prosodic stress has been shown to influence /s/
realization, with more reduction of /s/ in atonic syllables (cf. Alba 1982), as tempo-
ral and articulatory reduction could be considered part of the definition of atonic
syllables. Word length has been shown to condition /s/ articulation, with more
reduction of /s/ in longer words (cf. Terrell 1979). Register also influences /s/ artic-
ulation, with more retention in formal speech and reading tasks (cf. ­File-Muriel
2009). Syllable and word positions influence the realization of /s/ as well. In the
vast majority of dialects, /s/ is reduced more often in syllable-final position than in
syllable-initial position, and in word-final position than in word-initial position.2
It might be for this reason that syllable- and word-final /s/ is the more-studied
variable among the two. Further, speaking rate influences /s/, with more reduction

. Whether syllable- and word-final /s/ reduction in Modern Spanish represents a sound
change in progress or stable variation falls outside the scope of this paper and should be taken
up elsewhere.
. However, Lipski (1994: 209) proposes that some varieties of Colombian Spanish are
“unique in the Spanish-speaking world in reducing /s/ more frequently in syllable-initial than
in syllable-final position.”
 Earl K. Brown

in faster speech. It should be noted that differentiating between style and speaking
rate can be difficult, as some styles are consistently delivered quickly, for example,
play-by-play sports commentary (cf. Lipski 1985).
The realization of /s/ is also affected by social variables. The geographic origin
of speakers has a profound effect, with the Spanish of some regions exhibiting high
rates of reduction (e.g. Caribbean and Andalusian varieties), while other regions
exhibit low rates of reduction (e.g. Mexican and Andean varieties). Additionally,
the articulation of /s/ has been shown to be a social marker that differentiates
socioeconomic classes, with more reduction of /s/ among lower-class and less-
educated speakers, even in dialects with extensive reduction among all speakers,
for example, in the Dominican Republic (Alba 2004). In addition to educational
level, Terrell (1981) found task and gender to be significant predictors of /s/ real-
ization in Santo Domingo, Dominican Republic. Male speakers are more likely to
reduce /s/ than female speakers, and female speakers are more likely to retain /s/
when reduction is stigmatized (cf. Fontanella de Weinberg 1973). Finally, the age
of the speaker has also been shown to condition this sound, with more reduction
among younger speakers (cf. Guillén Sutil 1992).
In addition to linguistic and social factors, usage-based factors have been
shown to influence the articulation of Spanish /s/. As alluded to above, lexical
frequency conditions the segment’s realization, with, in general, more reduction
in higher-frequency words (cf. Bybee 2002; Brown 2009b; Brown 2005; Brown &
Torres Cacoullos 2002; Brown & Torres Cacoullos 2003; File-Muriel 2009; File-
Muriel 2010; Minnick Fox 2006). However, Brown (2009a) finds a ceiling effect for
lexical frequency, such that among the dialects with the highest levels of overall
syllable- and word-final /s/ reduction, the effect of lexical frequency is weakened
or even absent. This result may be attributable to the simple fact that as a sound
change nears completion, there is less variation to model. The effect of contex-
tual ratio frequency on Spanish /s/ has received less attention. Brown (2004) and
Raymond and Brown (2012) find that word-initial /s/ in the colonial Spanish of
northern New Mexico and southern Colorado is reduced more in words that occur
frequently in the phonological context that most favors reduction in that variety (a
preceding non-high vowel). Brown (2009a) finds a similar result for word-final /s/
in Cali, Colombia, but only when the following context favors reduction for that
variety (a following consonant). To our knowledge, these few studies represent the
entirety of the literature documenting the influence of contextual ratio frequency
on Spanish /s/, whether in word-initial or -final positions.
In much of the literature on Spanish /s/, a bipartite or tripartite classifica-
tion of the /s/ sound has been used, whereby the realization of /s/ is placed into
one of two or three categories. Studies often simply report that a given token of
/s/ was either “maintained” or “reduced”, while other studies utilize symbols of
 The company that word-boundary sounds keep 

the ­International Phonetic Alphabet, such as [s], [z], [h], or Ø. In these studies,
the process by which a given token of /s/ is placed in a category is accomplished
impressionistically, that is, researchers simply reported what they heard, or more
accurately, what they thought they heard. These impressionistic codes may be
not completely reliable, as problems associated with transcriber bias have been
documented in the literature (cf. Boucher 1994; Erker 2010; Pouplier & Goldstein
2005). Further, flanking phonetic context can influence the perception of sounds.
For example, Repp and Williams (1985) report that syllable-final stop consonants
are perceived more accurately when the preceding vowel is shorter in duration and
File-Muriel and Diaz-Campos (2003) report that the listeners in their experiment
with synthesized speech achieved an accuracy rate of only 46% when categorizing
Spanish /s/ before a pause.
In addition to discrepancies between what listeners perceive on the one hand
and the acoustic reality of the sound signal on the other, subtle differences in
sounds are lost when two sounds are placed into the same category. For example,
two tokens of /s/ that might be labelled [s] would likely have different durations
and spectral qualities, a reality that could be meaningful, but these differences
would be ignored when both tokens are categorized as [s]. Widdison (1995) shows
that subtle acoustic differences of Spanish /s/ influence perception in a lexical
decision task even though the listeners are unable to identify the differences. Erker
(2010) demonstrates that in Dominican Spanish in New York City, significant dif-
ferences in duration and center of gravity of /s/ exist in tokens of /s/ that are cat-
egorized as [s].
The effects of transcriber bias as well as the loss of nuanced information with
categorization have motivated recent studies of Spanish /s/ to rely on acoustic cor-
relates rather than impressionistic coding. As alluded to, Erker (2010) measured
the influence of various factors on the reduction of coda /s/ in Dominican ­Spanish
by analyzing the duration in milliseconds and the center of gravity in Hertz of this
sound. Similarly, File-Muriel and Brown (2011) rely on acoustic correlates to study
/s/ in Cali, Colombia by measuring duration in milliseconds, center of gravity in
Hertz, and the percentage of voicing of this sound. In this matter, reduction of /s/
is studied as a gradient variable rather than a categorical one, allowing for a more
detailed portrait of the variable effect of the independent variables cited in the
literature.

The present study

The present study analyzes factors that contribute to the reduction of word-
final /s/ in a sample of Mexican Spanish spoken in Salinas, California. The main
focus of the paper is to analyze whether contextual ratio frequency exerts a
 Earl K. Brown

s­ignificant influence on the articulation of this word-boundary sound. As

Mexican Spanish is a variety of Spanish that regularly maintains syllable- and
word-final /s/, using a categorical coding scheme would likely prove unhelpful
as the vast majority of the tokens would likely be labelled as “maintained” or
as [s]. Thus, a nuanced and gradient approach is adopted in this paper. This
is accomplished, following the precedent of Erker (2010) and File-Muriel and
Brown (2011), by analyzing the duration of /s/ in milliseconds and its center
of gravity in Hertz. These measurements serve as indicators of reduction, as
reduced articulatory gestures result in shorter durations and lower center of
gravity measurements of fricatives.
The research questions motiving this paper are:

RQ1. What effect, if any, does contextual ratio frequency have on the d
­ uration of
word-final /s/ in Mexican Spanish?
RQ2. What effect, if any, does contextual ratio frequency have on the center of
gravity of word-final /s/ in Mexican Spanish?
RQ3. In general terms, does this sample of word-final /s/ in Mexican Spanish
support the usage-based assertion that words that occur more frequently
in contexts that favor the reduction of their word-boundary sounds are
indeed reduced more than other words?

Data and methods

In order to analyze the influence of contextual ratio frequency on word-final /s/

in Spanish, a corpus of sociolinguistic interviews of Mexican Spanish spoken in
­Salinas, California (Brown 2012) was accessed. The majority of the interviews in the
corpus were recorded with a solid-state Edirol R-1 recorder and a lapel-mounted
Audio-Technica microphone in a sound studio of a university in central California.
A few of the interviews were recorded with a solid-state Tascam DR-05 recorder in
a classroom of the university or in the home of the participants. As is the norm with
sociolinguistic interviews, the topics of discussion varied based on the interests of
the participants. Often, the interviewers were the friends and family members of
the participants, a situation that is conducive to accessing speakers’ vernacular
(cf. Labov 1984: 29). Each interview lasted approximately 30 minutes in duration.
A subset of the corpus was utilized to extract the data for this paper. A sequen-
tial set of at least 100 tokens of word-final /s/ starting at least ten minutes into the
recording was extracted from ten interviews: five men and five women. The speak-
ers were born and raised in one of the three west-central Mexican states of Guana-
juato, Jalisco or Michoacan, before migrating to the United States, usually directly
 The company that word-boundary sounds keep 

to California. The ages at which the participants emigrated from ­Mexico varied
between 13 and 23 years (mean = 15.4, SD = 2.8) while their ages at the time of
recording ranged from 20 to 27 years (mean = 23.3, SD = 1.8). They belonged to
a homogeneous social group, as all had already completed a Bachelor’s degree or
were in the process of doing so at the time of recording. The decision to limit
sociodemographic variability among the speakers was deliberate in order to iso-
late the usage-based variables of interest in this paper, especially contextual ratio
frequency.
While viewing both a waveform and a spectrogram of the sound files with
the acoustic software Praat (Boersma & Weenink 2017), the duration of /s/ was
manually marked, or in the case of deleted tokens (N = 95), a note was made.
Specifically, high-level frequencies visible in the spectrogram, generally above
5,000 Hertz, and aperiodicity in the waveform were used to delimit maintained
/s/. Subsequently, a Praat script automated the extraction of the middle 50% of
the tokens of /s/ and applied a high-pass filter to eliminate the lowest 750 Hertz
from the sound signal before measuring the center of gravity in Hertz, with a
power setting of 2. The motivation for using only the middle 50% of the tokens
and for removing the lowest 750 Hertz before measuring the center of gravity
was to avoid a skewing effect that occurs when voicing is present, whether in
the transitions between /s/ and surrounding sounds or from any voicing present
in the /s/ itself (for discussion on removing the lowest frequencies of fricatives
before measuring the center of gravity, cf. Silbert & de Jong 2008: 2772). Deleted
tokens of /s/ were assigned a duration of 0 milliseconds and a center of gravity
of 0 Hertz.
Rather than enter the raw duration and center of gravity values into the statis-
tical models reported below, normalized values were used in order to control for
individual as well as gender-based variation. It is likely that specific participants
speak, on average, more quickly than others. Additionally, in general, men have
longer vocal folds than women, and therefore lower voices, a reality that results in
lower measurements of center of gravity. In order to control for these confound-
ing factors, centered and scaled z-scores were created per speaker by subtracting
the mean duration of /s/ of that speaker from his or her values of /s/ duration, and
then dividing those values by the standard deviation of duration for that speaker.
The resulting normalized duration values were centered around zero, with values
above zero representing duration values higher than the mean for that speaker,
and values below zero representing duration values lower than the mean for that
speaker. The same procedure was used to create normalized center of gravity val-
ues for each speaker.
In addition to the two dependent variables, the following independent vari-
ables were coded: the preceding and following phonological contexts, speaking
 Earl K. Brown

rate, prosodic stress, the sex of the speaker, the lexical frequency of the word with
/s/, and the frequency with which words with /s/ occur in the phonological context
most favorable to reduction of word-final /s/ in this variety of Spanish, a vari-
able referred to in the literature as contextual frequency, Frequency in a Reducing
Context (FRC), and Frequency in a Favorable Context (FFC) for reduction. In
order to highlight the proportional aspect of this measure, this paper uses “con-
textual ratio frequency” throughout. The two frequency variables were measured
in a combined corpus of Mexican and Mexican-American Spanish that included
the corpus from which the tokens in this study were taken (Brown 2012) as well as
the Mexico City portion of the Habla Culta corpus (Lope Blanch 1971), the Habla
Popular of Mexico City (Lope Blanch 1976), a corpus of Spanish in the Southwest-
ern United States (Lope Blanch 1990), and a handful of interviews of spontane-
ous speech recorded in Irapuato, Guanajuato by Jeffrey Turley of Brigham Young
University in the early 2000s as well as eleven interviews of the corpus El habla de
Monterrey [Mexico] (Rodriguez Alfano 2005). The number of words in this com-
bined frequency corpus is 592,195.
Contextual ratio frequency was calculated as follows. All occurrences of each
word type with /s/ in our dataset were retrieved in the frequency corpus. It was
subsequently determined, based on orthography, how often tokens of each word
type were followed by a consonant as opposed to other sounds and pauses. This
procedure placed each word type on a scale between 0% and 100% with values
near 0% representing word types that rarely occur before a consonant and con-
textual ratio frequency values near 100% indicating word types that frequently
occur before consonants. The decision to consider a following consonant as the
phonological context most favorable to reduction is based on the distribution of
word-final /s/ in this sample of Mexican Spanish. On average, following conso-
nants condition shorter durations and lower center of gravity values of word-final
/s/ in this variety of Spanish. See Table 1.

Table 1. Distribution of word-final /s/ by following phonological context

Following context Mean duration of /s/ Mean center of gravity

Consonant 53 ms 5068 Hz
Vowel 82 ms 6440 Hz
Pause 99 ms 6369 Hz

It is assumed that there must be a minimum number of occurrences of a word

type for contextual ratio frequency to have the potential of exerting an influ-
ence. As such, tokens of /s/ whose contextual ratio frequency was based on fewer
 The company that word-boundary sounds keep 

than three word tokens in the frequency corpus were excluded (N = 31).3 This
exclusion left 1,028 tokens of word-final /s/ from 235 word types for the analyses
reported below.
In addition to the independent variables mentioned, an interaction term
between contextual ratio frequency and the following context was also entered
into the statistical model reported below, as contextual ratio frequency is based on
the cumulative exposure to the following context. Finally, instead of the raw lexical
frequency values, logarithmic frequency values were entered into the models in
order to control for the hugely larger frequency values of the most frequent words
in comparison to all others (cf. Gries 2013: 254). This practice is motivated by the
Power Law of Practice, which proposes that “performance improves with practice
but the amount of improvement decreases as a function of increasing practice or
frequency” (Bybee 2010: 34, based on Anderson 1982). Two mixed effects linear
regressions were performed, one for duration and one for center of gravity, with
speaker and lexical item specified as random effects, specifically as varying inter-
cepts, using the R package lme4 (Bates et al. 2015). Additionally, the R package
lmerTest (Kuznetsova, Brockhoff & Christensen 2016) was employed to perform
a backwards selection procedure in order to eliminate non-significant variables
from the models.

Results

The results show that contextual ratio frequency is a significant predictor of

both duration and center of gravity of word-final /s/ in this sample of Mexican
Spanish. Interestingly, the results do not return a significant effect for lexical
frequency.

Duration
The results of the mixed effect linear regression of the duration of word-final /s/
report that following context, speaking rate, and the interaction term between
contextual ratio frequency and following context exert a significant influence on
the prediction of word-final /s/ duration in these data. See Table 2.

. It should be noted that, admittedly, it is very likely that over their lifetimes the speakers
have had more experience with the words that were eliminated than the frequency corpus
leads to believe.
 Earl K. Brown

Table 2. Mixed effects model of normalized duration of word-final /s/

Random effect variable Variance SD Groups

word 0.057 0.239 235

speaker 0.000 0.000   10
Residual 0.640 0.800
Fixed effect variable Estimate SE df t-value p-value Sign.
(Intercept) 0.813 0.162 749.875 5.034 ≤ 0.001 ***
fol. context = C −0.473 0.161 928.626 −2.945   0.003 **
fol. context = pause 0.731 0.253 966.869 2.89   0.004 **
speaking rate −0.03 0.005 1002.922 −5.94 ≤ 0.001 ***
contextual ratio freq. 0.004 0.005 639.565 0.807   0.42 ns
con. ratio freq. x fol. −0.013 0.005 948.974 −2.463   0.014 *
context = C
con. ratio freq. x fol. −0.008 0.009 996.671 −0.859   0.391 ns
context = pause
N = 1,028; sigma = 0.800; log-likelihood = −1277.61; AIC = 2575.221; BIC = 2624.574;
deviance = 2555.221; df residual = 1018

With the reference level of following context specified as a vowel, a follow-

ing consonant conditions a shorter duration of word-final /s/ while a following
pause causes a longer one. As speaking rate increases, duration of word-final /s/
decreases. Contextual ratio frequency as a main effect was not selected as signifi-
cant. However, within the interaction term with the following context, contextual
ratio frequency exerts a significant effect. Specifically, with vowel as the reference
level for following context, contextual ratio frequency conditions shorter dura-
tions of word-final /s/ when that sound is followed by a consonant. Nevertheless,
when followed by a pause, contextual ratio frequency does not exert a significant
influence on the duration of word-final /s/.4

. It should be noted that when the tokens of unrealized /s/, that is, tokens with a duration of
0 milliseconds (N = 95), are excluded from the mixed effects linear regression, contextual ratio
frequency is not selected as significant. However, when these tokens are binned as “absent”
and all other tokens are binned as “present” (N = 933), a generalized mixed effects logistic re-
gression detects a significant effect from contextual ratio frequency. Consequently, it is unclear
whether contextual ratio frequency operates gradiently on phonetics or categorically on pho-
nemes. Future studies might provide possible answers to this question. Thanks are expressed
to an anonymous reviewer for pointing out this possibility.
 The company that word-boundary sounds keep 

The interaction between contextual ratio frequency and following context as

well as the significant influence of contextual ratio frequency when word-final /s/
is followed by a consonant can be appreciated in visual form in Plot 1.

Following context
3 #
V
C
2

1
Normalized duration

-1

-2

0 25 50 75 100
Contextual ratio frequency (%)

Plot 1. Normalized duration of word-final /s/ by contextual ratio frequency and

­following context

As seen in the plot, the only context in which contextual ratio frequency signifi-
cantly influences the duration of word-final /s/ in these data is when that sound
is followed by a consonant, as that is the only context in which a horizontal line
cannot be drawn entirely within the 95% confidence interval indicated by the gray
bands around the regression line drawn through the tokens of word-final /s/ fol-
lowed by a consonant, indicated by red points. In contrast, the confidence intervals
around the regression lines for the other two contexts allow for a horizontal line
to be drawn, a fact that suggests the lack of a significant slope for those other two
contexts. Further, the interaction between contextual ratio frequency and following
context is also indicated by the general position of the three groups of tokens in
the plot. Tokens of word-final /s/ followed by a pause, indicated by yellow points,
generally cluster in the upper-left region of the plot, and none of these tokens have
a contextual ratio frequency value above 48%. Tokens of word-final /s/ followed by
a vowel, indicated by dark green points, generally cluster in the center-left region
of the plot, and none of these tokens have a contextual ratio frequency value above
80%. Finally, the tokens of word-final /s/ followed by a consonant, indicated by red
 Earl K. Brown

points, are more widely dispersed among the center and lower regions, and span
the entire contextual ratio frequency range, that is, up to and including 100%.
The independent variables eliminated during the backward selection process,
that is, the variables not selected as making a significant contribution to the pre-
diction of the duration of word-final /s/ in these data are: lexical frequency, stress,
preceding context, and gender.
In summary, these data show that contextual ratio frequency exerts a signifi-
cant influence on the duration of word-final /s/, but only when this sound is fol-
lowed by a consonant, the context most favorable to reduction in this variety of
Spanish. Interestingly, the other usage-based variable, lexical frequency, was not
selected as significant.

Center of gravity
The results of the mixed effect linear regression of the center of gravity of word-
final /s/ show that speaking rate and the interaction term between contextual ratio
frequency and following context significantly condition the prediction of center of
gravity of word-final /s/ in these data. See Table 3.

Table 3. Mixed effects model of normalized center of gravity of word-final /s/

Random effect variable Variance SD Groups

word 0.052 0.229 235

speaker 0.000 0.000   10
Residual 0.843 0.918
Fixed effect variable Estimate SE df t-value p-value Sign.
(Intercept) 0.458 0.182 749.325 2.565   0.01 *
fol. context = C 0.036 0.182 922.735 0.195   0.845 ns
fol. context = pause 0.071 0.287 974.304 0.248   0.804 ns
speaking rate −0.014 0.006 1013.275 −2.448   0.015 *
contextual ratio freq. 0.005 0.005 637.326 0.864   0.39 ns
con. ratio freq. x fol. −0.023 0.006 948.611 −3.759 ≤ 0.001 ***
context = C
con. ratio freq. x fol. −0.009 0.011 999.921 −0.873   0.383 ns
context = pause
N = 1,028; sigma = 0.918; log-likelihood = −1411.77; AIC = 2843.541; BIC = 2892.895;
deviance = 2823.541; df residual = 1018

As speaking rate increases, center of gravity of word-final /s/ decreases. As main

effects, neither following context nor contextual ratio frequency were selected
as significant. However, the interaction term between the two was selected as
 The company that word-boundary sounds keep 

significant. With the reference level of the following context set as a vowel, con-
textual ratio frequency has a significant effect on word-final /s/ when that sound
is followed by a consonant. However, when followed by a pause, contextual ratio
frequency has no significant effect.
The interaction between following context and contextual ratio frequency as
well as the significant influence of contextual ratio frequency when word-final /s/
is followed by a consonant can be appreciated in visual form in Plot 2.

Following context
2 #
V
C
Normalized center of gravity

-2

-4

0 25 50 75 100
Contextual ratio frequency (%)

Plot 2. Normalized center of gravity of word-final /s/ by contextual ratio frequency and
­following context

As seen in the plot, the only context in which contextual ratio frequency signifi-
cantly influences the center of gravity of word-final /s/ in these data is when that
sound is followed by a consonant, as that is the only context in which a horizontal
line cannot be drawn entirely within the 95% confidence interval indicated by the
gray bands around the regression line drawn through the tokens of word-final /s/
followed by a consonant, indicated by red points. Conversely, the confidence inter-
vals around the regression lines for the other two contexts allow for a horizontal
line to be drawn, a fact that suggests the lack of a significant slope for those other
two contexts. In this respect, the influence of the interaction between following
context and contextual ratio frequency when analyzing the center of gravity of
/s/ is similar to the influence of the interaction between these two variables when
analyzing the duration of /s/.
 Earl K. Brown

The independent variables eliminated during the backward selection process

were: lexical frequency, stress, preceding context, and gender.
In summary, contextual ratio frequency has a significant influence on the cen-
ter of gravity of word-final /s/, but only when this sound is followed by a conso-
nant. Also notable is that lexical frequency was eliminated during the backward
selection procedure and was not selected as significant.

Discussion

The first research question (RQ1) asks whether contextual ratio frequency has an
effect on the duration of word-final /s/ in Mexican Spanish. The results of this sam-
ple of Mexican Spanish allow this question to be answered in the affirmative, as the
interaction term between contextual ratio frequency and following context was
selected as a significant fixed effect variable in the mixed effects linear regression.
Specifically, contextual ratio frequency exerts a significant effect on word-final /s/
duration when this sound is followed by a consonant, the phonological context
most favorable to reduction in this variety of Spanish. Words that have occurred
frequently in the phonological context most favorable to reduction in this variety
of Spanish have a significantly shorter duration of word-final /s/.
The second research question (RQ2) inquires whether contextual ratio fre-
quency conditions the center of gravity of word-final /s/ in Mexican Spanish.
This question can also be answered in the affirmative, as the interaction term
between contextual ratio frequency and following context was again selected as
making a significant contribution to the prediction of center of gravity of /s/ in
these data.
The result that contextual ratio frequency exerts a significant influence only in
the phonological context favorable to reduction concurs with the result of Brown
(2009a). In that study of word-final /s/ in Cali, Colombia, it was found that contex-
tual ratio frequency is significant, but only in the phonological context favorable
to reduction, which was specified in that study as a following voiced consonant.
Specifically, when measured in favorable contexts for reduction, words with a high
contextual ratio frequency were reduced ten percentage points more often than
words with a low contextual ratio frequency, 73% versus 63% respectively, a dif-
ference that reached statistical significance in a chi-square test (χ = 15.22, df = 1,
p ≤ 0.001). However, when measured in contexts unfavorable to reduction, words
with a high contextual ratio frequency were reduced only two percentage points
more often than words with a low contextual ratio frequency, 23% versus 21%
respectively, a difference that did not reach statistical significance (χ = 0.47, df = 1,
p = 0.5). It should be noted that Brown (2009a) relied on categorization, as both
 The company that word-boundary sounds keep 

the dependent variable and contextual ratio frequency were categorical variables,
with levels “maintained” versus “reduced”, and “high” versus “low”, respectively.
In contrast, as detailed above, the present study utilizes two gradient dependent
variables of word-final /s/ (duration in milliseconds and center of gravity in Hertz)
and places contextual ratio frequency on a gradient scale between 0% and 100%.
Despite these methodological differences, it is noteworthy that in both studies
contextual ratio frequency exerts a significant conditioning effect only in the pho-
nological context favorable to reduction. This is especially true given the finding of
Brown and Alba (2017) that contextual ratio frequency conditions another voice-
less fricative, word-initial /f/, in Mexican Spanish only in phonological contexts
unfavorable to reduction. It is apparent that the interaction between contextual
ratio frequency and the phonological context upon which it is based should be
explored further in future studies.
The third and final research question (RQ3) queries whether, in general terms,
the analysis of word-final /s/ in this sample of Mexican Spanish supports the
usage-based assertion that words that occur more frequently in contexts that favor
reduction of their word-boundary sounds are indeed reduced more often than
other words. Again, as with the previous two research questions, this question
can be answered in the affirmative. Both dependent variables negatively correlate
with contextual ratio frequency, such that as contextual ratio frequency increases,
duration in milliseconds and center of gravity in Hertz decrease (dur: tau = −0.12,
z = −5.90, p ≤ 0.001; cog: tau = −0.12, z = −5.68, p ≤ 0.001). The results from both
the duration and the center of gravity of word-final /s/ in this sample of Mexi-
can Spanish suggest that the mental representation of words are malleable entities
whose word-boundary sounds are sensitive to the effects of the sounds that they
frequently co-occur with. In other words, this study shows that word-boundary
sounds are influenced by the company they keep.
As mentioned above, lexical frequency was not selected as making a signifi-
cant contribution to the prediction of word-final /s/ in these data. The question
arises as to the validity of either this result or the results of previous research that
have shown a significant effect from this variable on the articulation of word-final
/s/ in Spanish (e.g. Brown 2009b). It must be noted that a large difference between
previous studies and the present one is the lack of inclusion of contextual ratio
frequency in previous studies and its inclusion in this one. Could it be that the
previous research that found a significant effect for lexical frequency on word-
boundary sounds did so erroneously by not including contextual ratio frequency
in their analysis? While the results of the present study cannot support such a
claim, these results suggest that contextual ratio frequency is a better predictor
of word-boundary sound reduction than lexical frequency, at least on word-final
/s/ in Mexican Spanish. This assertion concurs with that of Brown and Raymond
 Earl K. Brown

(2012: 156) who, after analyzing the Modern Spanish outcome of word-initial
Latin /f/, conclude that “FFC [contextual ratio frequency] … is a more precise
measure and predictor of change than word frequency alone.”

Conclusions

The analysis of word-final /s/ in a sample of Mexican Spanish in Salinas, C

­ alifornia,
from five women and five men of the same socioeconomic class, provides evi-
dence that contextual ratio frequency plays a significant role in the reduction of
this word-boundary sound, but only when measured in the phonological context
favorable to reduction. This result concurs with that of Brown’s (2009a) study of
word-final /s/ in Cali, Colombia but contradicts that of Brown and Alba’s (2017)
study of word-initial /f/ in Mexican Spanish, as the latter study finds a signifi-
cant effect for contextual ratio frequency in unfavorable contexts. Consequently,
the interaction between contextual ratio frequency and the phonological context
upon which it is based, whether preceding or following context, depending on the
word position of the sound, should be addressed in future studies.
The results of the present study concur with previous studies that suggest that
contextual ratio frequency is a better predictor of word-boundary sound reduc-
tion than is lexical frequency. These results support Bybee’s (2002: 261) assertion
that words that occur more frequently in phonological contexts conducive to
reduction are reduced more often, even when the phonological context is brought
under statistical control, because the mental representation of words are malleable
cognitive entities which respond to usage-based factors.

References

Alba, Orlando. 1982. Función del acento en el proceso de elisión de la /s/ en la República
Dominicana. El español del Caribe, 15–26. Santiago, Dominican Republic: Universidad
Católica Madre y Maestra.
Alba, Orlando. 2004. Cómo hablamos los Dominicanos: Un enfoque sociolingüístico [Colección
Centenario]. Santo Domingo, Dominican Republic: Grupo León Jimenes.
Bates, Douglas, Mächler, Martin, Bolker, Ben & Walker, Steve. 2015. Fitting linear mixed-effects
models using lme4. Journal of Statistical Software 67(1): 1–48. doi: 10.18637/jss.v067.i01.
Boersma, Paul & Weenink, David. 2017. Praat: Doing Phonetics by Computer [Computer pro-
gram]. Version 6.0.28. <http://www.praat.org> (23 March 2017).
Boucher, Victor J. 1994. Alphabet-related biases in psycholinguistic inquiries: Considerations
for direct theories of speech production and perception. Journal of Phonetics 22(1): 1–18.
Brown, Earl K. 2009a. A Usage-based Account of Syllable- and Word-Final /s/ Reduction in Four
Dialects of Spanish [Lincom Studies in Romance Linguistics 62]. Munich: Lincom.
 The company that word-boundary sounds keep 

Brown, Earl K. 2009b. The relative importance of lexical frequency in syllable- and word-final /s/
reduction in Cali, Colombia. In Selected Proceedings of the 11th Hispanic Linguistics Sym-
posium, Joseph Collentine, Maryellen García, Barbara Lafford & Francisco Marcos Marín
(eds),165–178. Somerville, MA: Cascadilla Proceedings Project.
Brown, Earl K. 2012. Corpus of Mexican Spanish in Salinas, California [Electronic corpus].
<http://itcdland.csumb.edu/~eabrown/>
Brown, Earl K. & Alba, Matthew C. 2017. The role of contextual frequency in the articulation
of initial /f/ in Modern Spanish: The same effect as in the reduction of Latin /f/? Language
Variation and Change 29(1): 57–78.
Brown, Earl K. & Brown, Esther L. 2012. Syllable-final and syllable-initial /s/ reduction in Cali,
Colombia: One variable or two? In Colombian Varieties of Spanish, Richard File-Muriel &
Rafael Orozco (eds), 89–106. Madrid: Iberoamericana.
Brown, Esther L. 2004. The Reduction of Syllable Initial /s/ in the Spanish of New Mexico and
Southern Colorado: A Usage-based Approach. PhD dissertation, University of New Mexico.
Brown, Esther L. 2005. New Mexican Spanish: Insight into the variable reduction of “la ehe
inihial” (/s-/). Hispania 88(4): 813–824. doi: 10.2307/20063211
Brown, Esther L. 2015. The role of discourse context frequency in phonological variation:
A usage-based approach to bilingual speech production. International Journal of Bilingual-
ism 19(4): 387–406. doi: 10.1177/1367006913516042
Brown, Esther L. & Raymond, William D. 2012. How discourse context shapes the lexicon:
Explaining the distribution of Spanish f-/h words. Diachronica 29(2): 139–161.
doi: 10.1075/dia.29.2.02bro
Brown, Esther L. & Torres Cacoullos, Rena. 2002. ¿Qué le vamoh aher? Taking the syllable
out of Spanish /s/ reduction. In University of Pennsylvania Working Papers in Linguistics:
Papers from NWAV 30, Daniel Ezra Johnson & Tara Sanchez (eds), 17–32. Philadelphia PA:
­University of Pennsylvania Press.
Brown, Esther L. & Torres Cacoullos, Rena. 2003. Spanish /s/: A different story from beginning
(initial) to end (final). In A Romance Perspective in Language Knowledge and Use [Current
Issues in Linguistic Theory 238], Rafael Núñez-Cedeño, Luis López & Richard Cameron
(eds), 22–38. Amsterdam: John Benjamins. doi: 10.1075/cilt.238.05bro
Bush, Nathan. 2001. Frequency effects and word-boundary palatalization in English. In Bybee
& Hopper (eds), 255–280.
Bybee, Joan. 1985. Morphology: A Study of the Relation Between Meaning and Form. [Typologi-
cal Studies in Language 9]. Amsterdam: John Benjamins. doi: 10.1075/tsl.9
Bybee, Joan. 1998. Usage based phonology. In Functionalism and Formalism in Linguistics,
Vol. I: General Papers [Studies in Language Companion Series 41], Michael Darnell, Edith
Moravcsik, Frederick Newmeyer, Michael Noonan & Kathleen Wheatley (eds), 211–242.
Amsterdam: John Benjamins.
Bybee, Joan. 2001. Phonology and Language Use [Cambridge Studies in Linguistics 94].
­Cambridge: CUP. doi: 10.1017/CBO9780511612886
Bybee, Joan. 2002. Word frequency and context of use in the lexical diffusion of phonetically
conditioned sound change. Language Variation and Change 14: 261–290.
doi: 10.1017/S0954394502143018
Bybee, Joan. 2006. From usage to grammar: The mind’s response to repetition. Language 82(4):
711–733. doi: 10.1353/lan.2006.0186
Bybee, Joan. 2007. Frequency of Use and the Organization of Language. Oxford: OUP.
doi: 10.1093/acprof:oso/9780195301571.001.0001
 Earl K. Brown

Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: CUP.

doi: 10.1017/CBO9780511750526
Bybee, Joan & Hopper, Paul (eds). 2001. Frequency and the Emergence of Linguistic Structure.
[Typological Studies in Language 45]. Amsterdam: John Benjamins. doi: 10.1075/tsl.45
Bybee, Joan & Scheibman, Joanne. 1999. The effect of usage on degree of constituency: The
reduction of don’t in American English. Linguistics 37: 575–596. doi: 10.1515/ling.37.4.575
Du Bois, John W. 1985. Competing motivations. In Iconicity in Syntax [Typological Studies in
Language 6], John Haiman (ed.), 343–365. Amsterdam: John Benjamins.
doi: 10.1075/tsl.6.17dub
Erker, Daniel G. 2010. A subsegmental approach to coda /s/ weakening in Dominican Spanish.
International Journal of the Sociology of Language 203: 9–26.
Ferguson, Charles A. 1990. From esses to atiches: Identifying pathways of diachronic change.
In Studies in Typology and Diachrony for Joseph H. Greenberg [Typological Studies in Lan-
gauge 20], William Croft, Suzanne Kemmer & Keith Denning (eds), 59–78. Amsterdam:
John Benjamins. doi: 10.1075/tsl.20.06fer
File-Muriel, Richard. 2009. The role of lexical frequency in the weakening of syllable-final ­lexical
/s/ in the Spanish of Barranquilla, Colombia. Hispania 92(2): 348–360.
File-Muriel, Richard & Brown, Earl K. 2011. The Gradient Nature of s-Lenition in Caleño
­Spanish. Language Variation and Change 23(2): 223–243. doi: 10.1017/S0954394511000056
File-Muriel, Richard & Diaz-Campos, Manuel. 2003. Grammatical judgments and phonetic
reality: A study of internal constraints in phonological variation. The Journal of the Acous-
tical Society of America 114(4): 2335–2336. doi: 10.1121/1.4781065
File-Muriel, Richard J. 2010. Lexical frequency as a scalar variable in explaining variation.
­Canadian Journal of Linguistics/Revue Canadienne de Linguistique 55(1): 1–25.
doi: 10.1353/cjl.0.0065
Fontanella de Weinberg, Maria Beatriz. 1973. Comportamiento ante -s de hablantes femeninos
y masculinos del español bonaerense. Thesaurus 25(1): 12–22.
Givón, Talmy. 1979. On Understanding Grammar. New York NY: Academic Press.
Guillén Sutil, Rosario. 1992. Una cuestión de fonosintaxis: realización en andaluz de la /s/ final
de palabra seguida de vocal. Anuario de Estudios Filológicos 15: 135–153.
Hooper, Joan Bybee. 1976. Word frequency in lexical diffusion and the source of morphopho-
nological change. In Current Progress in Historical Linguistics, William M. Christie (ed.),
96–105. Amsterdam: North Holland.
Hopper, Paul. 1998. Emergent grammar. In The New Psychology of Language: Cognitive and
Functional Approaches to Language Structure, Michael Tomasello (ed.), 155–175. Mahwah,
NJ: Lawrence Erlbaum Associates.
Jurafsky, Daniel, Bell, Alan, Gregory, Michelle & Raymond, William D. 2001. Probabilistic rela-
tions between words: Evidence from reduction in lexical production. In Frequency and the
Emergence of Linguistic Structure [Typological Studies in Language 45], Joan Bybee & Paul
Hopper (eds), 229–254. Amsterdam: John Benjamins. doi: 10.1075/tsl.45.13jur
Kuznetsova, Alexandra, Brockhoff, Per Bruun & Christensen, Rune Haubo Bojesen. 2016.
lmerTest: Tests in Linear Mixed Effects Models [Computer software]. Version 2.0.33.
<https://CRAN.R-project.org/package=lmerTest> (3 December 2016).
Labov, William. 1984. Field methods of the project on linguistic change and variation. In Lan-
guage in Use: Readings in Sociolinguistics, J. Baugh & J. Sherzer (eds.), 28–53. Englewood
Cliffs NJ: Prentice Hall.
Labov, William. 1994. Principles of Linguistic Change: Internal Factors. Oxford: Blackwell.
 The company that word-boundary sounds keep 

Lipski, John. 1985. /s/ in Central American Spanish. Hispania 68: 143–149. doi: 10.2307/341630
Lope Blanch, Juan M. 1971. El habla de la Ciudad de México: Materiales para su estudio. México
DF: Universidad Nacional Autónoma de México.
Lope Blanch, Juan M. 1976. El habla popular de la ciudad de México: Materiales para su estudio.
México DF: Universidad Nacional Autónoma de México.
Lope Blanch, Juan M. 1990. El español hablado en el suroeste de los Estados Unidos: Materiales
para su estudio. México DF: Universidad Nacional Autónoma de México.
Minnick Fox, Michelle Annette. 2006. Usage-based Effects in Latin American Spanish Syllable-
final /s/ Lenition. PhD dissertation, University of Pennsylvania, Philadelphia PA.
Moonwomon, Birch. 1992. The mechanism of lexical diffusion. Paper presented at the Annual
Meeting of the Linguistic Society of America, Philadelphia PA.
Phillips, Betty. 2006. Word Frequency and Lexical Diffusion. Houndmills: Palgrave MacMillan.
doi: 10.1057/9780230286610
Pouplier, Marianne & Goldstein, Louis. 2005. Asymmetries in the perception of speech produc-
tion errors. Journal of Phonetics 33(1): 47–75. doi: 10.1016/j.wocn.2004.04.001
Raymond, William D. & Brown, Esther L. 2012. Are effects of word frequency effects of context
of use? An analysis of initial fricative reduction in Spanish. In Frequency Effects in Lan-
guage Learning and Processing, Stefan T. Gries & Dagmar Divjak (eds), 35–52. The Hague:
­Mouton de Gruyter. doi: 10.1515/9783110274059.35
Repp, Bruno H. & Williams, David R. 1985. Influence of following context on perception of the
voiced-voiceless distinction in syllable-final stop consonants. The Journal of the Acoustical
Society of America 78(2): 445–457. doi: 10.1121/1.392467
Rodriguez Alfano, Lidia. 2005. Investigacion sociolinguistica el habla de Monterrey: Su trayectoria
en una página electrónica. Monterrey, Mexico: Editorial Trillas.
Scheibman, Joanne. 2000. I dunno: A usage-based account of the phonological reduction of
don’t in American English conversation. Journal of Pragmatics 32(1): 105–124.
doi: 10.1016/S0378-2166(99)00032-6
Schuchardt, H. 1885. Über die Lautgesetze: Gegen die Junggrammatiker. Berlin: Oppenheim.
Silbert, Noah & de Jong, Kenneth. 2008. Focus, prosodic context, and phonological feature spec-
ification: Patterns of variation in fricative production. The Journal of the Acoustical Society
of America 123(5): 2769–2779. doi: 10.1121/1.2890736
Terrell, Tracy D. 1979. Final /s/ in Cuban Spanish. Hispania 62: 599–612. doi: 10.2307/340142
Terrell, Tracy D. 1981. Diachronic reconstruction by dialect comparison of variable constraints:
s-aspiration and deletion in Spanish. In Variation Omnibus, David Sankoff & Henrietta
Cedergren (eds), 115–124. Edmonton, Alberta: Linguistic Research.
Widdison, Kirk A. 1995. An acoustic and perceptual study of the Spanish sound change s > h.
Rivista di Linguistica 7(1): 175–190.
 Cumulative exposure to phonetic reducing
environments marks the lexicon
Spanish /d-/ words spoken in isolation

Esther L. Brown
University of Colorado Boulder

Reduced pronunciation variants of words commonly arise in discourse contexts

promoting lenition. Words differ in their likelihood of occurrence in reducing
contexts. We test whether words’ cumulative exposure to reducing environments
significantly predicts phonological reduction in experimentally elicited data.
Cumulative exposure is measured as the proportion of times words arise in
phonetic contexts promoting reduced pronunciation variants (FRC). Results of
instrumental analyses of the 503 Spanish word-initial /d-/ tokens produced by
18 native speakers show significant correlations between articulatory strength of
the onset consonant (d-) and words’ production histories (FRC), independent of
the production context. Additionally, linear modeling reveals a significant effect
of FRC in predicting onset intensity. Results suggest a cumulative effect on the
lexicon of words’ production histories.

1. Introduction

Language is shaped by the way in which it is used (Bybee 2010). A certain amount
of online reduction is expected in production due to neuromuscular requirements
of speaking (Raymond, Dautricort & Hume 2006), and studies of phonological
variation and change make clear that durational and segmental lenition are abun-
dant in speech (Ernestus 2014). Such variable forms of words, then, are “the natu-
ral result of the automation of linguistic productions” (Bybee 2002: 269). These
pronunciation variants of words that are the natural by-product of use are regis-
tered in the mental lexicon (Bybee 2001; Phillips 2006; Ernestus 2014).
Each episodic trace of a word is encoded with fine-grained acoustic and
articulatory detail, as well as information regarding speakers and words’ contexts
of use (Drager & Kirtle 2016). As such, the mental representation of every word
is comprised of a cloud of experienced tokens or exemplars and the strength of
representation of individual exemplar types is affected by patterns of use of each

doi 10.1075/slcs.192.06bro
© 2018 John Benjamins Publishing Company
 Esther L. Brown

word (Foulkes & Hay 2015: 303). Thus, lexical storage is affected not just by the
frequency with which speakers experience words in production and perception,
but also by the specific phonetic shape those words typically take (Johnson 1997).
Bybee (2001) details this view of the lexicon with the modified Exemplar Model of
lexical representation.
The phonetic shape and variable pronunciation of words is largely constrained
by the production context. Studies of phonological variation and/or change almost
categorically include factors of flanking sounds on target realizations because
neighboring sounds can shape target sounds through coarticulation and assimila-
tion (Fowler 2005). For instance, an increase in overlap or changes in the phasing
of articulatory gestures can effectively hide sounds acoustically (and perceptually).
Decreases in magnitudes of gestures can also yield variant forms in cases of target
undershoot whereby, for example, a stop ([d]) can be variably realized in a word
as a fricative or an approximant ([ð]) (Browman & Goldstein 1992: 173). Not all
instances of use of words, therefore, are equal. Many online factors (phonetic, con-
textual, extralinguistic) conspire to impact the phonetic shape words ultimately
take in use and, importantly, production contexts vary in the extent to which they
promote gestural reductions of sounds in words.
Clearly, then, contexts of use vary, and the ways in which contextual variation
shapes pronunciation variants is both widely studied and fairly well understood.
It is also the case, nevertheless, that words’ proportion of use in reducing con-
texts varies significantly (Brown 2013). That is, some words are used proportion-
ally more in contexts that promote reduction of segments, while others are used
predominantly in contexts unfavorable to reduction. A word used commonly in
a discourse context conducive to reduction logically will have more opportunity
to reduce than a word used frequently in contexts inhibiting reduction. Experi-
ence with words in reducing and non-reducing contexts accumulates in memory
and yields variant lexical representations. Such usage effects have the potential to
impact lexical representation by increasing or strengthening the number of reduced
exemplars stored in memory for certain words more than others (Bybee 2012).
In this way, subsequent articulations of words reflect both online effects of
phonetic contexts during articulation, as well as cumulative (lexicalized) effects
of words’ experience in specific discourse contexts affecting pronunciation (Bybee
2002; Drager & Kirtley 2016). The connection between experience and lexical rep-
resentations reflects a type of recursion principle (Guy 2014) or a feedback loop.
Articulatory reduction produces perceived exemplars that may serve as targets for
later articulations, which can then undergo further reduction.
Thus, if we accept the important role of production contexts on variation, and
acknowledge that variation is registered in memory, we must take seriously the
notion proposed by Bybee (2002) that words’ cumulative context histories need to
 Cumulative exposure to phonetic reducing environments marks the lexicon 

be considered in studies of variation and change. However, few studies attempt to

model the impact of cumulative usage in specific discourse contexts. Recent work
attempts to address this gap in the research. For instance, Brown and ­Raymond
(2012) and Raymond and Brown (2012) quantify this cumulative effect as the
proportion of times a word is used in discourse in a phonetic context conducive
to reduction, called its Frequency in a Reducing Context, or FRC. FRC has been
shown in synchronic and diachronic analyses based upon corpus data to be a bet-
ter predictor of phonological reduction than word frequency. However, it remains
to be explored whether the significant FRC effects reflect speaker experience with
reduced tokens of words, reflect online predictability effects from likelihood of
occurrence in context encouraging reduction, or reflect both these scenarios
(­Raymond, Brown & Healy 2016: 193–194).
That is, when speakers are planning an articulation of a word, that word’s like-
lihood of use in reducing contexts might unconsciously incite a speaker to plan a
reduced articulation (independent of the exemplar make-up of that word). Addi-
tionally, production can be biased toward specific variants as social and contextual
information can activate phonetic exemplars to which specific forms are indexed
(Drager & Kirtley 2016). Speakers, then, may be influenced to select an exemplar
that best conforms to the expectations of that production context. Such expecta-
tions reflect episodic linguistic and extralinguistic knowledge of that word’s usage
(Hay & Foulkes 2016; Drager & Kirtley 2016) in addition to the linguistic and
extralinguistic factors operative in the production context. Produced forms, then,
may not reflect the strongest or most numerous exemplars in memory, but rather
the most aptly suited to the context. It could be argued that we need not posit vari-
ant forms stored in memory. Rather, reduction rates could solely reflect predict-
ability of the context.
This study attempts to remove the expectations arising in contexts to deter-
mine whether there is evidence in support of the role of the lexical representation
of the word. It is the purpose of this work, therefore, to attempt to tease apart to the
extent possible, independent effects of lexical representations and predictability of
context to explore the role of lexicon (as decontextualized as possible) in phono-
logical variation (and, ostensibly, change). To do this, we remove any online con-
text from the equation by creating an elicitation task devoid of context; a word-list
style production task. Any remaining effects, we will argue, are indicative of lexical
representation as predicted by Bybee (2002).
We designed an experiment to test the effect of cumulative exposure to reduc-
ing environments (FRC) on decontextualized data immune from potential pre-
dictability effects found in naturally occurring discourse. The following outlines
in more detail the FRC theory and methods, followed by a brief discussion of
the dependent variable used in this study; variable realizations of the Spanish
 Esther L. Brown

­ ord-initial voiced, dental, obstruent /d-/. Section 3 outlines the data and meth-
w
ods used in the present project. Results are presented in Section 4. Lastly, conclu-
sions and discussion are presented in Section 5.

2. Background

2.1 Lexicalized effects of words’ context histories (FRC)

Studies of production and perception support the notion of a lexicon in which rich
linguistic, social, and contextual detail is accumulated for lexical items. Accumu-
lated episodic traces of words are shaped by specific production contexts. Factors
of production affecting lexical shape include online phonetic factors constraining
articulation, sociolinguistic factors of the context, as well as discourse-level fac-
tors such as predictabilities, repeated mention, and speech rate. Such factors act
together to impact the phonetic shape words ultimately take in use, and this expe-
rience with words in contexts accumulates in memory and yields variant lexical
representations. Such stored forms, then, can become planned targets for future
articulations, independent of production contexts. As noted by Hay and Foulkes
(2016), “an increasing number of lenited exemplars in the exemplar store shifts
the overall distribution of remembered forms, thus gradually increasing the likeli-
hood of selecting a lenited form as the target for speech production” (301).
One of the most frequently analyzed factors shaping lexical representations in
usage-based studies of phonological variation is word frequency. As Bybee (1999)
argues, each time a word is used in speech it is exposed to phonetically reducing
pressures inherent to articulation. Words used frequently (high frequency words) are
thus more vulnerable to the reduction born out of routinized articulatory gestures.
The lexical representation of high frequency words thus shifts as episodic traces of
reduced forms accrue, matching speaker experiences with words in production and
perception. Such effects of high frequency words leading reductive processes are
widely reported (see, for example, works cited in Hay & Foulkes 2016).
Less widely researched is a separate prediction made by Bybee regarding
the impact of words’ frequency of use in specific (leniting) discourse contexts on
variation and change. Such leniting contexts, Bybee argues, include sociolinguis-
tic factors such as informal speech styles which promote reduction (1999: 221),
or phonetically reducing contexts across sandhi which promote gestural over-
lap or undershoot (2002). In an analysis of word-final -t/d deletion in English,
Bybee (2002) predicts that words used frequently in discourse contexts promot-
ing deletion (pre-consonantal contexts in the case of this linguistic variable) will
delete more often overall, independent of the target production context. Indeed,
Bybee (2002) finds a significant effect of context histories on rates of deletion
 Cumulative exposure to phonetic reducing environments marks the lexicon 

whereby words used ­commonly in deleting contexts delete at a higher rate than
words used infrequently in contexts for deletion. She argues that a word’s pro-
pensity for use in reducing contexts logically yields more opportunity for reduced
articulations of that word. Speakers’ experiences leave episodic traces of reduced
forms in memory which shapes future realizations.
Despite recognizing a connection between factors constraining online pro-
ductions and episodic traces stored in memory, few studies include contextually
informed measures of words’ context histories beyond straightforward word fre-
quency counts. Studies that do measure words’ distributions in specific discourse
contexts, measuring either word classes’ frequency of use in reducing contexts
(Bybee 2002) or word types’ likelihood of use in reducing contexts (Raymond &
Brown 2012; Brown & Raymond 2012), find cumulative context histories predict
rates of reduction in natural speech corpora. Words, therefore, are produced in
line with their context histories, independent, it seems, of the production context.
Words frequently used in contexts promoting reduction reduce more readily (or
at a higher rate) than words used infrequently in discourse contexts promoting
reduction, even when analyzed in identical contexts.
The measure of the cumulative exposure of words to specific contexts (pho-
netic environments) that promote reduction is what has been called a word’s
Frequency of use in a Reducing Context (or FRC). Methodologically, FRC is a
measure of cumulative exposure to reducing environments: a ratio measuring a
word’s rate of occurrence in a context conducive to reduction out of that word’s
overall rate of occurrence (word frequency). The calculation attempts to capture
each word’s proportion of use in specific leniting contexts (see Brown 2015 for
more detail on FRC methods), which in turn is argued to be reflected, via episodic
traces, in the lexicon. What constitutes a reducing environment will depend upon
the variable under study (see for example Earl Brown 2008; Eddington & Channer
2010). The phonetic reducing environment for the object of study in this work is
described below.

2.2 The dependent variable: Spanish /d-/ lenition

This work uses as an object of study the word-initial realization of the voiced
dental obstruent /d/ in Spanish. Voiced stops in standard Spanish are said to
have two primary allophones in complementary distribution. The dental obstru-
ent is prescriptively realized as a stop [d] after pause, nasal, or /l/, as illustrated
in ­Example (1), and as an approximant [ð] elsewhere (Martínez-Gil 2014), as is
shown in (2).
(1) ¿Ud. nunca ha ido al dentista?  [Davies 2002-, Habla Culta: Havana]
You have never gone to the dentist?
 Esther L. Brown

(2) Su padre, su padre es dentista  [Davies 2002-, España Oral: CCOn034A]
Your father, your father is a dentist

Thus, in the discourse context of (1), a voiced, dental stop [d] is predicted owing to
the preceding phonetic context, whereas in (2), the voiced, approximant [ð] is the
anticipated realization. Words, then, are commonly realized with both onset types,
although the contexts for stop realizations are more limited (Barrutia & Schwegler
1994: 114; Teschner 2000: 96).
However, these realizations are variable (Hualde, Simonet, & Nadeu 2011).
Dialectological research of Spanish /d/ pronunciation attests wide variation both
across and within geographic areas as well as social groups in the Spanish speak-
ing world (Lipski 2011: 81–82). Variation includes distribution of the two primary
allophones with regard to the production contexts; stop articulations in contexts
prescriptively reserved for approximants and, conversely, approximant realiza-
tions where a stop is expected (Amastae 1989).
Additionally, recent work challenges the traditional binary description of the
articulations of /d-/ (Eddington 2011). Acoustic study suggests that rather than
two allophones for /d/ ([d], [ð]), a continuum of realizations exists ranging from
stops to deletions with differing degrees of constriction (Carrasco, Hualde &
­Simonet 2012: 151). Continuous measures of intensity are employed to calculate
the scalar variation in strength of the voiced /d/ in Spanish.
Despite findings indicating considerable regional, social, lexical and acoustic
variability of the voiced dental stop /d/ in Spanish, realizations are “conditioned
primarily by the preceding segment” (Carrasco, Hualde & Simonet 2012: 151).
Thus, although not uniform, the nature of the preceding segment in the discourse
context is largely determinative of the variant realized in natural speech. The
cumulative impact of the phonetic context, therefore, should mark the lexicon.
The FRC measurement used in this analysis, therefore, reflects words’ distribu-
tions in contexts promoting stop articulations (as Example 1) compared to those
promoting approximant articulations (Example 2) in standard Spanish. This is
described further in next section.

3. Data and methods

3.1 Speakers
The data used in this study form part of an ongoing project (Raymond & Brown
in preparation) in which a total of twenty-two native speakers of Spanish were
recorded. Data from eighteen of these speakers, selected at random, make up the
 Cumulative exposure to phonetic reducing environments marks the lexicon 

tokens used in this study. The 18 speakers hail from 8 different Spanish-­speaking
countries (Argentina, Bolivia, Colombia, Ecuador, Mexico, Peru, Spain, and
­Venezuela), which reflect distinct dialect zones (Amastae 1995; Canfield 1981;
­Lipski 1994). It can be assumed, therefore, that in spontaneous speech, these
speakers exhibit the type of variation discussed in Section 2 above. All of the sub-
jects work and reside near the University of Colorado and have native or near-
native proficiency in English as a simultaneous or second language. The subjects
range in age from 24–63, with the average age being 33 (N = 18, SD = 9.60). The
data is split for gender: there are nine male and nine female speakers.

3.2 Materials
All the results reported upon in this study are based upon acoustic analyses of
28 different Spanish /d-/ initial words read in isolation in a word-list style. Previ-
ous research has determined that multiple linguistic factors constrain realizations
of /d/ (and other voiced obstruents) in Spanish (as outlined in 2.2). To control
and loosely balance potential effects of previously identified linguistic factors, the
stimuli were created with attention paid to (1) following vowel, (2) stress pattern,
(3) number of syllables, and (4) word frequency. Fourteen pairs of words (N = 28)
were created that loosely balanced these factors (1–4), but which differed in the
cumulative measure that is the focus of this work, (5) FRC_approx. Each of these
linguistic factors is detailed below.
Previous research on stop articulations in Spanish has found that the quality
of adjacent vowels can significantly impact realizations (Cole, Hualde & Iskarous
1999). Harper (2014), for instance, in an acoustic analysis of 1,574 /d/ realizations
finds that the following non-high vowel (/a/) significantly favors approximantiza-
tion, presumably owing to the lower tongue height, while the high front vowel had
an inhibitory effect on approximantization. The initial /d-/ consonant in the 28
words used in this study precedes four different vowels (/a/, /e/, /i/, /o/) as well as
the glide in the diphthongs /ja/, /je/.
With regard to lexical stress, Carrasco, Hualde & Simonet (2012) find that for
all three voiced stops in Spanish (/b/, /d/, /g/) there are increased rates of approxi-
mantization in unstressed syllables. Eddington (2011) also notes that stressed syl-
lables inhibit lenition. There are both stressed and unstressed word types in the
list. The majority of the words have an unstressed onset, with just four word types
having a stressed initial syllable (dañan, daño, d ­ iablo, diario).
An attempt was made to match the 14 word pairs by number of syllables.
Previous studies of consonant reduction in Spanish (Terrell 1979; Brown & Brown
2012; Méndez Dosuna 1985) find increased rates of reduction with increased word
 Esther L. Brown

length. The words range in length from two syllables (for example deber ‘should,
ought to’) to five syllables (dominicana ‘Dominican’).
Word frequency (per million) was calculated for each of the 28 word types
using the oral data in the Corpus del español online (Davies 2002-), which repre-
sents 5,113,249 words. Previous work on Spanish /d/ articulations finds a positive
correlation between increased rates of lenition and word frequency (Bybee 2001,
2002; Waltermire 2010; Eddington 2011). A range of word frequencies are repre-
sented in the data set, from a very low frequency per million of .24 (despegó ‘took
off ’) to a frequency of 301 (difícil ‘difficult’). The average frequency per million of
the word list is 42 (N = 28, SD = 61.87).
Consistently, however, studies report that previous phonological context is the
factor that most significantly constrains variable realizations of /d/ across varieties
of Spanish (Carrasco, et al. 2012). Recall, for Spanish /d-/, a preceding homorganic
consonant (/n/,/l/)1 or a pause favors stop articulation. All other preceding con-
texts (consonantal or vocalic) prescriptively favor approximant [ð]. In this study,
all tokens are read in identical contexts. Each token is preceded by a pause, or
cessation of speech, of approximately two seconds (described in more detail below
in 3.3).
The allophonic distribution mentioned above underscores the significant
effect of the discourse context on realizations of /d-/ initial words in natural
speech. Whether a stop [d] or approximant [ð] is produced is largely determined
by the terminal sound of the preceding word (or by the preceding pause in the
case of post-pause or utterance-initial /d-/). Thus, a word used frequently in a
context for approximantization has more opportunity to be realized as [ð] than a
word used frequently in the non-reducing contexts (#, /l/, /n/, /m/). The exemplar
representation of that word, in turn, would be predicted to have more (or stron-
ger) approximant-initial exemplars than words used frequently in non-reducing
contexts which could serve as targets for future articulations, independent of the
production context. The potential effect of cumulative experience on the lexicon is
calculated as an FRC measure.
The exemplar make-up (the episodic traces) of each word is estimated based
upon typical discourse contexts of use in a large corpus of oral speech. Contexts
of use for each word in this study are calculated in the oral portion of the Corpus
del español (Davies, 2002–). A calculation of the number of instances of use in a
reducing context divided by the overall number of tokens of that word expresses
a percentage: 100, the lexical item categorically occurs in a context conducive to

. Both /n/ and /l/ undergo homorganic assimilation to following consonants in Spanish
(Martínez-Gil 2014: 112)
 Cumulative exposure to phonetic reducing environments marks the lexicon 

reduction and 0, the word never occurs in a reducing context (see Brown 2015
for more detail on the calculation). The measure, consequently, is not a frequency
count, but rather captures a ratio of occurrence in reducing contexts. The value
is each word’s probability of occurring in a context promoting an approximant,
labelled in this work as FRC_approx.
The twenty-eight words, thus, are selected to be loosely matched for these
linguistic factors (following vowel, stress pattern, number of syllables), but which
differ in their FRC_approx within pairs. For example, ostensibly similar words
despedí “I fired” and despegó ‘3rd sing. took off ’, which share commonalities in
form [following vowel /e/, stress pattern (unstressed syllable), number of syllables
(3)], differ in their patterns of use in discourse contexts that favor approximanti-
zation (FRC_approx). In this case, despedí is higher in both FRC_approx and word
frequency. However, an attempt was made to ensure that across pairs, not all high
FRC words were also higher in word frequency. Thus, in the pair debido (high
FRC_approx) / dejado (lower FRC_approx), the former is not higher in word fre-
quency than the latter. In this way, fourteen pairs of /d-/ initial words were chosen.
The full list of word types is listed in the Appendix with corresponding values.

3.3  Recording
The 28 /d-/ words that form the basis of this study were embedded randomly into a
list with 112 distractor words (non-/d-/ words). The list was transferred to a Power-
Point presentation with one word per slide. Two versions of the slides were created
that differed in order of presentation of the words in the list (Condition A, Condi-
tion B). Both conditions were generated randomly. Speakers read an introduction
in Spanish instructing them to read each word from the list aloud and then press
Return to advance to the next word. A two second pause was created before each
word appeared on the screen. Speakers’ productions of each word were recorded
in a sound attenuated booth in the University of Colorado Boulder Phonetics Lab.
Each word read one time by 18 different native Spanish speakers provides 504
tokens for analysis. One token is excluded for disfluency, bringing the total number
of examples used in this analysis to 503 instances of word-initial /d-/ for analysis.

3.4 Acoustic measurements

Each of the 503 /d-/ words was analyzed manually in Praat (Boersma & ­Weenink
2005). The onset consonant (where one is articulated), the vowel, and the coda
consonant (where one is present) of the first syllable in each token are delineated
to enable calculations aimed at determining strength of the initial obstruent
/d-/. The calculations include intensity measurements of consonants and vowel.
Previous studies have reported intensity difference between the onset ­consonant
 Esther L. Brown

and the following vowel as a determinant of degree of constriction (Cole et al.

1999; Ortega-Llebaria 2004; Eddington 2011; Carrasco, Hualde & Simonet
2012). Maximum, minimum and mean intensity measurements, indicating the
intensity of the speech signal, are calculated for each onset and following vowel
in Praat. Following Eddington (2011), we subtract intensity values of the onset
consonant from the intensity values of the following vowel. This provides the
Intensity Difference measurement used in this study. As Eddington (2011) notes,
“differences farther from zero indicate an approximant that is more stop-like,
while those closer to zero indicates [sic] highly lenited (or completely deleted)
phones that are more vowel-like” (6).
Additionally, for the linear model, each token of /d-/ was coded for linguistic
factors identified in previous studies, and summarized above, as constraining real-
izations of the voiced obstruent. These factors include (a) following phonological
environment: high front vowel /i/ (discoteca), glide /j/ tokens (diablo), mid front
vowel /e/ (dejado), central low vowel /a/ (dañan), and mid back round vowel /o/
(doctorado) (b) Stress: a binary classification of whether the onset syllable has lexi-
cal stress (stressed: diario) or not (unstressed: diecinueve), (c) syllable structure:
whether the onset syllable has a coda (closed: despegó) or not (open: dominicana),
(d) log of the word frequency per million (Davies 2002-), and (e) list position: The
28 randomly embedded /d-/ words were numbered by their order of appearance
in each of the two conditions. The following sections outline the results of the
analyses of /d-/ words investigating the cumulative impact of extralexical phonetic
contexts on words: specifically previous phonetic context promoting approximan-
tization of /d-/.

4. Results

4.1 Variation
The study is designed to enable an examination of decontextualized tokens of
Spanish /d/ initial words, while controlling for multiple factors known to impact
realizations. Contextual variables such as speech rate, stylistic variation, phonetic
context, and discourse topic are constrained by the experimental design. All tokens
are read once in isolation following several seconds of silence. Recall, the nature
of the data elicitation task would predict limited variation in the word-initial /d/
forms. The phonetic context, utterance-initial (or post-pause) context, is the least
conducive to reduction or approximantization of the stop (Eddington 2011). The
formality of the speech context (e.g. word list style) might encourage hyperar-
ticulation which is typical in laboratory speech (Nielson 2011; Recasens 2016), as
could the orthographic influence of [d] from the word list. The design stacks the
 Cumulative exposure to phonetic reducing environments marks the lexicon 

deck, so to speak, against variable forms. Despite this, based upon inspection of
the waveform and spectrogram, we find a total of five types of onsets.
In the read word-list style, the speakers produced canonical [d], prenasalized
[nd], voiceless stop [t], approximant [ð], and deleted onsets Ø. As an illustration
of onset variability, the following Figures 1–5 show the first syllable (/dis-/) of the
word discoteca (‘disco’). Tokens classified as canonical [d] articulations were deter-
mined by the appearance of a clear stop bar. These tokens characteristically have a
short period of prevoicing, a burst, and a short period of frication before onset of
the vowel. This is illustrated in Figure 1.

Figure 1. Stop [d] articulation in syllable /dis-/ of discoteca

In some cases the /d-/ words were realized with a prenasalized stop. For these
tokens a clear stop bar is observable. Voicing and weak formant structures are
observed prior to the burst. This is illustrated in Figure 2.

Figure 2. Prenasalized stop [nd] articulation in syllable /dis-/ of discoteca

Additionally, some stops, with clear bursts, were articulated with minimal voicing.
For these tokens there is no prominent voicing bar prior to the burst and they have
a negative voice onset time. As such, these articulations were labeled [t]. This can
be seen in Figure 3.
 Esther L. Brown

Figure 3. Voiceless stop [t] articulation in syllable /dis-/ of discoteca

Some onsets were realized with an absence of a burst closure. The initial /d-/ words
that lacked closure and which had formants with growing intensity into the vowel
were labeled approximants. This can be seen illustrated in Figure 4.

Figure 4. Approximant [ð] articulation in syllable /dis-/ of discoteca

Lastly, and unexpectedly, some tokens were realized with no discernible onset.
These examples were coded as having a fully lenited (deleted) obstruent. This is
illustrated in Figure 5.
This variation of onset types evident in the data holds across different words
and speakers. With regard to the 28 different word types used in the stimuli, all are
realized with a canonical voiced obstruent [d] (100%), and most words are also
realized with prenasalized, voiceless, or approximant onsets (89%, 93%, and 89%
respectively). Only in the case of deletions (Ø) are a minority of the words impli-
cated, with approximately a third being realized with no initial consonant (36%).
This is summarized in Figure 6.
 Cumulative exposure to phonetic reducing environments marks the lexicon 

Figure 5. Deleted onset (Ø) in syllable /dis-/ of discoteca

100 Words (N = 28)

% Words / Speakers

75 Speakers (N = 18)

50

25

0
[d] [nd] [t] [ð] Ø
Word-initial /d/ realizations

Figure 6. Distribution of onset types across words and speakers

Similarly, all speakers produce the prescriptive [d] as an onset for the words in
the list, and most use either a voiceless stop [t] or an approximant [ð] at least
once (72% and 67% respectively). Eight different speakers produced prenasalized
stops in the data, and seven different speakers completely deleted a word-initial /d/
(44% and 39% of the speakers). The variation described in this work, therefore, is
not restricted to specific lexical items or speakers.
Given the data elicitation task and design, the majority of tokens, not surpris-
ingly, were realized as a canonical, prescriptive, voiced obstruent. A total of 327 of
all tokens (65% of the data) were realized as [d], the most frequent form. Alterna-
tive realizations of the stop onset (pre-nasalized [nd] and voiceless dental stop [t])
together account for an additional 22% of the data. Nevertheless, in a production
context strongly disfavoring reduced forms, a full 13% of the data are reduced;
 Esther L. Brown

10% realized as approximant [ð] and 3% of the initial /d/ tokens not realized. This
is summarized in Table 1.

Table 1. Word-initial Spanish /d/ realizations read in word-list style

Realization N of tokens % of data Average FRC

STOPS
[d] 327 65 .66
[nd] 47   9 .67
[t] 66 13 .64
REDUCED FORMS
Approximant [ð] 50 10 .73
Deletions 13   3 .73
Total: 503 100

Given the production context (a decontextualized, read list of words) the degree
of variation is noteworthy. In naturally occurring speech, the pronunciation of
/d-/ words in utterance-initial or post-pause position would be constrained by
additional stylistic and discourse factors which have been minimized (if not elimi-
nated) in this study. The variation evident here, then, perhaps more straightfor-
wardly reflects exemplar make-up than the type of variation evident in spontaneous
speech, where factors such as speech rate, speech style, discourse predictability,
repetition, etc. all conspire to promote use of one variant or another. What then
predicts the realizations?
If we accept that the onset types reflect a cline of weakening, with the approxi-
mants and deletions indicative of lower articulatory effort than the stops, an interest-
ing correlation emerges. An independent-samples t-test was conducted to compare
FRC_approx for stop onsets ([d], [nd], and [t]) and reduced onsets ([ð] and Ø). There
is a significant difference in FRC_approx for stops (avg. = .66) and reduced onsets
(avg. = .73); t(501) = 2.01, p < .05. The reduced variants are words with a higher pro-
portion of use in discourse contexts promoting reduction (contexts other than post-
pause, nasal, or /l/). This result is in line with the assumption that such a usage pattern
would predict emergence of more reduced forms stored in memory as a result of
use. No significant difference is found between reduced onsets (avg. = 35.4) and stop
onsets (avg. = 42.3) for word frequency; t(501) = −.932, p = .35.

4.2 Variation & FRC_approx

As was outlined above, multiple variants of /d-/ are reported in this study. The rel-
ative articulatory strength of the variants mirrors the relative cumulative ­pattern
 Cumulative exposure to phonetic reducing environments marks the lexicon 

of use in reducing and non-reducing contexts of the /d-/ words. The stops are
realized in this study on words used significantly more often in discourse contexts
disfavoring approximantization. This result is in line with the suggestion that the
exemplar make-up of words used commonly in contexts disfavoring reduction
will likely have fewer reduced forms stored in memory.

4.2.1 Relative strength of [d] articulations

Intensity difference between the amplitude of the /d-/ consonant and the ampli-
tude of the following vowel is used as a continuous measure of constriction degree
and consonant strength (Eddington 2011). The greater the intensity difference, the
stronger the onset. If a word were frequently used in a discourse context promot-
ing approximantization, the strength and/or number of stored reduced exemplars
would be relatively higher than a word not commonly used in contexts for approx-
imant realizations. Thus, we could anticipate a correlation in the data between
higher FRC_approx and weaker [d] articulations (lower intensity differences), or,
put differently, stronger [d] articulations for words used more commonly in con-
texts for stop realizations (lower FRC_approx).
There is a significant negative correlation between FRC_approx and a measure
of intensity difference between the mean intensity of the onset consonant /d-/ and
the mean intensity of the following vowel [r(325) = −0.15, p = .007]. This is illus-
trated in Figure 7. A higher value (greater difference in intensity between the con-
sonant and vowel) indicates stronger closure.

FRC_approx
1.25
FRC_approx

0.75

0.5 R² = .0221

0.25

0
7.5 15 22.5 30
Intensity difference for [d] tokens

Figure 7. Significant negative correlation for intensity differences of [d] tokens and
FRC_approx

No significant correlation exists [r(325) = −0.08, p = .15] between word frequency

and intensity difference for these /d-/ tokens.
 Esther L. Brown

The other stop onset realization types [nd] (N = 47) and [t] (N = 66) were
examined individually. The correlation (r = −0.10) between intensity difference
and FRC_approx for [nd] tokens, although negative, was not significant (p = .32).
A measure of mean consonant intensity for the voiceless onsets was not possible
owing to the negative voice onset time. Thus, a measure of the burst amplitude was
calculated. The difference is calculated between the burst amplitude and the vowel
amplitude. Note, this is a different measure than the intensity difference used for
the other consonant onsets which calculated a mean amplitude of the segment.
The correlation (r = −0.05) for intensity difference for the voiceless consonants,
while also negative, is not significant (p = .11). These results are summarized in
Figure 8.

FRC_approx FRC_approx
1.25 1.25
FRC_approx FRC_approx
1 1

0.75 0.75
R² = 0.0021
R² = 0.0223 0.5
0.5
0.25 0.25

0 0
30 45 60 75 90
5 7.75 10.5 13.25 16 18.75
Intensity difference for [nd] Intensity difference for [t]
tokens tokens

Figure 8. Negative correlations for intensity differences of [nd], [t] tokens and
FRC_APPROX (n.s.)

4.2.2 Relative strength of approximant realizations [ð]

The same measure of intensity difference is applied to studies examining gra-
dient lenition of the voiced approximants in Spanish. Approximants [ð] with a
higher intensity difference between the onset and following vowel reflect greater
constriction and strength compared to [ð] articulations with low intensity dif-
ference. Thus, like stop articulations, a weaker onset is characterized by a lower
intensity difference. The measure FRC_approx used in this study estimates each
word’s cumulative exposure to the approximantizing context. The more a word
is used in such a context, the more practiced the [ð] onset becomes. As Bybee
(2001) notes, the pattern becomes practiced and can further reduce. As such, we
could predict a negative correlation between FRC_approx and intensity difference
for the approximants, as well.
As illustrated in Figure 9, there is a significant negative correlation between
FRC_approx and intensity difference between the mean intensity of the onset
 Cumulative exposure to phonetic reducing environments marks the lexicon 

c­onsonant [ð] and the mean intensity of the following vowel [r(48) = −0.34,
p = .02]. A higher value (greater difference in intensity between the consonant and
vowel) indicates stronger closure. Thus, this measure of strength of the [ð] also
supports the hypothesis regarding lexical impacts of context of use.

FRC_approx
1.25
FRC_approx
1

0.75

0.5
R² = 0.1182
0.25

0
0 7.5 15 22.5 30

Intensity difference for [ð] tokens

Figure 9. Significant negative correlation for intensity differences of [ð] tokens and
FRC_approx

No significant correlation exists (p = .29) between word frequency and intensity

difference for these [ð] tokens.

4.2.3  Linear modeling of onset variation

Although many factors typically having an impact on realizations in naturalistic
speech are controlled in the data elicitation employed here (eg. speech rate, previ-
ous phone), other linguistic factors could potentially promote stronger or weaker
articulations of the /d-/ words. To control for the possible independent effect of
such linguistic factors while exploring any role of FRC_approx in the variation, the
data are submitted to a linear mixed effects analysis. We employ R (R Core Team
2012) and the lme4 function (Bates, Maechler & Bolker 2015) to perform a lin-
ear mixed effects analysis fit by maximum likelihood of the relationship between
intensity ratios for [d], [nd], and [ð] and FRC_approx.
The range of variant possibilities ([d], [nd], [t], [ð], Ø) for the Spanish /d-/
words encumbers a unified measure of lenition across five onset types. The anal-
yses of intensity differences within onset types reveal significant correlations
between weaker articulations and higher FRC_approx (Figure 7, Figure 9). To
­calculate an intensity difference, by definition there needs to be a consonant onset
with an i­ntensity to measure. The deleted tokens (N = 13) have no ­discernable
onset and are thus not entered into the model. The voiceless stops [t] are also not
included (N = 66) because there is no mean intensity measure for these tokens,
 Esther L. Brown

only a burst intensity. The following analysis thus models the remaining 424
tokens of /d-/ words.
Other predictor variables entered into the mixed-effect model include lexical
factors specific to the word. These factors are following vowel, stress, presence/
absence of a coda consonant on the initial syllable, and the log of the word fre-
quency per million. To control for any potential elicitation effects, also included in
the analysis is the order in which the /d-/ word appeared in the elicitation stimuli
(list position). These factors are outlined in Section 3.4. We include speaker and
word as random effects. The following Table 2 summarizes factors determined to
significantly predict intensity ratio of onset consonants in the data set. The results
reported here reflect the best fit as determined by significant improvements in AIC
and BIC between models in a backwards stepwise regression.

Table 2. Final model predicting intensity difference (between onset & vowel) for [d],
[nd], [ð] tokens (N = 424)
Estimate Standard error

Intercept 11.91** (1.06)

Following Vowel (/a/)
/j/   3.05** (0.72)
/e/ −0.46† (0.94)
/i/ −1.11† (1.03)
/o/ −0.06† (1.01)
Realization([ð])
[d]   2.90** (0.48)
[nd]   2.47** (0.67)
FRC_approx −2.26** (0.73)
Stress (unstressed) −0.93† (0.83)
List position −0.03† (0.02)
Syllable (closed) −0.56† (0.39)
Log word frequency −0.35† (0.26)
Estimates and standard errors (in parentheses)
† p < .10
* p < .05
** p < .01

In the uniformly non-leniting context (word-list style, following a pause), the

nature of the following vowel predicts degree of constriction of the /d-/ onset.
A /d-/ word followed by a high front glide (which forms part of the diphthongs
/ja/ and /je/ in this study), has a significantly greater intensity difference ­compared
 Cumulative exposure to phonetic reducing environments marks the lexicon 

to the low central vowel /a/. The other following vowels that appear in this study
(/i/, /e/, /o/) do not differ significantly from /a/. A similar result is found in
Harper (2014).
Onset type also significantly predicts intensity difference in the experiment
words. Both [d] and [nd] have significantly higher intensity differences compared
to [ð].2 This result reflects the fact that both the [d] and [nd] have complete closure
in their articulations which diminishes the intensity of the onset consonant rela-
tive to the following vowel. The result is a greater intensity difference compared
to [ð].
Importantly, a significant main effect for FRC_approx is detected.3 A negative
relationship indicates that as FRC_approx increases (words with higher cumulative
patterns of use in contexts favoring approximantization), the intensity difference
decreases (indicating weaker onset consonants relative to vowel). Thus, in this data
set where minimal variation in form is predicted, the cumulative contextual his-
tories of the words significantly predict degree of closure (strength) of the onset
consonant. No other factors included in the analysis were significant.

5. Discussion

Results of instrumental analyses of the Spanish word-initial /d-/ tokens in this

study reveal a variety of onset types produced. A considerable portion of the data
(13%, N = 63) are realized without an initial stop, as is prescriptively expected in
utterance initial position and as would be expected in the artificially formal read,
word-list style. The onset types reflect a cline of articulatory strength. The weaker
onsets ([ð] and deletions Ø) are produced in words that on average have a higher
FRC_approx; words whose cumulative contexts of use reflect higher proportion of
use in discourse contexts promoting approximantization. This was summarized
in Table 1.
An examination within onset types finds significant correlations between
consonant strength (as measured by intensity difference) and FRC_approx. For
the words realized with a canonical voiced stop [d], there is a cline of strength
within these onsets such that as FRC_approx increases, strength of the constriction
decreases. The [d] articulations with the most complete closure (highest intensity
difference) are realized on words less often used in discourse contexts promot-
ing approximantization. There are gradient distinctions, therefore, within the stop

. A model that excludes realization type from analysis finds the same factors significant.
. No significant interactions are found with FRC_approx in any of the models.
 Esther L. Brown

category that mirrors cumulative discourse history. Similarly, for the [ð] tokens, a
significant correlation exists between the onset strength and the degree to which
those words are used in contexts promoting approximantization. The fine-grained
distinctions in approximant strength are also in line with FRC_approx.
A third result type that points to the importance of cumulative discourse con-
text as a significant predictor of phonological variation comes from the mixed
effect model. The strength of [d], [nd] and [ð] is significantly predicted by the
degree to which words are used in contexts promoting lenition. A significant nega-
tive effect was found between FRC_approx and intensity difference. No significant
effect of word frequency is found in any of the analyses.
These results show (i) a significantly higher FRC_approx in reduced onset types
compared to stops, (ii) significant correlations between FRC_approx and onset
strength within [d] and [ð] realizations, and (iii) a significant effect of FRC_approx
in linear modeling. Taken together, the results strongly support the hypothesis
outlined in Bybee (2002). The exemplar model of lexical representation (Bybee
2001) presumes our lexical representations proportionally reflect forms we experi-
ence in production and perception (­Eddington & Channer 2010). The exemplar
make-up, therefore, for words used frequently in discourse contexts promoting
approximantization (high FRC_approx words), can be assumed to have a greater
store of reduced forms in memory ­compared to words not as commonly used
in such discourse contexts (low FRC_approx words). When a speaker accesses the
lexicon to produce a high FRC_approx word, there is an increased likelihood that
a reduced onset form will be selected for production. Words’ histories yield pro-
duction patterns. This lexicalized effect is independent of the production context.
Unlike previous research that finds similar FRC results (Earl Brown 2008;
Eddington & Channer 2010; Brown & Raymond 2012; Raymond & Brown
2012), this work uses tightly controlled experimentally elicited data designed to
mitigate any potential effect of predictabilities. Thus, the significant relationship
between words’ cumulative proportion of use in context promoting reduction
(FRC) and degree of reduction produced in this data set can more straightfor-
wardly be argued to reflect the exemplar make-up of the stored forms. FRC find-
ings in previous research also implicate the lexicon, but reduced forms can be
argued to reflect the likelihood of occurrence in reducing contexts. That is, when
planning to produce a word, the probability of that word’s occurrence in reduc-
ing context might incite a speaker to plan a reduced form – even before the
­precise phonetic context is determined. Thus, without the artificiality of the data
elicitation used here, p
­ revious research notes it is difficult to adjudicate between
possibilities when context predictabilities are at play (Raymond, Brown & Healy
2016: 193–194).
 Cumulative exposure to phonetic reducing environments marks the lexicon 

This study, then, which quantifies an extralexical phonetically reducing

context (FRC_approx) provides experimental evidence for Bybee’s (2002) pre-
diction regarding the impact on the lexicon of words’ cumulative discourse pat-
terns of use. The ‘reducing context’ quantified in this work examines the effect
of the phonetic context (specifically, the context preceding the word-initial
/d-/). It is well known, however, that multiple factors (linguistic and extralin-
guistic) are brought to bear on pronunciation variation. Other factors of the
production context that impact degree of variation include discourse-level fac-
tors such as word and phone probabilities (Jurafsky, Bell, Fosler-Lussier, Girand
& Raymond, 1998; Jurafsky, Bell, Gregory & Raymond 2001a, 2001b), repeated
mention (Abramowicz 2007; Tamminga 2014), and speech rate (Gahl 2008;
Ernestus 2014).
A cumulative measure akin to FRC could capture words’ proportion of use in
contexts that promote reduction other than phonetic context (as done in this study),
such as words’ proportion of use in fast speech (Brown & Raymond 2014). Similarly,
Seyfarth (2014) examines word durations in words that are both typically predict-
able in speech (and have low-informativity) and typically unpredictable (and are as
such labeled high-informativity). Words that are low in informativity have shorter
durations in spontaneous speech compared to high-informativity words. Seyfarth
(2014) examines 41,167 word durations and finds low-­informativity words are sig-
nificantly shorter in all contexts (predictable or not). Similar to the interpretation of
the Spanish /d-/ results presented here, Seyfarth (2014) concludes that results sup-
port “representational models in which reduction is stored, and where sufficiently
frequent reduction biases later production” (140).
In addition to phonetic and contextual factors, sociolinguistic literature
demonstrates how social and stylistic factors significantly constrain phono-
logical forms and can equally be considered part of a ‘reducing context’. The
effect of stylistic and sociolinguistic factors of the production context on vari-
ant forms is also amply documented in sociolinguistic research (Tagliamonte
2006). A word used proportionally more often, therefore, in informal contexts
or within a speech community leading a reductive change, would register the
effect of this reducing context in the lexicon, which could then shape subse-
quent variant productions. Indeed, Hay & Foulkes (2016) propose such a feed-
back loop. Specifically these authors examine word-medial /t/ realizations in
New Zealand English to explore potential word level effects of cumulative use
among younger (more innovative) speakers. Hay and Foulkes find a statisti-
cally significant effect of words’ distribution among more innovative speakers
whereby the “population distribution of words affects representation, which in
turn affects perception and…production” (2016: 321–322). Thus, it seems as
 Esther L. Brown

Foulkes & Hay (2015: 304) note, “our e­ xperience with words extends far beyond
statistics regarding f­requency of use. The non-random association between
phonetic forms and social groupings, or social meanings, can also have long-
term consequences for the storage of words”.
Thus, independent lines of research are measuring lexicalized effects of
cumulative patterns of use in contexts promoting reduction. The factor used as
a measure of reduction can be phonetic (e.g. Raymond, Brown & Healy 2016),
discourse contextual (e.g. Seyfarth 2014), as well as extralinguistic (e.g. Hay &
Foulkes 2016). Each study demonstrates that patterns of use leave an imprint on
the lexicon (via episodic traces in memory). As Hinskens, Hermans & Oosten-
dorp (2014: 1) note,
usage-based models assume that language users store detailed phonetic infor-
mation about the words of their language each time that they are exposed to
them. These models stipulate redundant mental storage of bundles of maxi-
mally concrete articulatory, acoustic, grammatical, semantic and pragmatic
information concerning single occurrences (‘tokens’ or ‘exemplars’) of lexical
items, along with characteristics of both the speaker and the situation, organ-
ized in ‘clouds’.

These exemplars capturing cumulative experiences with words, in turn, shape

future articulations.
The factors constraining variation, therefore, that form the basis of myriad
studies of phonological variation and change, can be understood to not only have
an online effect in the production context (favoring or disfavoring, for instance, a
reduced form), but rather these same factors (phonetic, contextual, social/stylistic)
can be measured in a cumulative fashion (such as done here with the FRC_approx
calculation). This work adds to a growing body of literature, therefore, that sup-
ports a more multidimensional measurement of word histories. As proposed by
Bybee (1999, 2002), it seems that what needs to be central to studies of variation
and change is frequency+. That is, measures that move beyond “the simplistic fre-
quencies of (co-)occurrence” (Gries 2012: 5), and instead attempt to capture more
precisely speakers’ experiences with words.

Acknowledgements

I would like to thank Dr. Scarborough, Director of the University of Colorado Boulder Phonet-
ics Lab, for use of the sound booth. I also thank Devin Grammon and Alex McCallister for
assistance in data collection and coding. Additionally, I would like to acknowledge significant
contributions to this line of research by William D. Raymond. All errors are mine.
 Cumulative exposure to phonetic reducing environments marks the lexicon 

References

Abramowicz, Łukasz. 2007. Sociolinguistics meets Exemplar Theory: Frequency and recency
effects in (ing). Papers from NWAV 35. University of Pennsylvania Working Papers in
­Linguistics 13(2): 27–37.
Amastae, Jon. 1989. The intersection of s-aspiration/deletion and spirantization in Honduran
Spanish. Language Variation and Change, 1, 169–183. doi: 10.1017/S0954394500000053
Amastae, Jon. 1995. Variable spirantization: constraint weighting in three dialects. Hispanic Lin-
guistics 6: 267–286.
Barrutía, Richard & Schwegler, Armin. 1994. Fonética y fonología españolas. New York NY: John
Wiley & Sons.
Bates, Douglas, Maechler, Martin, Bolker, Ben & Walker, Steven. 2015. lme4: Linear Mixed-
Effects Models Using Eigen and S4. R package version 1.1–10. <http://CRAN.R-project.org/
package=lme4>
Boersma, Paul & Weenink, David. 2005. Praat: Doing Phonetics by Computer, version 5.2.18
[Computer program]. <http://www.praat.org> (1 March 2011).
Browman, Catherine P. & Goldstein, Louis. 1992. Articulatory phonology: An overview.
­Phonetica 49:155–180. doi: 10.1159/000261913
Brown, Earl. 2008. A Usage-based Account of Syllable- and Word-final /s/ Reduction in Four
Dialects of Spanish. PhD dissertation, University of New Mexico.
Brown, Earl K. & Brown, Esther L. 2012. Syllable-final and syllable-initial /s/ reduction in Cali,
Colombia: One variable or two? In Linguistic Studies in Colombian Varieties of Spanish,
Richard File-Muriel & Rafael Orozco (eds), 89–106. Madrid: Iberoamericana.
Brown, Esther L. 2013. Word classes in studies of phonological variation: Conditioning fac-
tors or epiphenomena? In Selected Proceedings of the 15th Hispanic Linguistics Symposium,
Chad Howe, Sarah Blackwell & Margaret Lubbers Quesada (eds), 179–186). Somerville
MA: Cascadilla Proceedings Project.
Brown, Esther L. 2015. The role of discourse context frequency in phonological variation:
A usage-based approach to bilingual speech production. International Journal of Bilingual-
ism 19: 387–406. doi: 10.1177/1367006913516042
Brown, Esther L. & Raymond, William D. 2012. How discourse context shapes the lexicon:
Explaining the distribution of Spanish f-/h- words. Diachronica 29(2): 139–161.
doi: 10.1075/dia.29.2.02bro
Brown, Esther L. & Raymond, William D. 2014. Contextual frequency effects in Spanish pho-
nology. Paper presented at the Georgetown University Round Table, Washington DC,
March.
Bybee, Joan L. 1999. Usage-based phonology. In Functionalism and Formalism in Linguistics,
Vol. 1 [Studies in Language Companion Series 41], Michael Darnell, Edith Moravcsik,
Frederick J. Newmeyer, Michael Noonan & Kathleen Wheatly (eds), 212–242. Amsterdam:
John Benjamins. doi: 10.1075/slcs.41.12byb
Bybee, Joan L. 2001. Phonology and Language Use. Cambridge: CUP.
doi: 10.1017/CBO9780511612886
Bybee, Joan L. 2002. Word frequency and context of use in the lexical diffusion of phonetically
conditioned sound change. Language Variation and Change 14: 261–290.
doi: 10.1017/S0954394502143018
 Esther L. Brown

Bybee, Joan L. 2010. Language, Usage and Cognition. Cambridge: CUP.

doi: 10.1017/CBO9780511750526
Bybee, Joan L. 2012. Patterns of lexical diffusion and articulatory motivations for sound change.
In The Initiation of Sound Change: Perception, Production, and Social Factors [Current
Issues in Linguistic Theory 323], Maria Josep Solé & Daniel Recasens (eds), 211–234.
Amsterdam: John Benjamins. doi: 10.1075/cilt.323.16byb
Canfield, D. Lincoln. 1981. Spanish Pronunciation in the Americas. Chicago IL: University of
Chicago Press.
Carrasco, Patricio, Hualde, José Ignacio & Simonet, Miquel. 2012. Dialectal differences in
­Spanish voiced obstruent allophony: Costa Rican vs. Iberian Spanish. Phonetica 69: 149–179.
doi: 10.1159/000345199
Cole, Jennifer, Hualde, José Ignacio & Iskarous, Khalil. 1999. Effects of prosodic and segmental
context on /g/-lenition in Spanish. In Proceedings of the Fourth International Linguistics and
Phonetics Conference, Osamu Fujimura, Brian D. Joseph & Bohomil Palek (eds), 575–589.
Thessaloniki: University Studio Press.
Davies, Mark. 2002–. Corpus del Español (100 million words, 1200s–1900s), <http://www.
corpusdelespanol.org>
Drager, Katie & Kirtley, Joelle. 2016. Awareness, salience, and stereotypes in Exemplar-Based
models of speech production and perception. In Awareness and Control in Sociolinguistic
Research, Anna Babel (ed.), 1–24. Cambridge: CUP.
Eddington, D. 2011. What are the contextual phonetic variants of /β ð ɣ/ in colloquial Spanish.
Probus 23: 1–19. doi: 10.1515/prbs.2011.001
Eddington, David, & Channer, Caitlin. 2010. American English has go? a lo? of glottal stops:
Social diffusion and linguistic motivation. American Speech 85: 338–351.
doi: 10.1215/00031283-2010-019
Ernestus, Mirjam. 2014. Acoustic reduction and the roles of abstractions and exemplars in
speech processing. Lingua 142: 27–41. doi: 10.1016/j.lingua.2012.12.006
Foulkes, Paul & Hay, Jennifer B. 2015. The emergence of sociophonetic structure. In The Hand-
book of Language Emergence, Brian MacWhinney & William O’Grady (eds), 292–313.
­Malden MA: John Wiley & Sons.
Fowler, Carol. 2005. Compensation for coarticulation reflects gesture perception, not spectral
contrast. Perception and Psychophysics 68(2): 161–77. doi: 10.3758/BF03193666
Gahl, Susanne. 2008. “Thyme” and “time” are not homophones: The effect of lemma frequency
on word durations in spontaneous speech. Language 84: 474–496. doi: 10.1353/lan.0.0035
Gries, Stefan T. 2012. Introduction. In Frequency Effects in Language Learning and Processing,
Stefan T. Gries & Dagmar Divjak (eds), 1–6. The Hague: Mouton de Gruyter.
doi: 10.1515/9783110274059.1
Guy, Gregory R., 2014. Linking usage and grammar: Generative phonology, exemplar theory,
and variable rules. Lingua 142: 57–65. doi: 10.1016/j.lingua.2012.07.007
Harper, David. 2014. An Analysis of Perceptual Factors in the Evolution of Spanish Approxi-
mants. PhD dissertation, University of Colorado, Boulder CO.
Hay, J. & Foulkes, P. 2016. The evolution of medial /t/ over real and remembered time. Language
92(2): 298–330.
Hinskens, Frans, Hermans, Ben & van Oostendorp, Marc. 2014. Grammar or lexicon. Or:
Grammar and lexicon? Rule-based and usage-based approaches to phonological variation.
Lingua 142: 1–26. doi: 10.1016/j.lingua.2014.01.005
 Cumulative exposure to phonetic reducing environments marks the lexicon 

Hualde, José Ignacio, Simonet, Miquel & Nadeu, Marianna. 2011. Consonant lenition and phono-
logical recategorization. Laboratory Phonology 2(2): 301–329. doi: 10.1515/labphon.2011.011
Johnson, Keith. 1997. Speech perception without speaker normalization: An exemplar model. In
Talker Variability in Speech Processing, Keith Johnson & John W. Mullennix (eds), 145–165.
San Diego CA: Academic Press.
Jurafsky, Daniel, Bell, Alan, Fosler-Lussier, Eric, Girand, Cynthia & Raymond, William D. 1998.
Reduction of English function words in Switchboard. Proceedings of the International Con-
ference on Spoken Language Processing 7: 3111–3114.
Jurafsky, Daniel, Bell, Alan, Gregory, Michelle & Raymond, William D. 2001a. The effect of lan-
guage model probability on pronunciation reduction. Proceedings of ICASSP-01 2: 801–804.
Jurafsky, Daniel, Bell, Alan, Gregory, Michelle & Raymond, William D. 2001b. Probabilistic rela-
tions between words: Evidence from reduction in lexical production. In Frequency and the
Emergence of Linguistic Structure [Typological Studies in Language 45], Joan L. Bybee &
Paul Hopper (eds), 229–254. Amsterdam: John Benjamins. doi: 10.1075/tsl.45.13jur
Lipski, John. 1994. Latin American Spanish. London: Longman.
Lipski, John. 2011. Socio-phonological variation in Latin American Spanish. In The Handbook
of Hispanic Sociolinguistics, Manuel Díaz-Campos (ed.), 72–97. Malden MA: Blackwell.
doi: 10.1002/9781444393446.ch4
Martínez-Gil, Fernando. 2014. Main phonological processes. In The Handbook of Hispanic Lin-
guistics, José Ignacio Hualde, Antxon Olarrea & Erin O’Rourke (eds), 111–132. Malden
MA: John Wiley & Sons.
Méndez Dosuna, Julian. 1985. La duración de S en los grupos SP, ST, SK: A propósito del orden
regular de difusión en algunos cambios fonéticos. In Symbolae Ludovico Mitxelena Septua-
genario Oblatae, Jose L. Melena (ed.), 647–655. Vitoria: Universidad del Pais Vasco.
Nielsen, Kuniko. 2011. Specificity and abstractness of VOT imitation. Journal of Phonetics 39:
132–142. doi: 10.1016/j.wocn.2010.12.007
Ortega-Llebaria, M. 2004. Interplay between phonetic and inventory constraints in the degree of
spirantization of voiced stops: Comparing intervocalic /b/ and intervocalic /g/ in Spanish
and English. In Laboratory Approaches to Spanish Phonology [Phonology and Phonetics 7],
Timothy L. Face (ed.), 237–253. Berlin: Mouton de Gruyter.
Phillips, Betty S. 2006. Word Frequency and Lexical Diffusion. New York NY: Palgrave Macmillan.
doi: 10.1057/9780230286610
Pierrehumbert, Janet B. 2001. Exemplar dynamics: Word frequency, lenition, and contrast. In
Bybee & Hopper (eds), 137–157.
R Development Core Team. 2012. R: A Language and Environment for Statistical Computing.
Vienna: R Foundation for Statistical Computing. <http://www.R-project.org/> (1 April
2011).
Raymond, William D., Dautricourt, Robin & Hume, Elizabeth. 2006. Word-medial /t-d/ dele-
tion in spontaneous speech: Modeling the effects of extra-linguistic, lexical, and phono-
logical factors. Language Variation and Change 18: 55–97. doi: 10.1017/S0954394506060042
Raymond, William D. & Brown, Esther L. 2012. Are effects of word frequency effects of context
of use? An analysis of initial fricative reduction in Spanish. In Frequency Effects in Lan-
guage, Vol 2: Learning and Processing, Stefan T. Gries & Dagmar S. Divjak (eds), 35–52. The
Hague: Mouton de Gruyter. doi: 10.1515/9783110274059.35
Raymond, William D. & Brown, Esther L. In preparation. Cumulative contextual effects on
­initial /d-/ articulations in Spanish.
 Esther L. Brown

Raymond, William D., Brown, Esther L. & Healy, Alice. 2016. Cumulative context effects and
variant lexical representations: Word use and English final t/d deletion. Language Variation
and Change 28: 175–202.
Recasens, Daniel. 2016. The effect of contextual consonants on voiced stop lenition: Evidence
from Catalan. Language and Speech 59: 139–161.
Tagliamonte, Sali. 2006. Analysing Sociolinguistic Variation. Cambridge: CUP.
doi: 10.1017/CBO9780511801624
Tamminga, Meredith J. 2014. Persistence in the Production of Linguistic Variation. PhD
­dissertation, University of Pennsylvania.
Terrell, Tracy. 1979. Final /s/ in Cuban Spanish. Hispania 62: 599–612. doi: 10.2307/340142
Teschner, Richard V. 2000. Camino oral: Fonética, fonología y práctica de los sonidos del español,
2nd edn. New York NY: McGraw-Hill.
Waltermire, Mark. 2010. Variants of intervocalic /d/ as markers of sociolinguistic identity among
Spanish-Portuguese bilinguals. Spanish in Context 7: 279–304. doi: 10.1075/sic.7.2.06wal

Appendix. /d-/ initial word types with FRC_APPROX and frequency per million*
Proportion of
use in leniting Word frequency
Pair Word ­context – FRC_approx per million

1 desespera – (s/he dispairs)    .88   1.89

1 demandaron – (they demanded)    .80   1.18
2 directo – (direct)    .60 53.86
2 difícil – (difficult)    .92 300.96
3 domingo – (Sunday)    .52 120.01
3 doscientas – (two hundred)    .57 31.42
4 diabéticos – (diabetic, diabetics) 1   0.94
4 diecinueve – (nineteen)    .82 39.69
5 debido – (due to)    .59 47.96
5 dejado – (left)    .77 77.25
6 dañan – (they harm)    .80   1.18
6 daño – (damage)    .50   4.72
7 diario – (daily)    .40 52.44
7 diablo – (devil)    .24   8.03
8 dominicana – (Dominican)    .92   3.07
8 doctorado – (doctorate)    .30 23.62
9 destino – (destiny)    .37 30.47
9 dentista – (dentist)    .40   5.91
10 decoración – (decoration)    .89 12.76
10 departamento – (department)    .20 92.37
11 discoteca – (disco) 1   3.07
 Cumulative exposure to phonetic reducing environments marks the lexicon 

Proportion of
use in leniting Word frequency
Pair Word ­context – FRC_approx per million
11 diccionario – (dictionary)    .12 18.19
12 distancia – (distance)    .96 34.25
12 diciembre – (December)    .80 65.67
13 dejar – (to leave)    .87 121.19
13 deber – (ought to, should)    .50 22.21
14 despedí – (I fired, said ‘bye’    .88   1.89
14 despegó – (3rd sing. took off) 1   0.24
Average    .66 42.02
* Calculations made in oral portion of Corpus del español (Davies 2002-).
 A usage-based account for the historical
reflexes of ain’t in AAE

K. Aaron Smith
Illinois State University

This paper is a comparative and historical study of the grammatical status of

auxiliary ain’t in African-American English (AAE). Drawing on the representation
of AAE speakers in works of fiction from the 19th century and present-day, the
paper confirms that the broad range of uses of auxiliary ain’t are nearly all available
in historical AAE. The study also presents data that certain of the uses (specifically
those not documented or only infrequently documented for non-AAE varieties)
have increased in fictive representations of AAE. The analysis shows that the
increase is led by a specific verb type (and even a specific verb) and thus suggests
an exemplar-based increase and extension of ain’t. Moreover, the paper also
shows a formal morphosyntactic convergence of several uses of auxiliary ain’t and
motivates that convergence through exemplar merging, suggesting an association
of that merged morphosyntactic form with sociolinguistic identity.

1. Introduction

The form ain’t has an infamous reputation in the English language as a malum in
se instance of bad usage (see Donauer & Katz 2015 for a recent collection of papers
on ain’t and its difficult history and current status in English). Despite relentless
efforts on the part of would-be language purists to eradicate it, ain’t has shown
admirable resilience. Dating back to at least the 18th century, ain’t appears in vari-
eties of English from all parts of the world.
In this paper, I will be concerned mostly with the use of ain’t in African-
American English (AAE), for which there is a robust literature (DeBose 1994;
DeBose & Faraclas 1993; Fasold & Wolfram 1975; Howe 1997; Howe & Walker
2000; Rickford 1977; Singler 1998; Weldon 1994, 2007; Winford 1998; Wolfram
& Schilling-Estes 2003, inter alia). Much of the work on AAE ain’t has centered
on diagnosticity, as defined in Poplack (2000: 17–18). In that sense, diagnosticity
is concerned with the way in which a given form might reveal its historical ori-
gins. Thus the morphosyntactic patterning of ain’t in AAE has been compared to
similar forms in creoles to argue for a creole origin (e.g. DeBose & Faraclas 1993;

doi 10.1075/slcs.192.07smi
© 2018 John Benjamins Publishing Company
 K. Aaron Smith

DeBose 1994) or to earlier English to argue for an English origin (e.g. Howe &
Walker 2000).
The current paper is also concerned with the diagnosticity of ain’t in AAE,
albeit on a smaller diachronic scale and for very different purposes. The data col-
lected here do not allow a full vetting of previous hypotheses and arguments con-
cerning the origins of AAE or AAE forms and instead accepts many of the findings
on the grammatical status and historical changes involving ain’t in AAE. The study
reported on here seeks to offer a usage-based explanation for those grammatical
and historical patterns. On the one hand, the data in this study confirm that all
of the uses of ain’t in Present-day African American English (PDAAE) appear in
nineteenth-century AAE (19-AAE). Furthermore, nearly all of the PDAAE uses of
ain’t were also available in nineteenth century non-AAE English (19-nonAAE) in
the southern US, thus suggesting longer English origins for most of the AAE uses
of ain’t.1 However, the data presented here will also support a divergence scenario
in which certain uses of ain’t, which although available in 19-AAE and only mar-
ginally available or not at all available in 19-non-AAE southern varieties, appear
to have increased in PDAAE, not only in terms of overall frequency but also in
terms of the types of verbs that occur with auxiliary ain’t in its various uses.2 Thus,
the use of ain’t is a good instance of the fact that PDAAE has become more dif-
ferent from other varieties of American English, and even from earlier varieties of
­African American English (Bailey & Maynor 1989; Rickford 1992).
The spread and increased use of ain’t appears on the surface to be a classic case
of secondary grammaticalization, that is the extension of a grammatical pattern
into a widening set of contexts (Traugott 2002; Kuryłowicz 1975). Such spread and
increase of an already grammatical item present a number of theoretical issues.
For one thing, when we look at the specific pattern of spread and increased fre-
quency for those AAE uses of ain’t that most support the divergence hypothesis,
we find that certain lexical verbs appear to “lead the pack”. Taking a usage-based
approach to the development of auxiliary ain’t in AAE, I argue in this paper that

. DeBose 1994 argues that ain’t in certain of its uses indicates a bisystemic configuration
for AAE, with sources in both an earlier creole and “Mainstream American English” (DeBose’s
term)
. While it can be shown from these data that certain uses of ain’t occur more frequently out
of roughly the same number of instances of ain’t collected within a given time period, which
certainly suggests increased frequency, strictly speaking one would need to consider all the
potential sites for ain’t usage to determine whether more of those sites are in fact expressed
with ain’t in later corpora (see Walker 2005). Since the fact of the increase of ain’t is well es-
tablished, the current paper’s focus is on a socio-functionally motivated explanation for the
distribution of ain’t auxiliary in AAE.
 A usage-based account for the historical reflexes of ain’t in AAE 

certain facts about that history are best accounted for within an exemplar model of
networked representations for language storage (Bybee 2006, 2010). In that model,
a speaker’s experience with a specific construction (i.e. phonetic form, types of
verbs, etc.) is stored in memory, and grammar then emerges (Hopper 1987) from
the network of stored exemplars and changes because of the attraction of newer
tokens to that exemplar, or cluster of exemplars.
However, the analysis of auxiliary ain’t in AAE is not finished here because its
place in the overall grammar of AAE presents another interesting phenomenon.
Secondary grammaticalization represents a later stage of grammatical development
and one that often sets the stage for renewal (Hopper & Traugott 2003), that is for
the development of new constructions to take over certain of the earlier grammati-
cal uses of an aging construction (Smith 2006). But, there does not appear to be any
such renewal for the tense/aspect distinctions lost as ain’t has spread and become
more frequent. Thus, the paper also argues for an analysis of AAE ain’t as an aux-
iliary underdetermined for tense/aspect and offers a usage-based explanation for
that undertermination analysis. As part of that analysis, I suggest that the general
motivation for the loss of tense/aspect specificity is a trade-off between distinctive-
ness in a certain part of the grammar of AAE and sociolinguistic identity, an iden-
tity that is also part of a speaker’s knowledge of the use of ain’t.

2. Linguistic status of ain’t in English overall

The historical source for ain’t in English is disputed. Limitation of space will not
allow a complete review of the various arguments, but see Anderwald (2002),
Cheshire (1981), Jespersen (1940), McDavid (1941), Palacios Martinez (2010),
Smith (2015), and Stevens (1954) for some proposals and reviews. In one view,
ain’t derives from various BE + not contractions (e.g. am not > amn’t) while in
others ain’t is believed to have developed from HAVE + not contracted forms
(e.g. hasn’t or haven’t), or some blend/merger of the two. Whatever its ulti-
mate historical origin, ain’t appears for all contracted forms of BE + not and
HAVE + not in their auxiliary uses, as shown in (1a–1d), taken from various
song titles: examples are easy to find in nearly any naturalistic corpus containing
colloquial speech.
(1) a. Progressive (BE + not): You ain’t goin’ nowhere (Song title by Bob Dylan)
b. Passive (BE + not): It ain’t called ‘Heartland’ for nothin’ (Song title by
Clay Walker)
c. Copular (BE + not): This ain’t a love song (Song title by Bon Jovi)
d. Perfect (HAVE + not): You ain’t seen nothing yet (Song title by
Bachman Turner Overdrive)
 K. Aaron Smith

In this paper, and following Payne (2010: 261ff.), I will include copular uses of
be, and thus copular uses of ain’t, as auxiliaries, since the latter’s syntactic behav-
ior is like that of an auxiliary verb in English (and see Smith 2015 for the more
detailed arguments of auxiliary status for copular ain’t), and since in sentences in
which be is the only verb, its function is to provide a site for inflection (Radford
1997: 61). For example, ain’t, like be, does not require do-support in negation; it
inverts with the subject in closed interrogatives, Ain’t you here yet? (again without
do-support); it is copied as the verb in tag-question formation in response to BE,
He’s your son, ain’t he?; and, it is stressed to emphasize the predication, He AIN’T
the boss! Additionally, ain’t cannot substitute as the main lexical verb for have not
in its possession meaning even in those varieties of UK English where have not,
or its contracted forms, can occur as the main predicate (Cheshire 1981: 366), as
shown in (2). Based on these syntactic facts, I consider ain’t to be an auxiliary verb
in those varieties of English in which it is used.
(2) He hasn’t any money on him right now.
*He ain’t any money on him right now.

While all of the uses given in (1a–d) above are wholly in accord with the putative
historical auxiliary sources for ain’t as derviving from BE not and HAVE not, or
some merger of the two, we will see in the next section that in some varieties of
English, auxiliary ain’t has an even wider distribution so that it has extended into
simple past and simple present contexts where it varies with DO not (doesn’t/don’t)
and DID not (didn’t).

2.1 Distribution of ain’t in AAE

In AAE, auxiliary ain’t may appear in the same contexts as those given above in
(1a–d), shown for AAE sources in 3a–d (see the appendix and Section 4 on the
sources for the current study).
(3)
a. Progressive: “Come here. I ain’t gon’ bite ’chew girl, dag,” the boy said
with a smile. (Omar Tyree, Flyy Girl, e-book)
b. Passive: Mrs. Poole say a baby ain’t born loving its mother. (Connie
Rose Porter, Imani all mine, e-book)
c. Copular: Well I’ll tell you, men ain’t shit, (Tu-Shonda Whitaker, The
Ex-Factor, e-book)
d. Perfect: You ain’t got no manners (Omar Tyree, Flyy Girl, e-book)3

. The example here is structurally perfect but semantically present and possessive. While
treated as a separate category in Weldon (1994), here it is combined with the perfect since the
current study is concerned with the structural historical reflexes of ain’t.
 A usage-based account for the historical reflexes of ain’t in AAE 

However, in addition to these uses, AAE also shows use of a “simple past” ain’t,
as in (4), in which case it varies with didn’t. This use is well documented in the
literature on ain’t in AAE (Wolfram & Fasold 1975: 69; Baugh 1988; Labov 1972;
Rickford 1977; DeBose & Faraclas 1993; Weldon 1994; Howe & Walker 2000).
(4) But I wanted to go. … Me and Imani ain’t get to go though. 
 (Connie Rose Porter, Imani all mine, e-book)
African American English also shows a “simple present” ain’t, as in the first
instance of ain’t in (5), in which use it potentially varies with don’t.
(5) That chick ain’t know shit about bein’ no queen and she damn sure ain’t
never treat me like no princess. 
 (Meesha Mink, Shameless hoodwives, e-book)
The uses in (4)–(5) are of particular interest for this study because they are unre-
lated to the putative historical sources of ain’t as contractions of BE + not and
HAVE + not, and therefore apparently represent extensions of the use of ain’t. In
sections 4 and 5, I will discuss the historical uses of ain’t where it will be shown that
although these uses were available in the earliest sources for AAE (but also other
varieties of American English), the frequency of present and past ain’t appears to
have increased. Before doing so, I will discuss the sources and methods used to
collect the data for this study.

3. Method and sources

The approach to the diachronic portion of this paper is not unlike diachronic
work carried out on older stages of many languages. In order to explore the his-
torical patterns of ain’t in AAE (as well as adjacent non-AAE varieties of southern
US English), I collected data from the on-line collection of mostly 19th-century
Southern literature available at the University of North Carolina’s Library website,
Documenting the American South (DAM) <http://docsouth.unc.edu>. The study
of past stages of a language is notoriously fraught with difficulties. Either one must
work from reconstructed languages or parts of languages, or, if one is fortunate,
there may exist written records of the language to work from. In the latter case,
however, there are still many problems since the records will be in the form of
written language, which almost certainly contains “idealized” forms caused by
standardization interference, constraints of print/web media or authorial imagina-
tion. Again, this is an issue that is to be found in historical language study broadly
and not only for the data collected in the current study. For these data, I have col-
lected tokens of usage of ain’t from works of creative fiction – mostly because of
accessibilty.
 K. Aaron Smith

While literary data may be objected to on the basis that authors fancifully
represent forms in dialect, which certainly may be true, the problem is not lim-
ited to historical (or even present-day) literary data. Regarding the use of the Ex-
Slave narratives, which on the surface would appear to be primary data on early
AAE, Poplack (2000: 3) makes the valid point that part of the difficulty in using
such data is “disentangling the linguistic system of the transcribers from that of
the speakers.” This problem is, obviously also a concern when using literary data.
While no source will ever present a complete picture, the data used for this study
is validated in that it comports generally with data collected from other sources
and reported on in the literature on AAE ain’t. In fact, the data for this paper has
turned up uses of ain’t, like the present tense uses illustrated in (5), that do not
appear in more traditional sociolinguistic data collection methods (Weldon 1994,
and see Howe & Walker 2000), despite its not infrequent use in AAE, although
the form does show up in other kinds of historical AAE material, like enclave
ex-slave community speech (see Howe & Walker 2000: 113). Again, the current
study accepts many of the findings on the grammatical status and spread of ain’t in
AAE and, as mentioned above, the data collected here support the findings of the
several other studies cited throughout this paper. Given the compatibility of these
data with those from other studies, the tokens compiled here are apprised as legiti-
mate for the usage-based hypotheses developed in the latter parts of this paper. As
with all hypotheses, further testing and analysis will help to evaluate the claims.4
The data taken from DAM was used to compile two databases; sources for the
historical data collected for this study are given in Appendix A. One database is
of AAE-speaking characters containing 300 tokens of ain’t. The other database is
a collection of 253 tokens from the representation of non-AAE speech (all white
southern characters/narrators). In order to compare like corpora, the present-day
data on AAE (PDAAE) was collected also from a corpus of fiction writings (in
this case exclusively by African American writers, although of course the level of
fluency in AAE on the part of any of the authors cannot be assessed), and not sur-
prisingly, authors in the present-day data show varying degrees of AAE features,
a fact which is also true of AAE users generally. The goal was to search texts until
about 300 tokens could be found. It will be noted that the amount of text needed
to arrive at that number varied. Non-AAE instance of ain’t were very difficult to
find and only 253 tokens are considered here. To achieve 300 tokens of ain’t from
19th-century AAE, more text was necessary than it was for present-day AAE. As I
point out later, this fact also supports claims for the increased frequency of certain
of the uses of ain’t in AAE diachronically.

. I will also mention that the use of literary texts to study the grammar of English and
creoles in diasporic communities is not without precedent. See for example Sebba 2004.
 A usage-based account for the historical reflexes of ain’t in AAE 

Each token in both the present-day and historical corpora was categorized
based on its putative historical origin, that is as a reflex of a BE + not contraction
or HAVE + not contraction, albeit sometimes with some distinctions in the more
specific type of auxiliary construction involved. On the one hand, the progressive,
copula and passive are categories where ain’t varies with the present-day reflexes
of its putative source BE + not. (Note that the analysis here is not concerned with
construction-internal variation to the degree it is in Weldon 1994 or Howe & Walker
2000). On the other hand, the perfect is a category where ain’t potentially varies with
the present-day reflexes of its putative source HAVE + not.5 Tokens were also coded
for those that appear in contexts not related to the putative historical sources, i.e.
those that potentially vary with didn’t (which I referred to earlier as “past tense ain’t”)
and those that potentially vary with don’t (referred to earlier as “present tense ain’t”).
In addition, another category of ain’t in potential variation with BE + not is
included: the existential construction. That construction has two variants, one
with the deictic there (or a variant thereof, th’/they,6 etc.), as in (6):
(6) They ain’t nothin’ thet a woman with a hairpin can’t do  (McEnery p. 57)
and, one without there, as in (7):
(7) Oh, sho, chile, afar ain’t nothin’ in dat lof ’ ‘cep’ de ‘taters  (Bonner p. 108)
As I will show below, the construction in (7) is another use of ain’t, in addition to
present and past ain’t, that increases in PDAAE.
Tag questions were also collected separately in order to test whether they coded
for all reflexes of ain’t. In my data, all tags coded for BE + not and HAVE + not
predicates. I will not have anything more to add about tag constructions in this
paper, yet they are still presented as a separate, but not frequent category. Finally,
one use of ain’t in the PDAAE data that was not in the historical data is an existen-
tial use of it ain’t (and see Green 2002: 80 on that use). As it only occurs once in my
data, I will not have anything more to say about it here.

4. Data

Perfect 100 30%

Progressive 51 17%
Copula 94 29%
Passive   5 .016%
(Continued)

. Compound perfect constructions like those involving perfect-progressive forms are

coded as perfect in these data.
. They in this context is taken to be the non-rhotic variant of there.
 K. Aaron Smith

(Continued)
Past 24 .08%
Present   6 .04%
Existential with ‘there’ 16 .05%
Existential without ‘there’   1 .003%
Tag   3 .01%
19th century AAE (n = 300)
Perfect 55 22%
Progressive 31 12%
Copula 123 48%
Passive   3 .011
Past   2 .007
Present   0 .0
Existential with ‘there’ 29 14%
Existential without ‘there’   0 0
Tag   7 .027%
19th century non-AAE (N = 253)
Perfect 41 .14%
Progressive 56 .18%
Copula 90 .30%
Passive   4 .013%
Past 61 .20%
Present 21 .09%
Existential with ‘there’   4 .013%
Existential without ‘there’ 16 .05%
It ain’t existential   1 .003%
Tag   6 .03%
PDAAE (N = 300)

5. Discussion

5.1 General
The data in Section 4 show that all of the uses of ain’t in PDAAE correspond-
ing to the putative historical sources of ain’t from BE + not and HAVE + not aux-
iliary uses were available in 19-AAE and 19-non-AAE. The data also show that
both present- and past-tense ain’t, forms not derivable from the putative historical
 A usage-based account for the historical reflexes of ain’t in AAE 

sources, although available in 19th-century AAE, both increase, past tense ain’t
more drastically, a finding that comports with Howe & Walker (2000: 120) and
Labov (1972: 284). As mentioned earlier in a footnote, these data can only suggest
an increase in frequency, but the increase scenario is supported when one consid-
ers that in order to arrive at 300 tokens of ain’t, significantly less text was required
from the PDAAE period than that for the 19th-century AAE, a point made earlier
as well. While present-tense ain’t makes a moderate increase from.04% in histori-
cal AAE to .09% in PDAAE, past-tense ain’t increases from .08% to .20%.7 From
those data, we can assert that the origin for both in PDAAE is historical AAE.
However, we see from the data that past tense ain’t also appears in the non-AAE
historical data, an example of which is given in (8).
(8) “I was thar all day and I ain’t seed you raise no special dust.” (Glasgow, p. 318)

Uses of past tense ain’t in historical and present-day non-AAE varieties of southern-
­US English are attested (Feagin 1979), and although the current study is not cen-
trally interested in the diagnosticity question beyond the establishment of PDAAE
uses in historical AAE, the historical and present-day occurrences of past tense ain’t
in non-AAE suggest English origins (and see Howe & Walker 2000). The absence
of present tense ain’t in historical non-AAE, on the other hand, suggests a different
origin, however, perhaps creole, I point a will return to below.
Another use of ain’t that shows increase in PDAAE is the existential use with-
out there. In many varieties of English, both AAE and non-AAE, the use of ain’t
is available in existential constructions with the deictic element “there,” or some
form of it, as illustrated above in (6). However, the existential construction with
bare-ain’t only appears once in 19-AAE data, the example of which is given in (7).
However, whereas it only occurs once in the 19-AAE data, it appears 18 times in
the PDAAE corpus, as in the example in (9).
(9) Ike?” Starr snapped. “Ain’t no Ike over here 
 (Whitaker, The Ex Factor: A Novel, ebook)

Similar uses of ain’t to that in (9) are attested in creoles, like Gullah, an example
of which is given in (10), a co-occurence that is suggestive of a creole origin, but
more research is necessary to address the diagnosticity question more directly.

. Given the lack of studies on present-tense ain’t, it is in need of more research to establish
its status and use in PDAAE. My sense from collecting data for this study is that collecting
present-day ain’t from the appropriate lect would yield many more examples and reveal its
spread in PDAAE to be quite robust.
 K. Aaron Smith

(10) Ain nobody ga worry wid you

‘There’s nobody that will worry with you’  (Mufwene 2004: 368)

If, however, the apparent increase of existential ain’t (1/300 –> 18/300) truly
represents an increase, i.e. a greater realization of potential existential sites of
meaning with ain’t, then it is another area in which PDAAE is diverging from
historical AAE.

6. Analysis

6.1 Present-tense ain’t

While the numbers on present tense ain’t are not robust here, and as I indicated e­ arlier
have not shown up with any great regularity in other studies of ain’t employing other
methods of collection (e.g. sociolinguistic interview/participant-observer ­recording),
the instances that do occur support an exemplar-based configuration for the emer-
gence and spread of that construction. Bybee (2006: 716) notes that “[c]onstructions
emerge when phrases that bear some formal similarity as well as some semantic
coherence are stored close to one another.” This kind of cluster is centered around
a (usually) high frequency token that triggers prototype effects (Bybee 2006: 717),
attracting other tokens, which are stored near that high frequency prototype.
Bybee and Eddington (2006) studied quedarse + adjective constructions,
finding that quedarse inmóvil immobile was a frequent token and that it attracted
other uses of quedarse + adjective with meanings similar to inmóvil. Bybee and
Eddington offer several types of evidence to suggest that quedarse inmóvil serves as
the organizing exemplar for the emerging category. They note that the frequency
of the inmóvil adjective type is highest among the general quedarse + adjective
combination, and that instances with lesser token frequency cluster around it,
e.g. quedarse parado stopped, atrapado seized, de piedra like stone, etc. The most
frequent adjective inmóvil then acts as a base and the other adjectives attracted
to it form a family resemblance set wherein the ­adjectives share certain semantic
features, in this case immobility. Bybee and Eddington were also able to carry
out an acceptability test in which they found that speakers of Spanish rated high
frequency adjectives occurring with quedarse as most acceptable, again suggest-
ing that such adjectives were the most accessible in that construction and serve as
central members of the emerging construction.
In the data collected for this study, present tense ain’t shows a similar catego-
rial configuration. Table 2 gives the lexical verbs in the present tense ain’t con-
struction and the number of occurrences for each of those verbs.
 A usage-based account for the historical reflexes of ain’t in AAE 

Table 2. Lexical verbs with present tense ain’t in PDAA

Lexical verb #of occurrences

know 11
care 1
give a shit 1
give a damn 1
like 3
want 3
have 2
have to 1
feel 2
look like 1
wear 1
go to school 1
say 1

Based on frequency alone the lexical verb know is a likely candidate for the
central exemplar of the set of present tense ain’t tokens. That centrality is sup-
ported by the fact that the nearly all of the other less frequent examples share
a stative semantic profile with know, except for go to school and say, which are
of a more dynamic nature. Thus, the more specific present-tense ain’t category
may have emerged based on the central and organizing token of ain’t know, and
from there attracted other stative verbs. Diachronic evidence for the central-
ity of ain’t know is also available in the data collected for this study in so far as
among the 6 instances of present tense ain’t in the 19-AAE data, 5 occur with
the verb know.
Earlier I suggested that since present tense ain’t does not appear in the
19th-century non-AAE (and not in any non-AAE varieity of PDE that I am
aware of), a creole origin may be likely. A comparison of the use of present-
tense ain’t in l­ iterary works with characters of creole or partial-creole speaking
identity is instructive for the analysis provided here on AAE present-tense ain’t.
In the collection of short stories by Colin Channer in his novel Passing Through,
written in the voice of various Caribbean island speakers often with some sus-
tained contact with AAE speakers in the US, present tense ain’t appears very
frequently. Table 3 shows the distribution of present-tense ain’t for the first 100
uses in the novel.
 K. Aaron Smith

Table 3. Lexical verbs with present tense ain’t in Passing Through8

Lexical verb # of occurrences

know 34
want 15
have 16
have to 58
like 4
care 4
go 2
think 2
see 2
trust 2
read 2
mean 2
mind 1
try 1
take 1
trouble 1
act 1
complain 1
do 1
disbelieve 1
feel 1
cuss 1

As the PDAAE data in Table 2 showed, know is the most frequent token with
like, want and have following. The data in Table 3 show the same overall pattern
of frequency and supports the hypothesis that know is the central exemplar
around which the other examples cluster for present-tense uses of ain’t. The fact
that the same general pattern holds for PDAAE and partially c­ reolized ­varieties
of English is again suggestive of a creole origin for present-tense ain’t.
In the next section, I raise some questions as to the specificity of present tense
ain’t in the overall tense and aspect system of AAE since tokens of present tense

. Here I have counted have and have to as separate verbs given the quasi-modal status of
the latter.
 A usage-based account for the historical reflexes of ain’t in AAE 

ain’t share an identical from with other possible tense/aspect meanings. The result,
as I argue, is the emergence of a larger, more abstract construction in which auxil-
iary ain’t is underdetermined for tense/aspect distinctions.

6.2 Underdetermined ain’t

The wide array of uses of ain’t in PDAAE has led researchers such as DeBose
(1994) to analyze ain’t as a “tense/aspect neutral negator.” However, the analysis
of ain’t as “neutral” is not supportable given that, for example, ain’t never occurs
as a past auxiliary, varying with weren’t/wasn’t; in other words, it does show some
constraints on its availability for some tense/aspect contexts (Howe and Walker
2000: 118). Furthermore from the data collected for this study, existential ain’t,
either in the bare form or with an expletive subject there, never occurs in the past.
In Smith (2015), I analyzed the form as “underdetermined,” by which I meant
that certain of the constructions it is used in are not formally distinct and there-
fore cannot distinguish certain tense/aspect meanings by form alone. For instance,
perfect, past and present uses of auxiliary ain’t can all share the same syntactic
form, ain’t + base form of the verb, as shown in (13 a-c), and therefore tense/
aspect meaning is only retrievable from other co-textual or contextual cues. In
other words, if more specific tense/aspect meaning is resolved, it must be done on
the pragmatic level.
(13)
a. “You ain’t see how ugly Arlene get.” 
 (From Newark’s Armadillo,9 Arlene J. Ramsey, p. 49)
b. “I fell right where I was and landed on some girl who ain’t say nothing.”
 (Imani all mine. ebook)
c. “Why don’t you stop hating on Mommy and let her live her life? She
ain’t want your father; hell is she with mine? Just get the hell over it.
Period.”  (The Ex-factor, ebook)

In the three examples, 13a is an instance of an anterior meaning, reporting on

something not having been seen before the present reference of how ugly Arlene
gets. Anterior meaning is regularly expressed by the Perfect have + past participle
construction, and instances with a past participle occur frequently in these and
other data: ain’t seen, ain’t seed, etc. Thus, the construction with ain’t + the bare
infinitive is only one variant for expression of that meaning. 13b is a past refer-
ence, ain’t say being the third verb in the series of past tense verbs, fell, landed, ain’t
say, and 13c is a present tense instance as evidenced by its occurrence with other

. Newark’s Armadillo is not one of the novels used for data in this study. Parts of the novel,
however, are meant to represent AAE as is this instance from the novel.
 K. Aaron Smith

­ resent tense verbs, don’t…stop, let, ain’t want, is, get. Again, we note that while
p
the use in 13a is, at least semantically, related to the historical sources of ain’t from
have not, 13b and 13c are not related to the historical sources.
The formal identity of the three instances in (13a–c) suggests too that the
constructions are stored near one another from which a more general construc-
tion emerges with the abstract form ain’t + bare infinitive. In fact, it may be the
case that present and past tense uses of ain’t in 13b and 13c attracted perfect uses,
thus motivating the use of a bare infinitive instead of past participle forms, as in
examples like that in 13a. The effect of that convergence of forms results in an
underdetermined status for auxiliary ain’t in such uses.
In this way, ain’t is not unlike underdetermined structures in other languages.
Gil (2009) for instance reports on a rather extreme case of underdeterminism in
Riau Indonesian. In that language, the two words ayam makan ‘chicken eat’ is
underdetermined for grammatical relationships so that the simple syntactic string
can mean ‘I ate/eat/have eaten the chicken’, ‘the chicken was eaten’, ‘the chicken
that we ate’, etc. Underdetermined structures like that offered in Riau Indonesian
by Gil have been central in discussions on language complexity, and in the next
section I turn to the complexity issue in the tense/aspect system of AAE.

6.3  Undetermination and linguistic complexity

On the surface of things, it would appear then that underdetermined ain’t as pre-
sented in the last section is an instance of simplification in the grammar of AAE. Lan-
guage simplification/simplicity has a vexed history in linguistics, and the usual default
assumption has been towards an equi-complexity hypothesis (Trudgill 2011: 15–16)
in which languages are thought to work according to a principle by which attenuation
in one part of the system is offset by growth in another, thus maintaining the overall
complexity of the linguistic system. While the assumption of an equicomplexity prin-
ciple has in recent years come under scrutiny (see the papers in Sampson, Gil & Trud-
gill 2009, for example), it is relatively uncontroversial that certain subsystems in a
language can undergo simplication. I should be explicit here in saying that my point is
not that PDAAE (or historical AAE for that matter) is a “simplified” language or vari-
ety of a language. The recent AAE development of a number of futurate forms with
highly nuanced meanings for futurity, obligation and intention, e.g. ima, finna (Smith
2009) and tryna attest well to the evolving complexity of the AAE verbal system.10

. An instance of the futurate trying (often written as tryna) can be seen in (i).
(i) “I’m tellin’ you, cuz’, leave that nut bitch alone.”
“You crazy. I ain’t hardly tryin’ to let her go,” Bruce snapped.
Tryin’ here talks about future intention and not attempt, i.e. the lexical meaning of try.
 A usage-based account for the historical reflexes of ain’t in AAE 

It should also be clear that underdetermination does not mean undetermin-

able. To return to the underdetermined structure in Riau Indonesian mentioned
above, Gil (2009: 24) argues that while speakers may often not desire to or need to
determine grammatical relations in a syntactic string like ayam makan, they may
well do so on occasion. I suggest the same is true generally for underdetermined
ain’t, but it would seem likely that tense/aspect resolution of formally identical
instances like those in (13a–c) are often resolved since speakers of AAE have forms
that vary with underdetmined ain’t and which are formally determined for pres-
ent, past and perfect (resultative and anterior) meanings. Therefore it could be
argued that the distinctions are more likely to be resolved in the face of under-
determined constructions like those in (13a–c) if speakers are already primed to
“listen for understanding” (Slobin 2003).
As discussed briefly in the introduction, the very general profile of ain’t
as an auxiliary in PDAAE (but also other varieties of English) makes one ask
whether the stage has been set for renewal (Hopper 1991). That is, are new aux-
iliary constructions developing to revivify the distinctions lost by the conver-
gence of present, past and perfect (anterior and resultative) meanings in the
same ain’t + bare infinitive form? To my knowledge there are not, and in the final
part of this paper, I suggest a functional explanation for the loss of tense/aspect
distinction in AAE through an extension of the network/exemplar model into
the sociolinguistic realm.

7. The divergence hypothesis and usage-based grammar: A conclusion

A salubrious feature of the exemplar/network theory is that it does not rely on

the langue-parole or competence-performance distinction of many 20th- and
21st-century linguistic theories; in fact it challenges any such dichotomy (Bybee
2006). The current paper extends the performance inclusivity of the exemplar/
network theory to explain the apparent divergence of AAE from non-AAE vari-
eties and earlier AAE (See too Bailey & Maynor 1989; Butters 1989; Labov 1985
and Rickford 1992 on PDAAE divergence). We note for instance that the spread
and increased frequency of ain’t in PDAAE has led to a loss of specificity, in which
case we can talk about the generalization of meaning for ain’t, observable in other
situations in which a form extends its range of uses (Givon 1975; Bybee, Perkins
& Pagliuca 1994: 6; Gabelentz 1891: 241). From a purely grammatical viewpoint,
then, the morphosyntactic generality of ain’t in PDAAE represents a loss in the
grammar, and thus arguably a simplification in the tense/aspect system of that
variety of English. In order to motivate this apparent loss of tense/aspect specific-
ity, I argue that a usage-based framework (Bybee 2006, 2010), which is compati-
ble with sociolinguistic theory (e.g. Sampson, Gil & Trudgill 2009; Trudgill 2011),
 K. Aaron Smith

can also motivate this kind of change. That is, through usage, specific linguistic
forms/constructions, in this case certain uses of auxiliary ain’t (perhaps even with
specific verbs), have become associated with and even indexical of certain non-
linguistic features, such as speech community identity, thus increasing the social
indexicality of auxiliary ain’t in AAE despite the concomitant loss of distinct aux-
iliaries for tense/aspect.
The data presented here, and in other published studies cited throughout this
paper, have shown that PDAAE has some uses of ain’t that are either absent from
non-AAE (present tense aint’ and bare-ain’t existentials) or that are more limited
in non-AAE (past tense ain’t). Furthermore given that all of these uses are avail-
able in historical AAE, but that their frequency increases in PDAAE, the data sup-
port the position that PDAAE has diverged further from non-AAE varieties of
English and even historical AAE.
As discussed earlier, in a usage-based model of grammar, a speaker’s expe-
rience with language impacts stored forms of the language. Bybee (2006: 717,
and see Johnson 1997) indicates that some of that experience may be, strictly
speaking, non-linguistic, having to do instead with sociolinguistic factors.
If constructions are associated with sociolinguistic experience at the site of
storage, then the increased frequency of AAE ain’t, the clustering together of
underdetermined uses and the apparent loss of tense/aspect distinctions are all
motivated in so far as the form itself has become a marker of community and
personal identity – something that makes good sense given the important role
of racial politics throughout history, and the foreseeable future, in the use of
American English.

Primary resourses

Nineteenth Century African American English (19-AAE)

William Francis Allen, b. 1830. Slave Songs of the United States, pub. 1867. New York NY:
A. Simpson & Co.
William Wells Brown, b. 1814? Narrative of William W. Brown, an American Slave written by
Himself, pub. 1849. London: C. Gilpin.
Ellen Anderson Gholson Glasgow, b. 1873. The Battle-Ground, pub. 1902. New York NY:
­Doubleday, Page & Co.
Will N. Harben, b. 1858. Northern Georgia Sketches, pub. 1900. Chicago IL: A.C. McClurg & Co.
Nineteenth Non-African American English (19-Non-AAE)
Sherwood Bonner, b. 1849. Dialect Tales, pub. 1883. New York NY: Harper & Brothers.
Joel Chandler Harris, b. 1848. Uncle Remus, His Songs and His Sayings: The Folk-Lore of the Old
Plantation, pub, 1881. New York NY: D. Appleton and Company.
Ellen Anderson Gholson Glasgow, b. 1873. “The Battle-Ground,” pub. 1902. New York NY:
­Doubleday, Page & Co.
 A usage-based account for the historical reflexes of ain’t in AAE 

John Patterson Green, b. 1845. Recollections of the Inhabitants, Localities, Superstitions, and
­KuKlux Outrages of the Carolinas. By a “Carpet-Bagger” Who Was Born and Lived There,
pub. 1880. Publisher unknown.
Ruth McEnery Stuart, b. 1856. In Simpkinsville: Character Tales, pub. 1897. New York NY:
Harper & Brothers.
Present-day African American English (PDAAE)
Mink, Meesha and De’nesha Diamond. 2008. Shameless Hoodwives: A Bentley Manor Tale.
New York NY: Simon and Schuster
Porter, Connie Rose. 2014. Imani All Mine. Orlando FL: Houghton Mifflin Harcourt.
Tyree, Omar. 2001. Flyy Girl: An Urban Classic Novel. New York NY: Simon and Schuster.
Whitaker, Tu-Shonda. 2008. The Ex-Factor. London: One World.

References

Anderwald, Lieselotte. 2002. Negation in Non-standard British English: Gaps, Regularization and
Asymmetries. New York NY: Routledge. doi: 10.4324/9780203167502
Bailey, Guy & Maynor, Natalie. 1989. The divergence controversy. American Speech 64(1):
12–39. doi: 10.2307/455110
Baugh, John. 1988. Langauge and race. Some implications for linguistic science. In Linguistics:
The Cambridge Survey, IV: Language: The Socio-cultural Context, Frederick J. Newmeyer
(ed.), 64–74. Cambridge: CUP.
Butters, Ron. 1989. The Death of Black English: Divergence and Convergence in Black and White
Vernaculars. Frankfurt: Peter Lang.
Bybee, Joan L. 2010. Language, Usage and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan L. & Eddington, David. 2006. A usage-based approach to Spanish verbs of ‘becom-
ing’. Language 82: 323–355. doi: 10.1353/lan.2006.0081
Bybee, Joan L., Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar: Tense,
Aspect and Modality in the Languages of the World. Chicago IL: University of Chicago Press.
Bybee, Joan L. 2006. From usage to grammar: The mind’s response to repetition. Language
82(4): 711–733. doi: 10.1353/lan.2006.0186
Cheshire, Jenny. 1981. Variation in the use of ain’t in an urban British English dialect. Language
in Society 10(3): 365–381. doi: 10.1017/S0047404500008848
DeBose, Charles E. 1994. A note on ain’t versus didn’t negation in African American Vernacular.
Journal of Pidgin and Creole Languages 9(1): 127–130. doi: 10.1075/jpcl.9.1.16cha
DeBose, Charles E. & Faraclas, Nick. 1993. An Africanist approach to the linguistic study of
Black English: Getting ot the roots of the tense-aspect-modality and copula systems in
Afro-American. In Africanisms in Afro-American Language Varieties, Salikoko S. Mufwene
(ed.), 364–387. Athens GA: University of Georgia Press.
Donauer, Trish & Katz, Seth. 2015. Ain’thology: The History and Life of a Taboo Word. Newcastle
upon Tyne: Cambridge Scholars.
Fasold, Ralph & Wolfram, Walt. 1975. Some linguistic features of the Negro dialect. In Black
American English. Its Background and its Usage in the Schools and in Literature, Paul Stoller
(ed.), 49–89. New York NY: Dell.
 K. Aaron Smith

Feagin, Crawford. 1979. Variation and Change in Alabama English: A Sociolinguistic Study of the
White Community. Washington DC: Georgetown University Press.
von der Gabelentz, George. 1891. Die Sprachwissenschaft. Ihre Auggaben, Methoden, und bisheri-
gen Ergebnisse. Leipzig: Weigel.
Gil, David. 2009. How much grammar does it take to sail a boat? In Language Complexity as an
Evolving Variable, Geoffrey Sampson, David Gil & Peter Trudgill (eds), 19–49. Oxford: OUP.
Givȯn, Talmy. 1975. Serial verbs and syntactic change: Niger-Congo. In Word Order and Word
Order Change, Charles Li (ed.), 890–925. Austin TX: University of Texas Press.
Green, Lisa. 2002. African American English: A Linguistic Introduction. London: Routledge.
doi: 10.1017/CBO9780511800306
Hopper, Paul. 1991. On some properties of grammaticization. In Approaches to Grammatical-
ization, Vol. 1 [Typological Studies in Language 19:1], Elizabeth Traugott & Bernd Heine
(eds.), 17–36. Amsterdam: John Benjamins. doi: 10.1075/tsl.19.1.04hop
Hopper, Paul. 1987. Emergent grammar. Berkeley Linguistics Society 13: 139–157.
doi: 10.3765/bls.v13i0.1834
Hopper, Paul & Traugott, Elizabeth Closs. 2003. Grammaticalization. Cambridge: CUP.
doi: 10.1017/CBO9781139165525
Howe, Darin M. 1997. Negation in the history of African American English. Language Variation
and Change 9(2): 267–294. doi: 10.1017/S0954394500001903
Howe, Darin, M. & Walker, James A. 2000. Negation and the creole-origins hypothesis: Evi-
dence from early African American English. In The English history of African American
English, Shana Poplack (ed.), 109–140. Malden MA: Blackwell.
Jespersen, Otto. 1940. A Modern English Grammar on Historical Principles, Part V Copenhagen:
Eijnar Munksgaard.
Johnson, Keith. 1997. Speech perception without speaker normalization. In Talker Variability
in Speech Processing, Keith Johnson & John W. Mullennix (eds), 145–165. San Diego CA:
Academic Press.
Kuryłowicz, Jerzy. 1975[1965]. The evolution of grammatical categories. Reprinted in Jerzy
Kuryłowicz, 1976, Esquisses linguistiques, Vol. 2, 38–54. Munich: Fink.
Labov, William. 1985. The increasing divergence of Black and White vernacular English. In Afri-
can-American English: Structure, History and Use, Salikoko Mufwene, John Rickford, Guy
Bailey & John Baugh (eds), 110–153. London: Routledge.
Labov, William. 1972. Language in the Inner City: Studies in the Black English Vernacular. Phila-
delphia PA: University of Pennsylvania Press.
McDavid, Raven I. 1941. Ain’t I and aren’t I. Language 17: 57–59. doi: 10.2307/409463
Mufwene, Salikoko S. 2004. Gullah: Morphology and syntax. In A Handbook of Varieties of
­English: Morphology and Syntax, Bernd Kortmann, Kate Burridge, Rajend Mesthrie, Edgar
W. Schneider & Clive Upton (eds), 356–373. Berlin: Mouton de Gruyter.
Palacios Martinez, Ignacio M. 2010. “It ain’t nothing to do with my school.” Variation and
pragmatic uses of ain’t in the language of British English teenagers. English Studies 91(5):
548–566. doi: 10.1080/0013838X.2010.488841
Payne, Thomas E. 2010. Understanding English Grammar. Oxford: OUP.
doi: 10.1017/CBO9780511778988
Poplack, Shana. 2000. The English History of African American English. Malden MA: Blackwell.
Radford, Andrew. 1997. Syntax: A Minimalist Introduction. Cambridge: CUP.
doi: 10.1017/CBO9781139166898
 A usage-based account for the historical reflexes of ain’t in AAE 

Rickford, John R. 1977. The Question of Prior Creolization of Black English. In Pidgin and Creole
Linguistics, Albert Valdman (ed.). Bloomington IN: Indiana University Press.
Rickford, John R. 1992. Grammatical variation and divergence in Vernacular Black English. In
Internal and External Factors in Syntactic Change, Marinel Gerritsen & Dieter Stein (eds),
175–200. Berlin: Mouton de Gruyter. doi: 10.1515/9783110886047.175
Sampson, Geoffrey, Gil, David & Trudgill, Peter (eds). 2009. Language Complexity as an Evolving
Variable. Oxford: OUP.
Sebba, Mark. 2004. British Creole: Morphology and syntax. In A handbook of varieties of ­English:
Morphology and syntax, Bernd Kortmann, Kate Burridge, Rajend Mesthrie, Edgar W.
Schneider & Clive Upton (eds.), 196–208. Berlin/New York: Mouton de Gruyter.
Singler, John. 1998. What not new in AAVE. American Speech 73(3): 227–256. doi: 10.2307/455824
Slobin, Dan I. 2003. Language and thought online: Cognitive consequences of linguistic rela-
tivity. In Language in mind: Advances in the study of language and thought, D. Gentner &
S. Goldin-Meadow (eds.). Cambridge MA: MIT Press.
Smith, K. Aaron. 2006. The universal tendency for renewal among grammatical expression for
anterior and related aspect. Journal of Universal Language 7: 139–160.
Smith, K. Aaron. 2009. The history of be fixing to: Grammaticization, sociolinguisitc distribu-
tion and emerging literary spaces. English Today 97:25(1): 12–18.
doi: 10.1017/S0266078409000030
Smith, K. Aaron. 2015. Historical development and aspectual nuances of ain’t periphrases. In
Ain’thology: The History and Life of a Taboo Word, Trish Donauer & Seth Katz (eds), 72–94.
Newcastle upon Tyne: Cambridge Scholars.
Stevens, Martin. 1954. The derivation of ain’t. American Speech 29(5): 196–201. doi: 10.2307/454240
Traugott, Elizabeth Closs. 2002. From etymology to historical pragmatics. In Studies in the His-
tory of the English Language, Donka Minkova & Robert Stockwell (eds), 19–49. Berlin:
Mouton de Gruyter. doi: 10.1515/9783110197143.1.19
Trudgill, Peter. 2011. Sociolinguistic Typology: Social Determinants of Linguistic Complexity.
Oxford: OUP.
Walker, James. 2005. The ain’t constraint: Not-contraction in early African American English.
Language Change and Variation 17: 1–17. doi: 10.1017/S0954394505050015
Weldon, Tracey. 1994. Variability in negation in African American Vernacular English. Lan-
guage Variation and Change 6(3): 359–397. doi: 10.1017/S0954394500001721
Weldon, Tracey L. 2007. Gullah negation: A variable analysis. American Speech 82(4): 341–366.
doi: 10.1215/00031283-2007-023
Winford, Donald. 1998. On the origins of African American Vernacular English – A creolist
perspective. Diachronica 15(1): 99–54. doi: 10.1075/dia.15.1.05win
Wolfram, Walt & Schilling-Estes, Natalie. 2003. Language change in “conservative” dialects: The
case of past tense BE in southern enclave communities. American Speech 78(2): 208–227.
doi: 10.1215/00031283-78-2-208
Wolfram, Walt & Schilling-Estes, Natalie. 2006. American English. Malden MA: Blackwell.
 Gradient conventionalization of the Spanish
­expression of ‘becoming’ quedar(se) + ADJ
in seven centuries

Damián Vergara Wilson

University of New Mexico

The present chapter focuses on the evolution of semantic categories in the Spanish
expression of ‘becoming’ quedar(se) + ADJ and examines how this analogical
extension is affected by the conventionalization of prominent category members.
To this end, this study operationalizes conventionalization through examining
factors such as previous usage, widespread usage, and frequency in developing a
gradient index of conventionalization of specific verb + adjective combinations in
a diachronic corpus. Through developing an index that can be used to compare
the conventionalization of a specific combination from one century to the
next, we see that a robust centralized member will contribute to the ongoing
maintenance and productivity of a semantic category even in the face of overall
frequency reduction of the construction.

Introduction

How do varying levels of conventionalization interact with the categorization of

adjective types in verb + adjective change of state constructions in Spanish? Rec-
ognizing the importance of speaker experience of language in context, Bybee
(2006: 711) defines grammar as “the cognitive organization of one’s experience
with language” and highlights the importance of frequency in the conventionaliza-
tion of specific constructions and in categorization. She argues that speakers are
sensitive to frequency of usage, and it follows that speakers are also sensitive to the
conventionalization of certain constructions. In usage, the degree of conventional-
ization is gradient (Bybee 2006) and interacts with frequency effects such as chunk-
ing and constructional productivity. The present chapter focuses on the evolution
of semantic categories through analogical semantic extension and examines how
this extension is affected by conventionalization of prominent category members.
To this end, this study operationalizes constructional conventionalization through
examining factors such as previous usage, widespread usage, and frequency in

doi 10.1075/slcs.192.08wil
© 2018 John Benjamins Publishing Company
 Damián Vergara Wilson

developing a gradient index of conventionalization of specific verb + adjective

combinations of the change-of-state construction quedar(se) + ADJ in a dia-
chronic corpus. Through developing an index that can be used to compare the
conventionalization of a specific combination from one century to the next, we
gain insight into the interaction of analogical extension, category formation and
the organizing effect of conventionalized types.
Although the main focus of the present investigation is on the expression of
‘becoming’ quedar(se) + ADJ (1), it is important to highlight that the mechanism
of linguistic evolution analyzed here, analogical extension, entails a consideration
of competing synonymic constructions with similar structure: fincar(se) + ADJ (2)
and ponerse + ADJ1 (3).
Cuando ha de dar un consejo se siente en grave apuro y, si no sale bien, queda
(1) 
el consejero humillado y desacreditado.
‘When one must give advice one feels themselves in a serious bind and, if it
does not come out well, the advisor becomes humiliated and disaccredited.’
 (D. Juan Manuel, El Conde Lucanor: 14th c., O’Neill 1999)
Vendian a sus vezindades: en guisa que fincaron muy ricos de alli adelante.
(2) 
‘They sold their surrounding land: this is how they became rich from
then on.’  (Anónimo, Crónica del Cid: 15th c., CDE)
Y el Mochuelo se puso más triste todavía, pensando que cuatro semanas
(3) 
después él se iría a la ciudad…
 ‘And Mochuelo got even sadder still thinking that four weeks later he’d go
to the city …’  (Bybee & Eddington 2006: 342, Ex. 26 )

Despite etymological differences and general polysemy of these verbs, when used in
these constructions they are pseudo-copulas (Delbecque & Van Gorp 2012, 2015;
Van Gorp 2012, 2015) that mean ‘to become’ while lexical meaning is guided by
the adjective. In the data there is overlap as adjectives may co-occur among differ-
ent constructions of ‘becoming’. More than just variation in usage, co-­occurrence
is indicative of diachronic adjective movement as types are attracted away from
one construction and into another through analogical extension. By examining
two categories of adjectives appearing in quedar(se) + ADJ – one category that
centers on solo ‘alone’ and another on types that denote gradient degrees of happi-
ness – this study provides evidence that conventionalized combinations contribute

. In the notation for this construction the middle-voice pronoun ‘se’ is maintained in
­parentheses to indicate its variable usage in fincar(se) and quedar(se) + ADJ. An analysis of its
variability is beyond the scope of this paper. However, parentheses are absent with the change
of state construction ponerse + ADJ because ‘se’ is used categorically.
 Gradient conventionalization of the Spanish expression 

to retention of types in the categories to which they belong compared to categories

that lack highly conventionalized types.
The focus of this investigation is on the change-of-state construction
quedar(se) + ADJ used with an animate subject in a diachronic corpus of writ-
ten Spanish from the 1200s through the 1800s. The data primarily comes from
the Quedar(se) + Adjective Database (henceforth QAD) and secondarily from
Davies’ Corpus del español (2002: henceforth CDE). This study is an extension of a
detailed diachronic analysis of quedar(se) + ADJ described in Wilson (2014), itself
an extension of Bybee and Eddington (2006). The data cover seven centuries but
the analysis focuses on the time period where quedar(se) + ADJ begins to wane
from peak productivity in the 1600s to diminished productivity in the 1800s
Quedar(se) + ADJ is an ideal construction for the examination of frequency
effects on analogy-based categorization in a diachronic setting (Wilson 2014).
First, quedar(se) + ADJ presents challenges to the researcher because the fixed
element, the verb quedar(se) can convey opposite meanings of ‘stay/remain’ or
‘become’ depending on the context of usage. While the ‘stay/remain’ sense is
the more canonical etymological usage, the ‘become’ sense, is more innovative
and arose partially through inferencing (e.g., Traugott & Trousdale 2013: 202) of
ambiguous verb + adjective combinations in early periods. Also, its predecessor
fincar(se) + ADJ conveyed this ‘remaining/becoming’ duality, which appears to
be an inherited semantic characteristic. Second, the data clearly demonstrate that
analogical extension is a mechanism of productivity; quedar(se) + ADJ becomes
increasingly schematic over time accepting the widest variety of adjective types
in the 1600s. Productivity, Bybee (2010: 94–96) proposes, is the likelihood that a
pattern will extend to new forms and type frequency is the chief determinant of
the level of productivity of a construction. The evolution of quedar(se) + ADJ is
similar to the expansion of the way-construction (Israel 1996) in which schematic
categories expand through analogical extension to previously established usages
thereby causing massive constructional change over time. Third, adjective usage
in this construction has shown a great deal of covariation with synonymic expres-
sions of ‘becoming’.
The current study represents a development in previous studies of frequency
and categorization in expressions of ‘becoming’ (Bybee & Eddington 2006; Wilson
2009; Wilson 2014) by deriving a gradient Conventionalization Index that can be
applied to particular combinations of quedar(se) + ADJ. This index is a usage-based
measurement that takes into account the following factors: usage of a combination
in previous centuries, the number of authors who use the particular combination,
token frequency, and the overall normalized frequency of the construction in a
given century. We see that categories of adjectives with highly conventionalized
central members may retain a high degree of productivity despite the recession
 Damián Vergara Wilson

of the overall construction quedar(se) + ADJ. Also, because certain combinations

maintain high levels of conventionalization, this demonstrates that highly con-
ventionalized instances of quedar(se) + ADJ, or prefabs, have longevity. However,
the lack of a robust central member leaves a semantic category of adjectives more
vulnerable to losing members to competing constructions, such as ponerse + ADJ.
Highly conventionalized types serve as central members of enduring categories,
provide category cohesion and, depending upon the level of conventionalization,
may stem the erosion of a category.

Previous examinations of verbs of ‘becoming’

Spanish has a rich inventory of change-of-state expressions and there is a great

deal of potential variation. However, verb+adjective expressions of ‘becoming’
in Spanish have garnered attention from scholars and grammarians who failed
to provide an explanatorily adequate account of their distribution in usage.
Eddington (2004: 54–55) draws attention to the shortcomings of several stud-
ies (Coste & Redondo 1965; Crespo 1949; Eberenz 1985; Fente 1970) observing
that “criterion for distinguishing the contexts in which one or more change-of-
state verbs is defined, followed by a handful of sentences that exemplify that
the criterion works” (54–55). The richness of variation and lack of usage-based
analysis of these expressions became an invitation for innovative functionalist
work in expressions of ‘becoming’ (Eddington 1999, 2002; Bybee & Eddington
2006; Wilson 2009, 2014, see also Delbecque & Van Gorp 2012, 2015; Van Gorp
2012, 2015).
There are four overarching manners of expressing the ways in which a speaker
can experience a change from one state to another: verb + adjective expressions of
‘becoming’, which are also referred to as pseudo-copular constructions in work by
Delbeque and Van Gorp (ibid) as in (4); lexical verbs mainly derived from adjec-
tives and used in the middle voice, as in (5); periphrastic constructions (6); and
verbs that mean ‘to become/to change’ (7).

Se fue la mujer y el hombre se quedó solo. ‘The woman left and the man was
(4) 
left behind.’
El mercader se volvió rico. ‘The merchant became rich’
El cocinero se hizo gordo. ‘The cook became fat’
El actor se puso triste. ‘The actor got sad’

(5) El mercader se enriqueció. ‘The merchant got rich’

El cocinero se engordó. ‘The cook got fat’
El actor se entristeció. ‘The actor got sad’
 Gradient conventionalization of the Spanish expression 

(6) Llegaron a ser dueños de un rancho. ‘They came to be owners of a ranch’

Se convirtió en estatua de sal. ‘S/he converted into a salt statue’
(7) 
 Las muchachas se transformaron en mujeres ‘The girls transformed into
women’

Therefore, speakers have a great deal of choice in describing changes from one state
to another (for more detail see Wilson 2014: Chapter 2). Here we focus on the four
verbs that are productively used in Modern Spanish as pseudo-copulas with adjec-
tival complements shown in example (4): quedar(se), hacerse, ponerse, and volverse.
Additionally, fincar was a predecessor to these four verbs that fell into complete
disuse by the 1500s.2 As previously stated, there is great deal of overlap of specific
adjectives that appear with multiple pseudo-copulas. For example, the adjective
rico ‘rich’ is attested in Davies’ Corpus del español (2002-) with all four modernly
used verbs of ‘becoming’ (and with fincar) to denote that the subject got rich.
Bybee and Eddington (2006) conducted a multi-faceted examination of fre-
quency effects and categorization of adjectives in expressions of ‘becoming’ with
the verbs quedar(se), hacerse, volverse, and ponerse used with animate subjects. One
of the distributional patterns that they highlight was the prevalent overlap of adjec-
tives across these verbs in their data; they hypothesized that accumulated categories
of usage guided the formation of verb + adjective pairs instead of rules determined
by the verb’s etymology. The overarching contributions of their study support the
exemplar model of cognitive representation in which language is organized through
experience into gradient, radial categories that guide usage and novel extension.
Wilson (2014) followed the work on categorization in Bybee and Eddington,
but focused on evolutionary mechanisms of semantic categories adjectives occur-
ring with quedar(se) from the 1200s to the 1900s. In examining frequency effects
one particular measure was operationalized: Conventionalized Instances of Con-
structions (CICs). The CIC measure was derived in order to identify combinations
of quedar(se) + ADJ that showed some degree of conventionalization in a way that
could be readily applied to all the data examined (Wilson 2014: 70–76); this mea-
sure is discussed below.

Quedarse + ADJ in eight centuries

This section provides summary analysis figures of the construction quedar(se) +

ADJ and an overview of the corpus compiled for this research, the QAD. In

. Interestingly, ficar is the only verb used in modern Portuguese verb + adjective expressions
of ‘becoming’.
 Damián Vergara Wilson

c­ ompiling this corpus of written works, the author attempted to extract approxi-
mately 250 tokens of quedar(se) + ADJ per century from Spanish Peninsular nar-
rative works while also extracting all instances of the construction in the works
examined. Table 1, reproduced here from Wilson (2014: 68), indicates the count
of tokens extracted from the corpus, the count of adjective types, the overall nor-
malized frequency (ONF) of quedar(se) + ADJ computed as the number of occur-
rences per 10,000 words, the ratio of tokens per type, and the word count of the
QAD corpus by century. The figures for the 1900s were calculated using Bybee and
Eddington’s (2006)3 data.

Table 1. Overview of quedar(se) + ADJ types, tokens, and frequency through time in the
QAD (adapted from Wilson 2014: Table 4.2, p. 68)
ONF: # per Token/Type # words in
Tokens Types 10,000 ratio corpus

1200s 43 29 0.42 1.48 1,031,755

1300s 76 40 0.96 1.9   789,426
1400s 196 121 1.18 1.62 1,658,165
1500s 271 153 4.85 1.77   558,475
1600s 280 171 6.00 1.63   463,529
1700s 253 135 5.82 1.87   435,055
1800s 255 141 2.64 1.80   967,252
1900s* 181 54 1.83 3.35   990,000
* 1900s From Bybee & Eddington (2006)

The token per type ratio provides key information in the analysis of some fre-
quency effects found in quedar(se) + ADJ. Although work by Renouf and Sinclair
(1991) and Bybee (2010: 131–132) made use of the inverse type per token ratio,
the token per type ratio was used by Butler (1998) to analyze specificity of collo-
cations in Spanish. In Wilson (2014) the token per type ratio provides a gauge of
the internal variability of categories of adjectives in quedar(se) + ADJ because it
reveals a trend for tokens to be concentrated into singletons or into adjective types
of high token frequency: being essentially an average, a lower token per type ratio
indicates fewer tokens per adjective type (i.e. more singletons and less specificity),

. Bybee and Eddington (2006) consulted a corpus of 1.1 million words representing spoken
Peninsular Spanish and a corpus of 15 novels totaling 990,000 words. For the sake of compat-
ibility with the QAD corpus, the calculations from Bybee and ­Eddington (2006) are from the
written portion of their corpus.
 Gradient conventionalization of the Spanish expression 

while a higher ratio reflects a higher degree of token frequency per type (i.e. more
specificity). By way of example, the 1200s show the lowest token per type ratio
(1.48) because of a predominance of singletons and a lack of types with more than
one token: 19 out of a total of 29 types (65.5%) were singletons. Yet, in the follow-
ing century only 32% of the data were singletons.
One of the most striking figures in Table 1 is the normalized frequency col-
umn (ONF) as it shows the rise and fall of the construction quedar(se) + ADJ
in usage. We see that in the earliest time period, the 1200s, the construction
occurred relatively infrequently in the texts examined at a rate of 0.42 occur-
rences per 10k words; in order to extract 43 tokens it was necessary to examine
1,031,755 words of narrative text. There is a steady rise up to the 1600s, where it
occurs at a rate of 6 tokens per 10k words, before experiencing a steady decline
into the 1900s.

Operationalization in corpus linguistics

The operationalization of a linguistic phenomenon in the analysis of linguistic cor-

pora is hardly a novel concept. Glynn (2010: 3) observes that operationalization
“is a basic analytical tool – the definition of an object of study by how one can
measure it” that allows for the researcher to examine the accuracy of the mea-
sure and falsify predictions. Operationalization can draw upon qualitative factors,
simple arithmetic, or rely on correlational statistics. Schmid (2010: 123) points
out that while having statistical power behind complex measures such as the ones
proposed in collostructional analysis certainly lends credence to claims of a cer-
tain phenomenon’s strength, an overreliance upon correlational values may mask
certain distributional patterns.
A purely qualitative operationalization is represented by Lakoff ’s (1987) dis-
tinction between similarity and difference in categorization, which allowed for
an operationalization of meaning (Glynn 2010: 7). Operationalization along the
lines of simple arithmetic are found in Altenberg’s (1998) identification of recur-
rent word combinations defined simply as sequences with more than one occur-
rence in the data, used also by Zeschel (2010: 208). In a slightly more complex
measure Bybee & Torres Cacoullos (2008, 2009) operationalized a grammatical-
ization index by quantifying adjacency, association and fusion of copula + gerund
constructions in Spanish. Likewise, the present research uses factors specific to the
usage of the constructions studied in order to arrive at a score that indexes the level
conventionalization. While correlational scores are not used, it is hoped that the
Conventionalization Index proposed here provides the explanatory power to shed
light on this gradient phenomenon (or at least provides an innovative b ­ eginning
 Damián Vergara Wilson

for further study). Before delving into the Conventionalization Index, we turn to
another operationalization that relies upon simple arithmetic.

Conventionalized Instances of Constructions (CICs)

Conventionalized constructional sequences are pervasive in discourse. Erman and

Warren (2000) advance the term prefab as “a combination of at least two words
favored by native speakers in preference to an alternative combination which
could have been equivalent had there been no conventionalization” (31) and pro-
vided evidence that they account for at least 55% of written and spoken discourse
(50). In a less conservative proposal Altenberg (1998) proposed that 80% of all
words in an oral corpus of British English occur in recurrent word combinations.
Recognizing the pervasiveness of formulaic language, Sinclair (1991: 110) pro-
posed the idiom principle: “… a language user has available to him or her a large
number of semi-preconstructed phrases that constitute single choices even though
they might appear analyzable into segments”. By all accounts formulaic language
is ubiquitous. Yet the assertions by these authors do not account for the gradient
nature of conventionalization.
The operationalization of Conventionalized Instances of Constructions
(CICs)4 in Wilson (2014) served as a threshold to identify conventionalized com-
binations in a way that could be readily applied to quedar(se) + ADJ. The CIC
threshold draws upon the token per type ratio as a starting point. Being an average
of the number of tokens appearing per type in a given century, it is a fairly conser-
vative estimate to propose that combinations that have twice the average number
of tokens should be considered conventionalized. Therefore, the CIC threshold
was calculated in Wilson (2014: 73) as the “(Token/Type ratio) x 2, rounded up or
down to the nearest whole number”.
There are three levels of conventionalization at play here: the overall normal-
ized frequency of quedar(se) + ADJ, the productivity of this highly schematic con-
struction in terms of generating a plethora of types, and the conventionalization
of specific verb + adjective combinations found in the data. Barðdal (2008) argues
that a construction with low type frequency will be most productive if these types
have both high token frequency and high semantic cohesion; frequently occur-
ring types will serve as analogical models for extending usage to semantically

. The phrase “conventionalized instances of constructions” was articulated in Bybee &

Torres Cacoullos (2008: 3) though not as a term that was operationalized to define a particular
set of tokens or types.
 Gradient conventionalization of the Spanish expression 

related types. Barðdal also proposes that productivity in constructions with high
type frequency will happen through an extension of overall categories regardless
of the presence of frequently occurring combinations. While these points will be
returned to below, it is relevant to the present discussion to highlight the impor-
tance of relative frequency of types and tokens. The data reveal that there is an
overall tendency for quedar(se) + ADJ to display high type frequency. Yet this
masks the finding that there are combinations with high token frequency that
appear consistently over the time span studied, such as quedar(se) solo. Therefore
conventionalization, to a degree, will be dependent upon the portion of tokens for
which a certain combination accounts given the spectrum of singletons to prefabs
in quedar(se) + ADJ.
The portion of tokens and types accounted for by CICs relates to the over-
all rates of the construction quedar(se) + ADJ. Table 2 provides overall figures
for the CICs identified in the data. In the centuries where the construction
quedar(se) + ADJ occurred at the highest rate per 10k words (the 1600s) CICs
accounted for the greatest portion of types (14.5%) and tokens (43%). Inversely,
the century with the lowest normalized frequency, the 1200s, had the lowest por-
tion of CIC types (10.5%) and tokens (21%). Therefore, when a schematic con-
struction, such as quedar(se) + ADJ, becomes more frequent in discourse (# per
10k words), conventionalized combinations will proliferate.

Table 2. Overview of quedar(se) + ADJ types and tokens above the CIC threshold (based
on Wilson 2014: 74, Table 4.4)
N total N CIC N total N CIC CIC
types types (%) tokens tokens (%) threshold

1200s 29   3 (10.5%) 43 9 (21%) 3

1300s 40 4 (10%) 76 21 (28%) 4
1400s 121 16 (13%) 196 74 (38%) 3
1500s 153 13 (8%) 271 93 (34%) 4
1600s 171   24 (14.5%) 280 119 (43%) 3
1700s 135   14 (10.5%) 253 94 (37%) 4
1800s 141 10 (7%)   255 81 (32%) 4

The Conventionalization Index

Identifying potential CICs provides a useful baseline for identifying formulaic

types in a way that excludes the least conventionalized ones, but does not index
the level of conventionalization of the types identified. Nor does it take into
 Damián Vergara Wilson

c­ onsideration previous usage, or widespread usage. As Bybee and Torres Cacoul-

los (2008) propose, instead of a discrete partition between fixed expressions and
productive formations, there is a continuum. The purpose of the Conventional-
ization Index is to operationalize the gradient level of formulaicity of these CICs
by employing usage-based dimensions. This index provides a score for the con-
ventionalization of a particular verb + adjective combination based on previous
usage, the number of works that included that type, the CIC threshold, and the
overall normalized frequency of quedar(se) + ADJ as seen in Table 3.

Table 3. Overview of factors used in calculating the Conventionalization Index

Factor Condition Score

Previous usage Combination occurred in a 1 point

preceding century
Number of works Number of works in which the type 1 point per work
appears
CIC threshold The number of tokens above the 0.5 points for each
CIC threshold
Overall Normalized The ONF of the century in which Multiplier for the sum of points
Frequency the combination appeared

Previous usage
Previous usage affects subsequent usage. Through the passage of time individual
speakers accumulate rich memories of experience with language that are nested
within a broader cycle of evolution at the level of the speech community (e.g.
Wedel 2006). If a certain combination of quedar(se) + ADJ appeared in a
previous century, it indicates that this combination did not arise spontaneously,
or as a novel type, but had been transmitted from one generation of speakers to
the next. Therefore, types with precedent in previous centuries were assigned
one point.

Number of works
Some types were exclusive to one work and may reflect author preference; an
individually preferred type might appear more robust in usage if a particular
author used it many times. If a work appeared in works by multiple authors,
however, it is a strong an indicator that its usage was widespread. As Schmid
(2010, 2015) notes, conventionalization is a social phenomenon that involves
speaker accommodation. Consequently, one point was assigned to a combina-
tion for each work in which it appeared.
 Gradient conventionalization of the Spanish expression 

CIC threshold
The CIC threshold provides a useful figure from which to depart in indexing con-
ventionalization. First, it is essentially a recognition that a certain combination
occurred at double the average rate. Second it considers the token per type ratio, an
indicator of the internal variability of a schematic construction (Butler 1998). One
half point (0.5) was assigned to each type for the number of tokens that appeared
in the data above the CIC threshold for the century in which it appeared. Because
there was a great range of token frequency among the CICs in a particular century,
(for example, in the 1800s there were 29 occurrences of quedar(se) solo ‘to be left
alone’ and only four occurrences of quedar(se) convencido ‘to be convinced’), the
rationale for applying one half point was twofold. Primarily, it was done in order to
reduce the range of the score. If one point were assigned for each occurrence, the
range is greatly exaggerated. Secondly, some types may have a score of zero for this
factor by occurring exactly at the CIC threshold, but may have points for others
factors that indicate conventionalization. By reducing the weight of this factor, the
types with lower token frequency but other indicators of conventionalization will
not appear as distant from frequently occurring combinations that may lack other
indications of formulaicity.

Overall normalized frequency

Finally, as frequency in usage is arguably one of the most important artifacts of
conventionalization (e.g. Langacker 1987)5, the normalized overall token fre-
quency of quedar(se) + ADJ for a given century was used as a multiplier of the
points assigned to each type. In using the normalized frequency as a multiplier,
this step recognizes the overall rate of usage of the construction as a key factor in
the conventionalization of the individual types. In other words, if quedar(se) + ADJ
accounts for a higher portion of textual data, recurrent combinations will conse-
quently be more conventionalized.
The number of CICs in each century varied. Table 4 displays the number
of CICs and the average Conventionalization Index score by century. The over-
all type frequency of quedar(se) + ADJ and the overall normalized frequency
(ONF) of occurrences of quedar(se) + ADJ per 10k words appear in the last two
columns.

. But note Schmid’s (2010: 116) criticism that Langacker “apparently conceives of frequency
in a vacuum”.
 Damián Vergara Wilson

Table 4. Conventionalization per century 1500s – 1800s

# CICs Avg. Conv. Index score Type frequency ONF

1300s 4 3.7 40 0.96

1400s 17 5.9 121 1.18
1500s 13 30 153 4.85
1600s 24 31.5 171 6
1700s 14 29.5 135 5.82
1800s 13 20 141 2.64

The centuries with the lowest overall normalized frequency, the 1300s, 1400s
and 1800s, were also the centuries with the lowest average Conventionaliza-
tion Index scores (3.7, 5.9, and 20 respectively). The inverse was true of the
centuries with the highest overall normalized frequencies with the 1600s hav-
ing the highest ONF, the highest type frequency, and the highest average Index
scores. The relationship between overall normalized frequency and average
Index score is largely attributable to the use of the ONF as the multiplier. At the
same time, a relatively low ONF does not prohibit certain types from yielding
a high score; the third highest score was ascribed to quedar(se) solo (72.6) in
the 1800s (ONF = 2.64). Finally, the average Conventionalization Index score
hovers at around 30 from the 1500s to the 1700s regardless of the ONF, which
is further evidence that using the ONF as a multiplier does not exaggerate
scores unreasonably. In the following sections, we examine the interaction of
the Conventionalization Index and two semantic categories of adjectives found
in quedar(se) + ADJ.

The case of quedar(se) solo ‘to be left alone’

The combination quedar(se) solo ‘to be left alone’ is one of the most enduring types
found in the data and, due to its relatively high token frequency, is the proposed
central member of a category that will maintain semantic cohesion through the
span of the QAD data. This category has synonymic and antonymic types that,
used in this change-of-state construction, denote a change in which the subject
has either gained (e.g. quedar(se) casado ‘to get married’) or lost (e.g. quedar(se)
viuda ‘to get become widowed’) human company. The proposed exemplar clusters
centering on quedar(se) solo show trends of increasing productivity throughout
the period of the data examined. Figure 1 displays a semantic map of the types of
 Gradient conventionalization of the Spanish expression 

adjective attributed to the solo clusters in the 1400s. As with the semantic maps in
Bybee and Eddington (2006), the size of the circles reflects the token frequency
of a particular type as a way of representing its robustness. Synonymic types are
placed in the same bubble and overlap of bubbles represents semantic similarity.
Arrows represent an oppositional semantic relationship. In Figure 1, we see a cat-
egory of adjective types used in quedar(se) + ADJ that denote changes regarding
the presence or absence of human company.

desamparado casado
‘separated, ‘married’ (1)
unprotected’ (1)

perdido
‘lost’ (2)
viudo
‘widowed’ (6)

huérfano
‘orphaned’ (2)

solo
‘alone’ (8)
sin + HUMAN
-heredero ‘heir’ (1)

Figure 1. The solo clusters in the 1400s (Wilson 2014: 99, Fig. 5.3)

During the centuries preceding the 1600s, Wilson (2014: 79–91) presents evi-
dence that quedar(se) + ADJ was in the process of attracting adjectives away
from the synonymic construction fincar(se) + ADJ as evinced by changing pat-
terns of adjective overlap and constructional productivity. In a comparison of
adjective types found in these two constructions from the 1200s to the 1400s in
samples from the CDE, Wilson (2014: 88) found that overlapping types repre-
sented 53% of all types found to occur more than once in quedar(se) + ADJ. One
of the types that showed a great deal of overlap was solo occurring nine times
in quedar(se) + ADJ and eight times in fincar(se) + ADJ (Wilson 2014: 88) in
these samples.
After the 1600s the construction quedar(se) + ADJ declines in terms of
overall normalized frequency yet the combination quedar(se) solo endures as
 Damián Vergara Wilson

does the broader category. In the 1800s, as seen in Figure 2, we see evidence
of the conventionalization of established adjective types and extension to new
ones. The adjectival types viudo ‘widowed’, huérfano ‘orphaned’, perdido ‘lost’,
and sin padre ‘without a father’ appeared in the data from the 1500s to 1800s,
which evinces category cohesion over time. In terms of extendability, six new
types appear in the solo clusters in the 1800s. Therefore, despite a fall in nor-
malized frequency of quedar(se) + ADJ from a peak of 6 occurrences per 10k
words in the 1600s to 2.64 in the 1800s, the solo categories remain predomi-
nant and productive.

bajo mi protección
‘under my protection’ (1)
desamparado
‘separated, unprotected’ (1)
rezagado ‘left behind’ (1)
abandonado ‘abandoned’ (2)
perdido ‘lost’ (2)

viudo
solo ‘widow(ed)’ (2)
‘alone’
huérfano
(29)
‘orphan(ed)’ (7)

a solas sin + human

‘alone’ -padre/madre
(1) ‘father/mother’ (3)
-hijo ‘child’ (2)
-ella ‘her’ (2)
-jefe ‘boss’ (1)
convidado
‘invited’ (1)

unido ‘united’ (1)

reunido ‘reunited’ (1)

Figure 2. The solo clusters in the 1800s (Wilson 2014: 105, Fig. 5.7)

Yet the question remains as to how this set of exemplar clusters would not fall prey
to the waning of the overall construction quedar(se) + ADJ. To address this ques-
tion we turn to the proposed Conventionalization Index, as shown in Table 5. The
century of occurrence is shown in the first column, the CIC threshold for that cen-
tury in the second column, and the number of tokens in the third. The subsequent
columns provide the figures used in calculating the Conventionalization Index for
quedar(se) solo.
 Gradient conventionalization of the Spanish expression 

Table 5. Conventionalization Index of quedar(se) solo 1300s – 1800s

CIC Prev. # # above Total Convent.
Ctry Thrshld # Tokens Ctry. Authors CIC points ONF index

1300 4 4 1 2 0 3 0.96 2.88

1400 3 8 1 7 2.5 10.5 1.18 12.39
1500 4 12 1 4 4.5 7 4.85 33.95
1600 3 10 1 10 3.5 14.5 6 87
1700 4 5 1 2 0.5 3.5 5.82 20.37
1800 4 29 1 14 12.5 27.5 2.64 72.6

The final column of Table 5 shows that the combination quedar(se) solo goes from
a very low score on the Conventionalization Index in the 1300s (2.88) to a peak
of 87 in the 1600s before falling to 72.6 in the 1800s. The Index score of 20.37 for
the 1700s is idiosyncratic considering the data from the surrounding centuries.6
In the early period one of the factors to consider in interpreting the low
Index score of 2.88 in the 1300s and 12.39 in the 1400s is the previously noted
overlap with the nearly synonymic construction fincar(se) + ADJ. Overall, the
verbs fincar(se) and quedar(se) shared a variety of semantic domains including
the seemingly dichotomous senses of ‘remaining’ and ‘becoming’ in randomized
samples of 200 tokens per century extracted from the CDE (Wilson 2014: 79–91).
In these samples, fincar(se) solo was prevalent and five tokens in the sample for the
1300s came from four different authors, which is evidence that this combination
was widespread. In examining overall rates of these verbs in the CDE, Wilson
(2014: 80) found that in the 1200s the verb fincar(se) in all of its senses occurred at
a rate of 6.07 times per 10k words while quedar(se) occurred at 0.84/10k. However,
by the 1500s the situation had changed inversely and the verb quedar(se) in all of
its senses occurred at a rate of 11.5/10k words while fincar(se) occurred at a rate
of 0.08/10k. In an overview of the two verbs in usage, Wilson (2014) documented
that as quedar(se) gained in overall frequency, extendability, and productivity, the
verb fincar(se) declined in frequency and underwent semantic narrowing. What
is more, overlap of adjectives between two verbs of ‘becoming’ reveals individual

. This idiosyncrasy of the 1700s has been seen in other aspects of the investigation docu-
mented in Wilson (2014). It is attributed to the manner of extracting roughly 250 tokens of
quedar(se) + ADJ per century in the QAD, which was met by extracting tokens from only
11 works comprising 435,055 words. This is vastly different from surrounding centuries (see
Table 1). Therefore, author-specific tendencies may not be masked in the 1700s to the same
extent that they are in other centuries.
 Damián Vergara Wilson

authors’ variation in usage: of the 98 works that comprised the randomized sam-
ples of fincar(se) and quedar(se) from the 1200s to the 1400s in the CDE (I exclude
the 1500s because of the attrition of fincar(se)), 16 of the authors demonstrated
variable usage of quedar(se) + ADJ and fincar(se) + ADJ as expressions of ‘becom-
ing’. In studies of sociolinguistic variation, variability is taken as evidence of a
change in progress and various methods are used in order to determine the direc-
tion of change. Here, we see that the direction of change favors the expansion of
quedar(se) + ADJ and ongoing conventionalization of quedar(se) solo. Therefore,
we see compelling evidence that the expansion of one construction happens at the
expense of another. Frequent combinations play a crucial role in this expansion.
As the combination quedar(se) solo becomes more conventionalized, reaching a
peak score of 87 in the 1600s, the category to which it belongs maintains semantic
cohesion in terms of the retention of previously used adjectives while productively
expanding to new types.
Past the 1600s, we see a similar phenomena to the one experienced by fincar(se)
There was a loss in type productivity of the construction quedar(se) + ADJ that
accompanied the fall in normalized frequency from 6 occurrences per 10k words
down to 2.64 in the 1800s. By the 1800s, other categories of adjectives had fallen
into disuse through being attracted into usage with other constructions. For exam-
ple, adjectives belonging to the rico/pobre ‘rich/poor’ clusters ceased to be used
with quedar(se) + ADJ while being attested in ponerse, hacerse, and volverse + ADJ.
Despite the narrowing of quedar(se) + ADJ and the concomitant expansion of
other expressions of ‘becoming’, particularly ponerse + ADJ (Wilson 2014: 91–92),
we see that quedar(se) solo retains a relatively high conventionalization score of
72.6 in the 1800s, which had an average Index score of 20 in this century. Even
though it is lower than the peak Index score of 87 in the 1600s, it remains a robust
combination showing the second highest score calculated.
The high degree of conventionalization of this central category ­member,
quedar(se) solo, appears to protect the category from falling into disuse
and contributes to ongoing productivity even in the face of constructional
decline. In the 1600s, there were five types in the quedar(se) solo category that
accounted for only 3% of all types in the data largely due to overall productiv-
ity of quedar(se) + ADJ (Wilson 2014: 110): most of the categories of adjectives
appearing in this construction showed maximum analogical extension in the
1600s as evinced by the fact that it is also the century with the highest type
frequency (N = 171). By the 1800s, the 16 types attributed to the quedar(se)
solo category accounted for 11.5% of all types (N = 141). What is more, 6 of
these 16 types had never appeared in the data previously, which is an indication
of enduring productivity through ongoing analogical extension. Even as the
construction quedar(se) + ADJ loses ground to other expressions of ‘becoming’,
 Gradient conventionalization of the Spanish expression 

the conventionalization of the prefab quedar(se) solo contributes to its ongo-

ing presence in usage, to the endurance of the exemplar category to which it
belongs, and to the productivity of this category in terms of generating new
combinations. According to Barðdal (2008), this would be a case where pro-
ductivity and extendability radiate from a conventionalized combination as it
guides novel usage and provides category cohesion.
The category centered on quedar(se) solo is highly centralized. During the
time span for which the Conventionalization Index was calculated, only two other
CICs identified in the data pertained to the solo category: quedar(se) viudo ‘to be
left widowed’ in the 1400s (Index Score: 8.85), and quedar(se) huérfano ‘to be left
orphaned’ in the 1800s (Index Score: 19.8). The low number of CICs in this dia-
chronic category highlights the finding that the solo clusters were highly central-
ized, and it appears that the development of the category was dependent on the
conventionalization and endurance of quedar(se) solo.
When schematic constructions wane and some semantic categories disappear
in usage, other persistent categories naturally account for a higher portion of the
data sample for a given century. With less competition, in other words, frequently
occurring combinations such as quedar(se) solo will surface more prevalently in
the data and the category surrounding it will gain greater relative prominence.
Erosion does not happen to all categories at the same rate and as the waves of
time recede pinnacles of more conventionalized combinations and their perti-
nent categories endure with less attrition as others are washed away. Therefore,
the high degree of conventionalization of quedar(se) solo protects related types
from being attracted into usage with competing expressions of ‘becoming’, namely
ponerse + ADJ.

The case of the quedar(se) alegre / satisfecho ‘to be left happy/satisfied’

clusters

The quedar(se) alegre / satisfecho ‘to be left happy/satisfied’ clusters follow a differ-
ent evolutionary path than the quedar(se) solo clusters. Instead of having a con-
stant, enduring central member, as with the solo clusters, the alegre / satisfecho
clusters show evidence of having both simultaneous and changing types serve
as organizing forces in expressing changes of emotional state, including sadness.
While alegre had an anemic presence in the early period with one token in the
QAD during the 1200s and one in the 1300s, and no other synonymic types in the
category, it gained in token frequency in the 1400s and there was a concomitant
expansion in type frequency to synonymic adjectives. However, this category does
not show the diachronic cohesion of the solo clusters.
 Damián Vergara Wilson

The adjectives in the alegre / satisfecho clusters denote emotional changes

along a gradient semantic range and by the 1600s this category had reached maxi-
mal productivity in terms of extension to new types. As seen in Figure 3, in the
1600s we see types such as ufano ‘cheerful’, con incredible contento ‘with incredible
happiness’, and brioso ‘with zeal’ linked to types such as quieto ‘still, calm’, sosegado
‘calm’ and con tal reposo ‘with such calmness’ through the intervening central
members contento ‘happy’, satisfecho ‘satisfied’, and alegre ‘happy’. This category,
therefore, is more extended analogically and less centralized than the solo clusters.

como una estaua

mudo ‘mute like a
statue’ (1) sin ninguna palabra
transformado en ‘w/o a word’ (1)
estatua ‘transformed mudo
into a statue’ (1) ‘mute’(3)

inmoble ‘motionless’
(1)

quieto
dormido ‘still, calm’ (1) satisfecho
‘asleep’ (1) ‘satisfied’ (9)
sosegado
muerto entregado ‘calm’ (1) contento
‘dead’ (12) al sueño ‘happy’ (7)
con tal quietud
‘delivered
‘w/such calmness’ (1) ufano ‘cheerful’ (4)
into sleep’ (1)
con tal reposo gozoso alegre
desmayado
‘w/such rest’ (1) ‘joyful’ (2) ‘Happy’ (3)
‘fainted’ (1)
con incredible contento
‘with incredible happiness’ (1)
brioso ‘with zeal’ (1)

despierto
‘awake’ (1) apasionado
atento ‘impassioned’ (1)
‘attentive’ (1) enamorado
‘in love’ (2)
cuidadoso
‘careful, wary’ aficionado
(1) ‘fond’ (2)

Figure 3. The alegre / satisfecho clusters in the 1600s (Wilson 2014: 144, Fig. 7.5)

There are two other important factors that characterize the alegre / satisfecho clus-
ters. First, as indicated above, the types listed in the previous paragraph become
connected through intervening types that serve as links to types that would
otherwise be unrelated (see Wilson 2014: Chapter 7). These chains of ­family
 Gradient conventionalization of the Spanish expression 

resemblance were also noted in the manifestation of these clusters in Bybee and
Eddington’s (2006) data, which by then were proposed to center on quedar(se)
quieto ‘to become calm’. In the QAD data for the 1600s types such as mudo ‘mute’
and brioso ‘with zeal’ are related through the following chain: brioso – alegre –
contento – satisfecho – quieto – mudo ‘with zeal – happy – happy – satisfied –
calm – mute’. Secondly, the fact that there are numerous synonyms to prominent
types led Wilson (2014) to propose that the attraction of one of these adjectives
into quedar(se) + ADJ indicates potential productivity in terms of the possibility
for expansion. This potential becomes kinetic evolutionary energy as the related
types appear in the construction in a multiplicitous fashion; there is a precipitous
increase in new types in the immediate alegre / satisfecho cluster from one type in
the 1300s to 11 synonymic types in the 1400s.
Three key CICs in this category show gains in the Conventionalization Index
that peak in the 1600s leading to a scenario in which a group of related types con-
tributes to the extension of semantic categories. Table 6 presents the relevant fig-
ures for these three CICs: quedar(se) alegre, satisfecho, & contento from the 1400s
to the 1800s.

Table 6. Conventionalization Index of quedar(se) alegre, contento, & satisfecho

CIC Prev. # # above Total Convent.
Ctry Thrshld # Tokens Ctry. Authors CIC points ONF index

Alegre:
1400 3 4 1 3 0.5 4.5 1.18 5.68
1600 3 3 0 3 0 3 6 18
Contento:
1500 4 7 1 5 1.5 7.5 4.85 36.4
1600 3 7 1 5 2 8 6 48
Satisfecho:
1500 4 6 0 4 1 5 4.85 24.25
1600 3 9 1 5 3 9 6 54
1700 4 7 1 3 1.5 5.5 5.82 32.01
1800 4 6 1 3 1 5 2.64 13.2

In the early period from the 1200s to the 1400s, the type quedar(se) alegre ‘to get
happy’ was a clear central member of this emerging category. In the QAD this
type appeared once in the 1200s and once in the 1300s. By the 1400s, however,
the combination quedar(se) alegre contributed four tokens from three authors
and the calculated Conventionalization Index score was 5.68. This elevation in
 Damián Vergara Wilson

c­ onventionalization occurs with a notable expansion to related types with eleven

new adjectives appearing in the central alegre category.7
This notable expansion is attributed partially to the potential productiv-
ity represented by the existence of many synonyms and types related to alegre,
which are plausible candidates for analogical extension. Also, many of these
synonyms were used in fincar(se) + ADJ. For example, the following types co-
occurred between fincar(se) + ADJ and quedar(se) + ADJ from the 1200s to the
1400s (Wilson 2014: 89): alegre ‘happy’, confortado ‘comforted’, sosegado ‘calmed’.
As canonical opposites are related (Bybee and Eddington 2006; Murphy 2003;
Murphy 2006), it is worth mentioning that triste was a type that co-occurred
between these two ‘become’ verbs and it contributed to the general expansion of
quedar(se) + ADJ.8
Yet alegre does not maintain its status as a clear central member. Although it
appeared once in the data in the 1400s, the combination quedar(se) contento ‘to
become happy’ appears in the 1500s above the CIC threshold with seven tokens
and displays an Index score of 36.4. In this same century, alegre did not appear at
a rate above the CIC threshold while a newly appearing type, quedar(se) satisfecho
‘to be left satisfied’ shows an Index score of 24.25. Although alegre does resurface
as a CIC in the 1600s with an Index score of 18, it is clearly no longer the central
member of this category.
As category evolution continues, the types alegre and contento do not endure
in quedar(se) + ADJ past the 1600s and it appears that the expansion of pon-
erse + ADJ is partially responsible. Contento shows evidence of overlap by also
occurring in ponerse + ADJ in the 1600s and 1800s in a study by Balasch (2008)
and in the 1900s (Bybee & Eddington 2006), as do the emotion-related types
serio ‘serious’ and triste ‘sad’. Three other emotion-related types also co-occurred
between these two ‘become’ verbs in the 1600s and 1800s: encendido ‘lit up (with
emotion)’, fuera de sí ‘beside him/herself ”, furioso ‘furious’. However, alegre is not
attested as occurring with any verbs of ‘becoming’ past the 1600s (Balasch 2008;
Bybee & Eddington 2006; Wilson 2014).

. These adjective types were: sin queja ‘without complaint’, sin pena ‘without grief ’, sosegado
‘calmed’, acabado ‘satisfied’, asosegado ‘calmed’, confortado ‘comforted’, contento ‘happy’, ufano
‘cheerful’, gozozo ‘joyful’, en paz ‘in peace’, and limpio de odio ‘clean of hate’.
. While opposites are not analyzed in detail here, they are related semantically.
Bybee and Eddington (2006) propose that they share the same general semantics but differ in
one key feature. Wilson (2014) found that participants rated opposite adjective pairs as more
similar than semantically unrelated ones in a Multi-Dimensional Scaling task.
 Gradient conventionalization of the Spanish expression 

In the quedar(se) alegre / satisfecho clusters we see that the type alegre
opened the door for many related types to be attracted into the schematic slot
of quedar(se) + ADJ. Alegre was surpassed in conventionalization by contento and
satisfecho in a cluster of adjectives that described positive emotional states. Once
again the Conventionalization Index gives clues as to the trajectory of a given type.
In the 1600s satisfecho had reached an Index peak of 54 and contento had reached a
peak of 48. Despite a similar level of conventionalization, contento did not appear
in the QAD data above the CIC threshold and went on to show covariation with
ponerse + ADJ. Quedar(se) satisfecho, on the other hand, retained a relatively high
but waning degree of conventionalization in the 1700s (Index: 32.01) and 1800s
(Index:13.2); it also appeared in the 1900s in Bybee and Eddington’s (2006) data.
An examination of the evolution of the alegre / satisfecho clusters in relation-
ship to conventionalized types sheds light on Barðdal’s (2008) proposal that pro-
ductivity of categories with high type frequency will occur through abstraction of
the general category. First, this set of adjectives appeared as a fairly narrow cat-
egory in the early centuries that grew in type frequency with the emerging conven-
tionalization of centralized members that express happiness: alegre, satisfecho, and
contento. Therefore, at first the expansion was more in line with the kind of expan-
sion that Barðdal describes as emanating from cohesive semantic categories with
fewer types but highly conventionalized central members. Over time the nature of
productivity appears to change. Due to the potential for expansion represented by
the large number of synonyms to these centralized members, there is an explosive
expansion in terms of type frequency. By the 1500s onward the productivity is
potentially explained by abstraction over the category. As time continues, conven-
tionalized combinations, such as quedar(se) satisfecho help to stem the rapid loss
of category members from quedar(se) + ADJ to ponerse + ADJ.

Concluding remarks

The present work highlights the utility in developing a gradient usage-based

scale of conventionalization that can be applied to verb + adjective expressions
of ‘becoming’ in different time periods. The study of change-of-state expressions
shows that variation is indicative of greater constructional changes in which one
verb attracts adjectives away from other constructions. Yet, the waxing and wan-
ing conventionalization of a given type does not happen in isolation and may
affect an entire category. A robust centralized member will contribute to the ongo-
ing preservation and productivity of a semantic category even in the face of overall
frequency reduction. While both exemplar categories analyzed here showed a ten-
dency for more conventionalized combinations to withstand waning trends of the
 Damián Vergara Wilson

construction, the loss of types documented in the alegre / satisfecho is more conse-
quential. The lack of a clearly predominant central member of a category may be a
function of the number of synonymical adjectives that serve as potential extenders
of the category, as is the case with the alegre / satisfecho clusters – and these extend-
ers may co-occur in semantically similar constructions. However, the number of
options available for speakers to express nuances of ‘happiness’ or ‘contentment’
may prevent members of this category from reaching the levels of conventionaliza-
tion reached by quedar(se) solo. The alegre/satisfecho clusters represent a semantic
category of adjectives that express changes into nuanced emotional states while
the concept of being left alone without human company is much more binary and
plausibly centralized. It follows that there is a prototype effect that elevates the level
of conventionalization of quedar(se) solo in a way that is not found in the alegre/
satisfecho category. Additionally, it appears that the differing category structures
may also partially account for the different patterns of type loss. We have evidence
here that less cohesive categories are prone to the encroachment of other construc-
tions, such as ponerse + ADJ while centralized ones endure, most likely due to the
conventionalization of a central member. However, even the highly conventional-
ized prefab quedar(se) solo is not immune to ongoing evolutionary trends and we
should be on the lookout for the emergence of new verbs of ‘becoming’.

References

Altenberg, Bengt. 1998. On the phraseology of spoken English: The evidence of recurrent word
combination. In Phraseology, Anthony P. Cowie (ed.), 101–122. Oxford: OUP.
Balash, Sonia. 2008. Construction of poner(se) ADJ: 17th–19th century. Final project for gradu-
ate seminar Frequency of Use and Emergent Grammar, Professor Bybee, UNM, Spring
2008.
Barðdal, Jóhanna. 2008. Productivity: Evidence from Case and Argument Structure in Icelandic
[Constructional Approaches to Language 8]. Amsterdam: John Benjamins.
doi: 10.1075/cal.8
Butler, Christopher. 1998. Collocational frameworks in Spanish. International Journal of Corpus
Linguistics 3: 1–32. doi: 10.1075/ijcl.3.1.02but
Bybee, Joan. 2006. From usage to grammar: The mind’s response to repetition. Language 82:
711–733. doi: 10.1353/lan.2006.0186
Bybee, Joan. 2010. Language, Usage and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan & Eddington, David. 2006. A usage-based approach to Spanish verbs of “becoming.”
Language 82: 323–355. doi: 10.1353/lan.2006.0081
Bybee, Joan & Torres Cacoullos, Rena. 2008. Phonological and grammatical variation in exem-
plar models. Studies in Hispanic and Lusophone Linguistics 1: 399–413.
doi: 10.1515/shll-2008-1026
 Gradient conventionalization of the Spanish expression 

Bybee, Joan & Torres Cacoullos, Rena. 2009. The role of prefabs in grammaticization: How the
particular and the general interact in language change. In Formulaic Language, Vol. 1: Dis-
tribution and Historical Change [Typological Studies in Language 82], Roberta Corrigan,
Edith A. Moravcsik, Hamid Ouali & Kathleen M. Wheatly, 187–217. Amsterdam: John
Benjamins. doi: 10.1075/tsl.82.09the
Coste, Jean & Redondo, Augustín. 1965. Syntaxe de l’espagnol moderne. Paris: Société d’Edition
d’Enseignement Superieur.
Crespo, Luis A. 1949. To become. Hispania 32: 210–12. doi: 10.2307/333076
Davies, Mark. 2002. Corpus del Español: 100 Million Words, 1200s-1900s. <www.corpusdeles-
panol.org>
Delbecque, Nicole & Van Gorp, Lise. 2015. The pseudo-copular use of the Spanish verbs hacerse
and volverse: Two types of change. CogniTextes 13. <https://cognitextes.revues.org/844>
Delbecque, Nicole & Van Gorp, Lise. 2012. Hacerse y volverse como nexo pseudo-copulativo:
Dos maneras de concebir el cambio en español. Bulletin Hispanique 114: 279–307.
Eberenz, Rolf 1985. Aproximación estructural a los verbos de cambio en Iberorromance. In Lin-
guistique comparée et typologie des langues romanes, 461–475. Aix en Provence: Université
de Provence.
Eddington, David. 1999. On “becoming” in Spanish: A corpus analysis of verbs expressing a
change of state. Southwest Journal of Linguistics 18: 23–46.
Eddington, David. 2002. Disambiguating Spanish change of state verbs. Hispania 85: 921–929.
doi: 10.2307/4141261
Eddington, David. 2004. Spanish Phonology and Morphology: Experimental and Quantita-
tive Perspectives [Studies in Functional and Structural Linguistics 53]. Amsterdam: John
Benjamins. doi: 10.1075/sfsl.53
Erman, Britt & Warren, Beatrice. 2000. The idiom principle and the open choice principle. Text
20: 29–62.
Fente, Rafael. 1970. Sobre los verbos de cambio o “devenir”. Filología Moderna 38:157–172.
Glynn, Dylan. 2010. Corpus-driven cognitive semantics: Introduction to the field. In Quantita-
tive Methods in Cognitive Semantics Corpus-driven Approaches, Kerstin Fischer & Dylan
Glynn (eds), 1–41. Berlin: De Gruyter Mouton. doi: 10.1515/9783110226423.1
Israel, Michael. 1996. The way constructions grow. In Conceptual Structure, Discourse and Lan-
guage, Adele Goldberg (ed.), 217–231. Stanford CA: CSLI.
Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind.
Chicago IL: University of Chicago Press. doi: 10.7208/chicago/9780226471013.001.0001
Langacker, Ronald. 1987. Foundations of Cognitive Grammar, Vol. 1: Theoretical Prerequisites.
Stanford CA: Stanford University Press.
Murphy, M. Lynn. 2003. Semantic Relations and the Lexicon. Cambridge: CUP.
doi: 10.1017/CBO9780511486494
Murphy, M. Lynn. 2006. Antonyms as lexical constructions: or, why paradigmatic construction
is not an oxymoron. Constructions SV1. <www.constructions-online.de>
O’Neill, John. 1999. Electronic Texts and Concordances of the Madison Corpus of Early Spanish
Manuscripts and Printings. Madison, WI: Hispanic Seminary of Medieval Studies, Lt.
Renouf, Antoinette & Sinclair, John. 1991. Collocational frameworks in English. In English Cor-
pus Linguistics: Studies in Honour of Jan Svartvik, Karin Aijmer & Bengt Altenberg (eds),
128–143. London: Longman.
 Damián Vergara Wilson

Schmid, Hans-Jörg. 2010. Does frequency in text instantiate entrenchment in the cognitive sys-
tem? In Quantitative Methods in Cognitive Semantics Corpus-driven Approaches, Kerstin
Fischer & Dylan Glynn (eds), 101–133. Berlin: De Gruyter Mouton.
doi: 10.1515/9783110226423.101
Schmid, Hans-Jörg. 2015. A blueprint of the Entrenchment-and-Conventionalization Model.
Yearbook of the German Cognitive Linguistics Association 3: 3–26. doi: 10.1515/gcla-2015-0002
Sinclair, John. 1991. Corpus, Concordance, Collocation. Oxford: OUP.
Traugott, Elizabeth Closs & Trousdale, Graeme. 2013. Constructionalization and Constructional
Changes. Oxford: OUP. doi: 10.1093/acprof:oso/9780199679898.001.0001
Van Gorp, Lise. 2012. Ponerse como pseudo-cópula: una manera específica de concebir el cam-
bio en español. Actas del XL Simposio Internacional y III Congreso de la Sociedad Española
de Lingüística, Feb. 2011, 395–401. Madrid.
Van Gorp, Lise. 2015. Pseudo-copular use of the Spanish verbs ponerse and quedarse: two types
of change. CogniTextes 13. doi: 10.4000/cognitextes.843.
Wedel, Andrew B. 2006. Exemplar models, evolution and language change. The Linguistic
Review 23: 247–274. doi: 10.1515/TLR.2006.010
Wilson, Damián V. 2009. From “remaining” to “becoming” in Spanish: The role of prefabs in
the development of the construction quedar(se) + ADJECTIVE. In Formulaic Language,
Vol. 1: Distribution and Historical Change [Typological Studies in Language 82], Roberta
Corrigan, Moravcsik, Edith A., Hamid Ouali & Kathleen M. Wheatly (eds), 273–295.
Amsterdam: John Benjamins. doi: 10.1075/tsl.82.13fro
Wilson, Damián Vergara. 2014. Categorization and Constructional Change in Spanish Expres-
sions of “Becoming.” Leiden: Brill. doi: 10.1163/9789004274457
Zeschel, Arne. 2010. Exemplars and analogy: Semantic extension in constructional networks.
In Quantitative Methods in Cognitive Semantics Corpus-driven Approaches, Kerstin Fischer
& Dylan Glynn (eds), 201–219. Berlin: De Gruyter Mouton. doi: 10.1515/9783110226423.201
 The evidence add ups
An affix shift study of prefabs

Clay Beckner
University of Canterbury

In a usage-based model of the lexicon, linguistic elements that repeatedly

co-occur in sequence come to form a processing unit (Bybee 2002b). The present
study supplements previous psycholinguistic research finding that multiword
sequences may form a ‘prefabricated’ sequence. I describe a new experiment
designed to elicit ‘affix shift’ speech errors (e.g., adds up → add ups), which
give evidence of holistic processing of a multiword sequence. Analysis of the
data indicates that affix positioning errors are predicted by a confluence of
factors – a high-frequency stimulus sequence (settle down) in tandem with low-
frequency component words (settle, down). These findings provide support for
a usage-based account, in which linguistic units are not fixed, but gradient and
changeable with experience.

1. Introduction

In recent years, a growing number of writers have argued that a words-and-rules

model of syntax is insufficient, and that speakers rely heavily on formulaic multi-
word chunks, or ‘prefabs’, during speech comprehension and production (Bolinger
1976; Wray 2002). Following Bybee (2010: 35–36), we may describe a prefab as a
multi-word expression which is processed as a unit, but which nevertheless may
retain some internal morphosyntactic structure.
Evidence for prefabs is found in phonetic reduction, semantic and pragmatic
changes, and the relative fixedness of multi-word expressions (Pawley & Syder
1983; Sinclair 1991; Bybee & Scheibman 1999; Erman & Warren 2000; Bybee 2001,
2002a, 2002b; Beckner & Bybee 2009). Additionally, recent behavioral experi-
ments have provided evidence that frequent multi-word sequences are processed
more quickly than their lower-frequency counterparts (Reali & Christiansen 2007;
Tremblay et al. 2011; Arnon & Snider 2010). Such findings are not predicted by
models of the lexicon in which words are assembled according to rules, with no
role for frequency (Pinker 1991).

doi 10.1075/slcs.192.09bec
© 2018 John Benjamins Publishing Company
 Clay Beckner

Even so, experimental evidence of processing boosts need not suggest ‘prefab-
ricated’ status for a multi-word sequence. As observed by Kapatsinski and Radicke
(2009: 500), ‘sensitivity to predictability does not necessarily imply that the pre-
dictor and the predicted fuse into a unit. Rather, co-occurrence may simply make
the co-occurring words prime each other.’ However, a different kind of evidence
may be drawn from monitoring studies, in which subjects are asked to monitor
for a target word (e.g., up) within word sequences that vary in frequency. Such
experiments by Vogel Sosa and MacFarlane (2002) and Kapatsinski and Radicke
(2009) demonstrate that when stimulus sequences are frequent, participants are
slower to recognize the target word. In sum, the behavioral evidence indicates that
there is a holistic component to the processing of frequent sequences: a multi-
word chunk is more readily accessible, while the accessibility of its component
words is diminished.
In this chapter, I investigate a new source of evidence for prefabs, based on the
positioning of affixes in speech. Studies of conversational errors show that speak-
ers at times insert an affix at the wrong position in a sentence, as in It probably gets
out a little → It probably get outs a little (Garrett 1980: 202). Such an error suggests
that get out is processed as a unit, insofar as apparently (1) there is diminished
activation of the component words (including the target verb, get, which fails to
be inflected as a verb), and (2) prefabricated production of the word sequence as a
wordlike unit, which receives a verbal or nominal inflection as a unit.
The relevance of affix positioning errors has been previously noted by Quirk
et al. (1985) in a discussion of multi-word verbs. Quirk et al. remark on speech
errors such as ‘The editor must do precisely as he see fits,’ observed during a radio
interview; the ‘shift of the inflection from the verb to the adjective testifies to a ten-
dency for speakers to perceive the multi-word verb as a single grammatical unit’
(1985: 1151, note a). Quirk et al. say that errors of this type ‘deserve attention’ in
the study of multi-word units, but do not present a study of such errors themselves
(1151, note a). Similarly, Wray (2008) writes that grammatical indications that a
multi-word sequence has ‘morpheme equivalent status’ (that is, that the sequence
lacks internal structure) ‘often come in the form of errors,’ such as the plural form
weapon of mass destructions (119). Moreover, affix shifts such as get-outs parallel a
diachronic process in which inflections may come to be ‘externalized,’ as in sisters-
in-law > sister-in-laws. Haspelmath (1993) writes that as certain expressions ‘come
to be felt as single words, speakers externalize the inflection’ (289).
Additionally, several studies (Stemberger 1984; Stemberger 1985; Stemberger
& MacWhinney 1986b) analyze the occurrence of ‘affix shift’ errors in naturalistic
speech. However, their analyses focus on the storage of affixes in the lexicon, and
do not consider a possible role for multi-word units in the occurrence of errors.
Additionally, these studies make no distinction between different types of affix
 The evidence add ups 

shifts that may be observed. That is, affix errors can be anticipatory, thus poten-
tially shifting inside a wordlike unit (dead_ ends > deads end_), or they may rep-
resent the ‘deflection’ of an affix outside a wordlike unit (adds up > add ups). The
hypothesis of the present study can only directly address the latter kind of error,
since holistic retrieval of a phrase would encourage affixes to attach outside, rather
than inside, a chunked group of words.1
In sum, a number of researchers have observed that affix errors may provide
a useful source of data. However, such discussions have either focused on isolated
examples, or have not performed analyses that would be relevant to the study of
multi-word units. This area is thus in need of systematic study.

1.1 Errors of interest, and theoretical background

Three different types of affix error are of interest in the present chapter, and will be
examined in an experimental study.

1.1.1 Full affix shifts

First, there are the full affix shifts, such as It get outs and She see fits. Errors of
this sort offer the most straightforward evidence in support of the prefab hypoth-
esis, that is, the view that a multi-word sequence may be processed as a unit, with
diminished activation of its components. In full shifts of this sort, the tendency for
inflections to be ‘deflected’ to the periphery of particular multi-word sequences
would attest that the sequence tends to be processed as a whole unit with respect
to morphosyntax, while the morphosyntactic status of component words within
the sequence is diminished.

1.1.2 Double-marked affixes

Secondly, there exist more complex affix errors which suggest that the processing
of a unit is multifaceted. These are ‘double-marked’ errors, such as It gets outs.
Errors of this type append an affix to a multi-word sequence and to one of its
words, implying that with respect to morphosyntax, the speaker has concurrently
activated the multi-word sequence as a whole unit, and as an assemblage of com-
ponent words. Note that double-marked affixes parallel a diachronic process, since

. Errors such as deads end must be attributed to some other mechanism, such as anticipa-
tion of a morpheme that has entered the speaker’s buffer memory for a later portion of the
sentence (Levelt 1989). Also, note that the descriptions in this overview are customized for the
grammar of English. In a language with inflectional prefixes, an early, rather than late, affix
would be of interest for investigating holistic retrieval.
 Clay Beckner

such usages are attested as an intermediate (and often overlapping) stage of inflec-
tional externalization (Haspelmath 1993).
In considering evidence from the distribution of double-marked affixes, I
proceed with the assumption that linguistic units of various types, from words to
syntactic constituents to constructions, are characterized by gradience rather than
sharp delineation (Bybee & Scheibman 1999; Bybee & McClelland 2005; Croft
2001; Beckner & Bybee 2009). Prefabs are no exception to this principle. Iden-
tifying a multi-word sequence as a prefab makes no claim that it has no internal
structure, nor that it can never be assembled word-by-word (Bybee 2010). In any
particular case, the component parts of a unit may be salient to varying degrees
(Hay & Baayen 2005). One underlying cause for such gradience is that representa-
tion of linguistic units is complex and redundant, and multiple modes of access are
in competition with one another.
Along these lines, errors such as gets outs will be taken as indicators that the
multi-word sequence of interest has been activated as a whole, while the morpho-
syntactic status of the individual words has also been recognized.

1.1.3  No-marking errors

When grammatical affixes are unintentionally omitted, such slips may be described
as ‘no marking errors’. No-marking errors are known to be more likely on low-
frequency forms (Stemberger & MacWhinney 1986a, 1986b), apparently because
it is easier to assemble (or retrieve pre-assembled) high-frequency base-plus-affix
pairs. Moreover, zero-marking of third person singular verbs is a feature of non-
standard varieties of English (Labov 1969; Green 2002), and it is possible that
absent verb marking will be a sociolinguistic variable.
While noting these general caveats, in certain contexts it is reasonable to
interpret no-marking errors as evidence of holistic retrieval. Consider a naturalis-
tic error such as (1):

What is the object of our study? The object of our study are, broadly speaking,
(1) 
fourfold: pronunciation, grammar, meaning, and attitudes toward language
change. – Seth Lerer, The History of the English Language, Part I in the Great
Courses audio CD lecture series.

In such an error, the speaker utters a sequence of interest in the discourse (object
of our study), and then – on reintroducing that sequence – fails to alter it to fit the
grammatical context. Wray (2006: 592) observes that holistic sequences are rela-
tively difficult to interrupt or modify. Thus, if a sequence has been retrieved as a
unit (with diminished analyzability), on a subsequent occurrence it may be more
resistant to alterations, including morphosyntactic ones. On this assumption, no-
marking errors can provide evidence of failed interruption, and thus evidence of
holistic retrieval.
 The evidence add ups 

The particular design of this chapter’s experimental task requires participants

to add an inflection to a stretch of speech they have just heard. I thus include
data from no-marking errors, on the assumption that if a word sequence has been
processed as a unit, it is more likely that the speaker will fail to interrupt it on a
subsequent utterance.

1.2 Predictions, and quantitative corpus metrics

The hypothesis of the present chapter is that affix errors will not be distributed
randomly, but will give some indication of which multi-word sequences are likely
to be prefabs. Speech errors may arise for a variety of reasons, but the question is
whether some errors are more likely than others (Fromkin 1973; MacKay 1979).
Specifically, it is predicted that patterns of errors should correspond to patterns
of usage, with the assumption that the formation of prefabs depends on the co-
occurrence of words in sequence. In the dictum of Bybee (2002b: 316), ‘items that
are used together fuse together.’
A question remains, however, regarding how we are to quantify the co-­
occurrence of words. The most intuitive metric would certainly be raw frequency
counts, that is, to count how often some sequence (e.g., a word sequence XY)
appears in a corpus. Indeed, there is broad support for such an absolute frequency
approach in the usage-based literature (Langacker 1987; Ellis 2002; Bybee 2006),
and the psycholinguistic studies cited previously find that raw corpus frequency
is a determinant in processing multi-word sequences (e.g., Kapatsinski & Radicke
2009; Arnon & Snider 2010).
However, an alternative approach is to consider a relative frequency inter-
pretation of co-occurrence (e.g., Hay 2001). In this view, we take note of a word
sequence X Y relative to all the other instances of the component words (that is, X
in the absence of Y, and/or Y in the absence of X). In this second interpretation, it
turns out that even if X Y is a relatively rare sequence, we might say that X and Y
frequently co-occur – as long as X and Y tend to occur together.
One relative frequency measure is Mutual Information (MI) which is said
to indicate a ‘stronger cohesion’ among words (Gregory et al. 1999: 9). More
specifically, for a two-word sequence (a bigram) this measure tells us how much
each word predicts the other. MI quantifies how often the words appear together,
in contrast with how often they occur separately; it is calculated as the frequency
of the bigram, divided by the product of the component-word frequencies2

. Other adjustments are also typically applied, including log-transformation. The quantity
described here is also known as the ‘pointwise mutual information’, whereas in Information
Theory, mutual information refers to a different metric.
 Clay Beckner

(­ Gregory 1999). A series of studies by Nick Ellis and colleagues (Ellis, Simpson-
Vlach & Maynard 2008; Ellis & Simpson-Vlach 2009) provides support for MI
as a determinant in processing multi-word units. In four experiments, they mea-
sure reaction time in an acceptability judgment task; measure fluency in reading
a sequence aloud; measure priming of the final word using voice onset time in
reading aloud; and measure comprehension in context, assessed through reac-
tion time in an accessibility task. All four experiments give evidence that high-
MI multi-word sequences are easier to process. However, there are null results
for absolute frequency.
Despite the apparent success of MI in experimental work, it may need some
fine-tuning as a corpus metric. Manning and Schütze (1999: 181–182) observe
that MI is, among collocational measures, especially sensitive to problems of
‘data sparseness,’ that is, the limits imposed by rare occurrences in small cor-
pora. The measure has a bias that is subject to undue influence from low-fre-
quency events, and, with respect to frequency, it systematically ranks items in a
counterintuitive way.
To counteract the low-frequency bias of MI, a number of researchers (Fon-
tenelle et al. 1994; Thanapoulos et al. 2002; Bouma 2009) have independently sug-
gested modifying the measure by squaring the frequency of the bigram. Such an
approach leads to the following equation, noting that F(w1w2) is the frequency of
the bigram, F(w1) and F(w2) are the frequencies of the component words, and N
is the corpus size.

(Equation 1) MD(w1w2)= log2 N ⋅ F(w1w2)2

F(w1)*F(w2)

I have labeled the quantity in Equation (1) as ‘MD’ to stand for ‘Mutual Depen-
dency,’ a term coined by Thanapoulos et al. (2002). In the experiment design below,
I make use of corpus-derived Mutual Dependency scores to quantify the relative
frequency of stimulus bigrams.
Regarding absolute frequency and relative frequency, it is possible to take an
inclusive approach, following Bybee (2002b: 317, 2010: 46), who argues that both
types of co-occurrence lead to the entrenchment of units. (See also Krug 2003;
Klein & Yu 2009.) The prediction of the current study, then, is that the affix errors
of interest will be more common on bigrams that are high in absolute frequency,
and more common on those high in Mutual Dependency.
To investigate these hypotheses, the present study will thus jointly investigate
Frequency and Mutual Dependency. This approach is not statistically ideal, insofar
as frequency is included in the definition of Mutual Dependency. However, it is
possible to identify multi-word sequences which are high in Mutual Dependency
 The evidence add ups 

while being low in Frequency (or vice versa), as will be evident in the stimuli out-
lined in Section 2.2.1. Moreover, previous research has tested the role of frequency
alongside higher-order measures for multi-word sequences (e.g., Ellis, Simpson-
Vlach & Maynard 2008), and such steps are necessary to test for the relative con-
tributions of different quantitative determinants.

2. Task design

In the task, participants were instructed to listen to a recorded stimulus sentence,

and to repeat back a modified version of the sentence which requires adding the
3rd person singular marker -s on the verb. Each sentence included the pronoun
subject they; female participants were asked to substitute the pronoun subject
she, and male participants were asked to substitute the pronoun subject he. Thus,
for instance, if the stimulus sentence was Despite the ads about switching to green
energy, they depend on contributions from the coal industry, a correct response
would be Despite the ads about switching to green energy, she depends on contribu-
tions from the coal industry.
Affix errors are quite rare in conversation, and thus the experiment was
designed with additional distracting factors in the hopes of increasing the error
rate. First, participants were asked to ‘shadow’ throughout the course of the
experiment (Marslen-Wilson 1973). That is, they were asked to begin echo-
ing back the stimulus sentence immediately, without waiting for the stimulus
presentation to end, thus requiring simultaneous listening and speaking dur-
ing most of the participant’s response. Speech shadowing provides one method
of overloading verbal capacities, thus providing ongoing interference with
participants’ abilities to use language introspectively (Hermer-Vasquez et al.
1999), and prompting more automatic, less carefully analyzed speech output.
In the present experiment, then, the additional requirement of shadowing was
intended to minimize the speaker’s resources for monitoring his or her own
planned speech output.
A second distracting element in the experimental task is based on the
Competing Plans Hypothesis of Baars (1980, 1992), which suggests that speak-
ers often formulate alternate, parallel plans for an utterance, and competition
between these plans sometimes results in errors. A wide range of experiments
have been developed in which errors are elicited by ‘creating competition
between alternative output plans,’ either in language or in other motor activi-
ties (Baars 1992: 130). Often, such approaches encourage errors by alternating
unpredictably between the type of response required from participants. In the
present experiment, we seek to increase the likelihood that participants will,
 Clay Beckner

with respect to affixation, chunk multiple orthographic words together. Thus,

the experiment intermingled bigram stimuli with distractor items which contain
more than one bound root, that is, compound verbs. In the bigrams of interest,
any inserted affixes will be expected on the verb inside the multi-word sequence,
as in gets out, depends on, and sees fit. The Competing Plans Hypothesis predicts
that we may encourage syntagmatic errors by priming an alternate affixation
strategy, in which roots inside a single lexical item must be passed over: sleep_
walks, safe_guards, play_acts, and so on. This alternate strategy may be especially
influential in cases where the first component of the compound verb may be
parsed as a verb (e.g., sleep-walk).

2.1 Materials and stimulus design

2.1.1 Frequency x Mutual Dependency bins
The present affix shift experiment is designed to investigate possible effects from
frequency and MD, both separately and together. The units of interest for this
experiment are all two-word sequences (bigrams). Thus stimuli are selected in
a 2 x 2 design, including High/Low MD bigrams, crossed with High/Low fre-
quency bigrams.
An extensive search was undertaken in order to identify suitable candi-
date stimuli. Bigrams were selected based on frequency counts and MD scores
gleaned from the 95-million-word spoken portion of the Corpus of Contem-
porary American English (COCA, Davies 2008-). Since the first word of each
bigram needs to be a verb, my COCA counts for the bigram’s first word are
constrained by part of speech. Ad hoc thresholds were defined for High and
Low bins, using a trial-and-error approach in order to (a) make each bin large
enough to allow an adequate repertoire of candidates, for purposes of match-
ing features between bigrams (as described below), while also (b) making the
bins designated ‘High’ as selective as possible, on the assumption that poten-
tially idiosyncratic bigram properties will be most evident at the high end of the
appropriate scale (see Alegre & Gordon 1999; Kapatsinski & Radicke 2009). This
approach led to the designation of High Frequency bigrams as those having a
frequency of 505 or more in the corpus, and High Mutual Dependency bigrams
are those having an MD value of 13.6 or more.
The process of identifying matches across the 2 × 2 categories involved a care-
ful, hand-selected search for items which were controlled for numerous features.
Note that if statistics alone are randomly used to select items (see, for instance,
Ellis et al. 2008: 380), it is likely that biases will be introduced, since bigrams will
tend to cross constituent boundaries in some bins more than others. The guide-
lines I used are listed below.
 The evidence add ups 

2.1.2 Bigram features matched across bins

1. Stimuli across categories were matched with respect to the part of speech
of the word following the verb. This approach helped to encourage cross-­
category uniformity for constituency of the sequence, given that similar
sequences (for instance, Verb + Preposition or Verb + Adverb) will tend to
have broadly similar structural features. An attempt was made to include a
range of structures in the searches (Verb + Pronoun, Verb + Mass Noun).
2. Additional heuristics were used to ensure that structural features were
uniform across all four bins. Beyond broad similarities imposed by matching
part of speech sequences, it is possible different bigrams will have varying
grammatical relationships with the surrounding text. Thus, items in the four
bins were matched using grammatical criteria for multi-word verb categories
by Quirk et al. (1985: 1152–1161), including, for instance, intransitive phrasal
verbs, transitive phrasal verbs, and prepositional verbs.
It should be noted that the same word sequence may fall into different cat-
egories depending on the context. For instance, wake up may be intransitive
(They wake up late) or transitive (They wake up the children/They wake them
up) (cf. Quirk et al. 1985: 1158). Where such variation is possible, I selected a
stimulus sentence which matched the structural features of each bigram to its
counterparts in other bins.
3. In all four bins, an attempt was made to include bigrams that exemplify a
range of idiomaticity. Verb + Particle sequences quite typically exhibit some
degree of semantic opacity on the particle. However, a number of bigrams
selected are especially idiomatic, insofar as the two words collectively have a
meaning that is not transparently related to the literal meaning of the verb on
its own (e.g., hold off, meaning ‘wait’). For these more idiomatic cases, match-
ing cases of idiomatic or metaphorical use were found for all four bins.

2.1.3  Bigram stimuli and distractors

The above selection criteria were used to select a total of 56 bigram stimuli.
There were 14 matched bigrams across the four bins. The items used are
presented in Table 1. It may be noted that the corpus metrics are not perfectly
matched across bins; thus, for instance, the High Frequency bigram wake up has
a notably higher MD score than the matched Low Frequency item settle down –
even though both stimuli are in the “High MD” category. Such asymmetries
are to some extent unavoidable due to the definitional link between frequency
and MD. However, this problem is mitigated somewhat below by including
statistical tests with frequency and MD as continuous (rather than categorical)
predictors.
  Clay Beckner

Stimulus bigrams used in the elicitation experiment. Frequency and MD values are based on COCA Spoken data (Davies 2008, 95
million words). Corpus counts for F(w1) are constrained by part of speech.
Low frequency High frequency

Bigram Bigram Bigram Bigram

Bigram freq MD F(w1) F(w2) Bigram freq MD F(w1) F(w2)

1. settle down 204 14.63 1734 90508 1. wake up 1865 19.28 2225 233482
2. screw up 119 14.20 308 233482 2. work out 1839 15.01 36216 270881
3. freak out 60 13.60 102 270881 3. hang out 695 15.61 3391 270881
4. wrap up 276 14.97 971 233482 4. add up 665 14.59 7344 233482
5. tear apart 21 13.76 679 4478 4621 20.18 6326 270881
6. read aloud 24 14.36 18227 143 6. make sure 10647 21.01 103464 49506
7. gain weight 93 16.39 1857 5165 7. stay home 546 14.59 22242 51792
HIGH MD

8. interfere with 383 15.51 550 546640 8. depend on 1103 16.85 1459 673655
211 13.93 4413 61799 9. pay for 4689 17.08 20493 737067
10. arrive at 363 14.26 1584 405469 10. worry about 3974 19.07 6638 411223
11. insist on 378 13.77 1446 673655 11 talk about 28166 21.64 56472 411223
12. trust me 504 14.61 4374 220828 12. call it 4677 15.72 27512 1401667
13. recover from 311 14.30 1390 329733 13. need to 28042 19.03 54108 2590553
14. walk through 393 14.61 8675 68125 14. look at 32791 21.65 76959 405469
MEAN 238.6 14.49 3307.9 224559.2 MEAN 8880.0 17.9 30346.4 572268.7
 The evidence add ups 

(Continued)
Low frequency High frequency

Bigram Bigram Bigram Bigram

Bigram freq MD F(w1) F(w2) Bigram freq MD F(w1) F(w2)
1. move up 214 10.10 17015 233482 1. get down 986 12.24 211305 90508
2. give in 296 6.67 52448 1560158 568 12.70 2963 1560158
204 12.28 12947 61799 3. let go 749 11.45 112634 169428
4. leave out 132 8.42 17881 270881 4. take on 2372 12.96 99948 673655
5. seek out 175 12.10 2455 5. cut out 567 12.87 15082 270881
6. smell bad 11 8.61 1051 28028 6. look good 833 12.59 76959 139555
7. buy food 69 11.10 13644 15179 7. see people 708 10.08 139252 316865
LOW MD

8. walk at 22 3.71 8675 405469 8. point to 824 12.20 5301 2590553

9. speak in 254 8.30 12507 1560158 9. know of 1631 8.26 401401 2055523
10. fear for 136 9.84 2606 737067 10. come with 813 10.15 101723 546640
35 5.93 19821 96897 11. agree on 1330 13.57 20638 673655
84 6.79 4341 1401667 12. forget it 618 11.79 7350 1401667
13. resolve to 33 4.60 1648 2590553 13. hope to 1773 12.51 19776 2590553
14. pay at 53 5.01 20493 405469 14. move to 1242 11.70 17015 2590553
MEAN 122.7 8.1 13395.1 688406.3 MEAN 1072.4 11.8 87953.4 1119300
 Clay Beckner

The stimulus bigrams were used to construct 56 stimulus sentences. Sentences

were loosely based on usages found in the COCA corpus. Sentences were matched
in groups of four, with respect to register and broad semantic domains. The sen-
tences were all approximately matched for length, measured in number of syl-
lables. Sentences were all similar in syntactic structure and complexity, with the
verb bigram occurring after an introductory, dependent clause.
An example of four matched stimulus sentences appears in Table 2.

Table 2. Example of stimulus sentences matched across the four bins. Sentences are
matched for semantic domains – in this case, referring to political or legal settings.
­Bigrams and the surrounding context are also matched for structural features, here
­encompassing a verb followed by a prepositional phrase.
To avoid the throngs of reporters on the steps, they walk at a brisk pace toward the packed
a. 
courtroom. [Low MD, Low Frequency]
As the senator listens and jots a few notes, they point to a variety of problems with the old law.
b. 
[Low MD, High Frequency]
To stall for time while the bill is being revised, they interfere with the process of bringing it to
c. 
the floor. [High MD, Low Frequency]
Despite the ads about switching to green energy, they depend on contributions from the coal
d. 
industry. [High MD, High Frequency]

As discussed in Section 2.1, the distractors in the experiment are verbs which
contain multiple lexical roots, that is, compound verbs. The 56 compound verbs
are listed in Table 3. The 56 compound verbs were used to construct 56 dis-
tractor sentences, matching the semantic domains for the stimulus (bigram)
sentences.

Table 3. Compound verb distractors used in experiment. Asterisks indicate compounds

in which the first root can function as a verb, including cases in which this usage is not
primary, or in which this usage arises from homonymy.
hang-glide* test-drive* house-sit strongarm jumpstart* moonlight* bankroll*
jam-pack* bench press flyfish* mass-produce bearhug* blackmail babysit*
panhandle* timetravel* deepfry fine-tune* tie-dye* earmark play-act*
fireproof* bookmark* blowdry* mastermind* overhear copyright* proofread
sleepwalk* spoonfeed* sidestep freeload* leapfrog* fundraise* skyrocket
wallpaper cherry-pick windsurf brainstorm* zigzag* shoplift* badmouth
backtrack* handwrite* waterski* doublecheck* spotlight* globetrot forcefeed*
daydream safeguard hotwire underestimate dryclean* wisecrack freezedry*
 The evidence add ups 

2.2 Participants and experiment setup

The 56 stimulus sentences and 56 distractor sentences were recorded by a female
native speaker of English, who was instructed to read them aloud at a normal,
casual rate of speech. The sentence audio files were presented to participants in
random order.
Before the experiment began, participants were given six practice sentences to
get them accustomed to the echoing and substitution task. Three of the six prac-
tice sentences contain a compound main verb (pickpocket, spearhead, breakdance),
matching the distribution of patterns in the main experiment. Since participants
had to resist a tendency to make errors such as she breaksdance(s), the inclusion of
compound verbs among the practice items helped to introduce interference from
the alternate affix-insertion strategy as early as possible.
During the experiment, participants listened to the stimulus sentences on
headphones, and gave vocal responses into a digital microphone attached at
the collar. The main part of the experiment took approximately 20 minutes for
participants to complete. A self-timed break was allowed halfway through the
experiment.
Volunteer participants were recruited at the University of New Mexico from
the Introduction to the Study of Language course, and received a small amount
of course credit for participating. A total of 29 participants enrolled in the study.
Out of this group, 27 participants reported that English was their first language.
The remaining 2 participants reported that they learned English before the age
of 6, speak English fluently, and use it daily. No participants reported a history of
speech or hearing disorders.

2.3  R
 esults and discussion: Affix shifts and other affixation errors on
bigram stimuli
2.3.1 Participant accuracy
The experiment trial encompassed a total of 1624 responses (56 stimuli × 29 par-
ticipants). I listened to all participant responses and coded them as No Error,
Affix-Shifted, Double-Marked, Zero-Marked, or Unclassifiable. Responses with
‘No Error’ were those that correctly positioned the -s suffix (they arrive at > she
arrives at). I also included cases where the participant mistakenly converted the
sentence to the past tense, but correctly positioned a past tense suffix (they arrive
at > she arrived at). Unclassifiable responses could arise for several reasons, for
example, if the participant failed to remember the crucial part of the sentence,
forgot to change the subject pronoun, or produced an irregular past form (they
give in > she gave in).
 Clay Beckner

Experiment participants were highly attentive to the task, and were gener-
ally able to give sufficient responses in spite of the time pressure. Overall, 142
responses had to be rejected as Unclassifiable, meaning that 1483 participant
responses (91%) were codable. The rejection rate was relatively stable across the
four bins in the experiment (chi-square = 4.82, p = 0.185).
Additionally, participants’ responses were, as expected, overwhelmingly accu-
rate. On average, 90% of the responses (1462 of the 1624) were coded as having No
Error. The least accurate participants (2 participants out of the 29 total) provided
accurate responses 77% of the time, and were deemed to be sufficiently accurate to
be retained in the study.

2.3.2 Overview of errors: Affix shifts, double-marking errors, no-marking errors

Despite the strong tendency toward accurate responses, the methodology did
induce the speech errors of interest to this study. The data contain 7 full affix shifts;
one example appears in (2).
Since there’s not much else to do before it’s time to go, she settle downs on the
(2) 
couch…before it’s time to go.

Additionally, the data contain 9 ‘double-marked’ affix errors, in which a suffix

appears on the verb as well as on the following word. A typical example is pre-
sented in (3).
To reward the students for completing more tedious… assignments, she reads
(3) 
alouds . . . from children’s books an hour each day.

One of the 9 double-marked errors in this set is somewhat more complex, and
requires some discussion:
Despite the ads about switching to green energy, she depends onned
(4) 
([dəpɛnzɑnd]) contributions from this, from the coal industry.

I have classified sentence (4) as a double-marking error, albeit a double-marking

that involves two different suffixes. Apparently the speaker activated two distinct
suffixes (3rd person singular -s, and the past -ed), and a failure to resolve compet-
ing plans for the sentence caused one affix to be applied to the verb (depends),
and the second affix to be applied to the bigram (dependson-ed). Such an occur-
rence is not especially surprising, given overall patterns in responses. First, it was
relatively common for participants to convert sentences into the past tense, in
spite of reminders not to do so during the experiment’s instruction phase. Among
the bigram items, there were 14 responses that shifted to past tense. Addition-
ally, the compound distractors generated 16 past tense responses. Thus, it seems
that p­ articipants may have needed to suppress an ongoing temptation to insert
 The evidence add ups 

-ed rather than -s. Additionally, there were two other stimuli on which a speaker
attached two different affixes to the same verb (talksed about, shopliftsted). Here, it
is clear that the speaker appended -s and -ed in succession, indicating that speak-
ers at times retrieve two different affixes for the same verb. Moreover, in two com-
pound distractor sentences, a participant produced -s and -ed on different roots of
the verb (jams-packed, spooned-feeds).3 In sum, it seems that we can we can make
inferences from the positioning of affixes, even when the affixes in question reveal
other, competing mechanisms at work.
Combining the 7 affix shifts with the 9 double-marked errors, there were 16
errors of affix positioning in the experiment results. Collectively, then, there were
16 errors out of 1624 attempts, yielding an error rate of just under 1% (0.98%). In
other words, an affix positioning error occurred on 1 out of every 102 sentences
attempted in the experiment. This error rate indicates that the experiment meth-
odology is effective at eliciting affix positioning errors, given how rare these errors
are in spontaneous conversation. For comparison, note that Deese (1984: 130)
estimates that approximately one out of 100 sentences in conversation contains a
speech error of any kind; this estimated frequency naturally includes phonologi-
cal and lexical errors, which are far more numerous than morphosyntactic ones.
Similarly, the catalog of 191 slips from the 170,000-word London-Lund Corpus
indicates that speech errors occur approximately once every 890 words (Garnham
et al. 1981). Again, the London-Lund speech errors are predominantly phonologi-
cal and lexical in nature, and the Garnham et al. collection contains no errors of
affix positioning. Thus the rate at which errors of interest are observed in the pres-
ent experiment – equivalent to one affix positioning error approximately every
2,000 words – indeed represents a marked increase from baseline error rates in
conversation.
In addition to the 16 affix positioning errors, there were five no-marking errors
observed. As noted previously, no-marking responses are also consistent with unre-
lated phenomena; for instance, in some speech communities, the absence of a 3rd
person singular -s may be normative. However, review of the experiment data as a
whole indicates that none of the speakers omit 3rd person singular -s systematically,
which helps to justify considering no-marking responses as errors. In the current
dataset, all 5 no-marking responses were by different speakers; that is, no speaker
was responsible for more than one of these omissions on his or her 56 responses to

. More broadly, the distractor sentences prompted a total of 57 errors such as plays-act,
spoons-feed, and spooned-feeds, in which an affix interrupts a compound word. This error
pattern is distinct from the affix shifts under consideration in this chapter, and provides evi-
dence regarding the analyzability of compounds. See Beckner (2013) for additional details.
 Clay Beckner

the stimuli. Thus, in the analyses below, I will include the 5 no-marking errors along
with the 16 affix errors previously described, for a total of 21 affix errors.4

3. Data and analysis

With respect to the MD and Frequency bins in this study, the affix errors were not
symmetrically distributed. Since all four bins are represented by an equal num-
ber of stimuli, if the distribution were random we would expect the errors to be
spread evenly across the four categories. However, this was clearly not the case, as
is evident from examining the affix errors presented in Table 4. The most striking
feature of the distribution is an overwhelming tendency (20 out of 21) for errors
to occur on high MD bigrams. This result is as predicted. We may verify that the
effect of MD is statistically significant using a two-tailed Fisher Exact test, which
yields a highly significant result (p < 0.001). However, inspection of Table 4 also
reveals that most of the affix errors (15 out of 21) are low in frequency, counter to
predictions. In a Fisher Exact test, the effect of low frequency is statistically signifi-
cant (p = 0.025).
The foregoing findings may be verified with additional statistical tests. Logis-
tic regression is a natural choice for analysis of data in which a dependent variable
is binary (Jaeger 2008; Harrell 2013), as is the case for the occurrence or non-
occurrence of a speech error. Note, however, that error events are quite rare as an
outcome in the current dataset – “zeroes” vastly outnumber the “ones” – and a
traditional regression analysis may raise concerns about biased model coefficients
(King & Zeng 2001). As a precaution, then, I use an alternate approach to logis-
tic regression which is well-suited to rare events datasets, namely, the bias reduc-
tion method5 (Firth 1993; King & Zeng 2001: 148, n3). Bias reduction has been

. All the statistical analyses to be discussed in Section 3 have also been performed using a
more conservative dataset that excludes the no-marking responses as unanalyzable. Analyses
are qualitatively similar in either case, and none of the results of this chapter hinge on the
inclusion or exclusion of the no-marking errors.
. A number of alternate analyses are available, but detailed discussion is beyond the scope
of the current chapter. For instance, results are essentially the same if we use more familiar
maximum-likelihood regression methods on continuous variables, e.g., mixed-effects models
with random effects for participants and items. This approach in particular is useful because it
is striking that certain stimuli recur in the set of observed errors. For instance, 6 different par-
ticipants made an error on read aloud, and 3 on tear apart. These surprising recurrences raise
the possibility that additional factors are at work in making affix errors more likely. However,
 The evidence add ups 

Table 4. Distribution of affix shift errors, double-marked errors, and no-marking

errors collected in the shadowing task
Low frequency High frequency

HIGH MD 15 errors/371 codable responses 5 errors/382 codable responses

5 full affix shifts: 2 full affix shifts:
gain weights make sures
settle downs make sures
settle downs
read alouds
tear aparts
7 double-marked affix errors: 2 double-marked affix errors:
wraps ups wakes ups
reads alouds depends onned
reads alouds
reads alouds
reads alouds
tears aparts
tears aparts
3 no-marking errors: 1 no-marking errror:
she… read aloud she ca-… call it
she gain weight
she screw up
LOW MD 0 errors/365 codable responses 1 error/365 codable responses
------ 1 no-marking error
he conveniently forget it

i­mplemented in an iterative algorithm available as an R package, called logistf

(Heinze et al. 2013, Heinze & Schemper 2002; Heinze & Ploner 2003).
In this regression analysis, I analyze the independent variables (MD and
Token Frequency) as continuous predictors. The values of interest are based
on the COCA Spoken Corpus, and appear in Table 1. As is common practice,
raw frequency counts are log-transformed. The regression model also tested
the effects of presentation order on the occurrence of errors, both as a continu-
ous variable (trial number) and categorical variable (first or second half of the
experiment). Such predictors were not significant, and thus dropped from the
model. The regression analysis does, however, again verify that high MD is asso-
ciated with an increase in affix errors (p < 0.0001), with a positive regression
coefficient (β = 0.62). Low bigram frequency is associated with an increase in

in a mixed-effect model, item-based variation is directly incorporated, and the model results
are nearly identical to those presented here.
 Clay Beckner

such errors (p < 0.0001), with a negative coefficient (β = −0.70). To assess the
model’s predictive success, we may calculate the coefficient of concordance, C,
between model predictions and observed responses. The concordance index is
0.86, thus surpassing the expected threshold of 0.80, indicating an acceptable
model (­Harrell 2013).
Thus, statistical tests indicate that affix positioning errors are less likely to
occur on high-frequency bigrams. This finding is quite surprising in the context
of the predictions of this study. However, it is worth noting that the preponder-
ance of low-frequency bigrams among the errors arises from bigrams that are
also high in Mutual Dependency. No errors are observed when frequency is
low if MD is also low. This pattern hints at an interaction between independent
variables in the experimental task, rather than a general effect from low bigram
frequency.
We may expand our logistic regression model to investigate the possibility
of variable interactions. However, the model is not improved by incorporating
an interaction between Token Frequency and MD. In a logistf model including
a Frequency * MD interaction, this interaction is not significant (β = −0.004,
p = .85). In the current dataset, it seems that an interaction cannot be verified,
apparently due to the sparsity of the data, and/or due to collinearity between
variables.6
In any case, it is quite clear that affix errors are more likely to arise under
a particular confluence of variables (Low Frequency and High MD). We may
confirm this pattern by comparing the results of models in which the vari-
ables are included in isolation, as opposed to jointly as presented above. If we
model the data with MD as the sole variable (without including frequency as
a separate factor), the predictor is still significant (β =  0.14, p = 0.01). How-
ever, the model’s concordance index is unsatisfactory (C = 0.69), and log-
likelihood comparison indicates that the model is superior if both frequency
and MD are included (chi-square = 39.97, df = 1, p < .0001). Similarly, a
model excluding MD indicates that log frequency is significant (β = −0.26,
p = 0.003). However, the model concordance is unsatisfactory (C = 0.68), and
log-likelihood tests favor the model with both variables (chi-square = 37.79,
df = 1, p < .0001). Such results indicate that, even absent a demonstrable vari-
able interaction, the best description of the error distribution relies on both
MD and frequency of the bigram.

. Indeed, diagnostic tests indicate that the MD * Frequency interaction introduces collin-
earity; the Variance Inflation Factor for this model has scores as high as 65, which surpasses
even the most lenient benchmarks (~10).
 The evidence add ups 

3.1 Post hoc analysis: Examining components of the MD metric

In sum, the foregoing analyses demonstrate that higher Mutual Dependency of
a bigram is associated with increased likelihood of affix errors. As predicted, the
experiment indicates that high-MD sequences are more likely to be retrieved as
units, and/or less amenable to interruption with inflections in a time-pressured
task. However, these analyses also indicate that higher frequency is associated with
a decrease in affix errors. As noted, this pattern is observable only among bigrams
high in MD, but logistic regression models cannot directly verify variable inter-
actions. Moreover, the ‘backwards’ frequency effect is a counterintuitive finding
which is in need of further investigation.
To help make sense of the frequency data, in this reanalysis I consider the
possibility that confounding variables may in fact be driving the reversed effects
from frequency. Indeed, retracing the experiment design reveals that additional
frequency differences may be relevant. Specifically, frequencies of the component
words in each bigram are a noteworthy factor in the experiment, and asymmetries
are apparent if we focus on the items which are most prone to affix errors (Low
Frequency, High Mutual Dependency).
For instance, comparing the frequency of the verb in the bigrams in the four
stimulus categories reveals that there are rather striking category differences. The
average verb frequency (from the COCA Spoken Corpus; see Table 1) for items in
the Low Frequency, High MD bin is 3308; this contrasts markedly with averages
of 13395, 87953, and 30346 for the other three bins. Similarly, the second word of
each bigram is, on average, lower in the Low Frequency, High MD bin. The aver-
age second-word frequency for this bin is 224559; this is markedly lower than the
averages in the other three bins (572268, 688406, and 1119300).
All four bins contain a considerable range of verb frequencies, and there
are overlaps in values across all the bins, but there is a clear overall trend: the
items which are most prone to affix errors contain lower-frequency words.
Note that component-word frequencies were not left to vary in an uncon-
trolled way in the experiment design. Rather, the wide range in component-
word frequencies is a necessary part of the experiment’s design. The bigrams
in the Low Frequency/High MD bin are, in large part, categorized as such due
to the lower frequencies of their component words. To understand why this
is the case, note from Equation 1 that the MD value is calculated on the basis
of three frequency measures, involving the bigram, the bigram’s first word
(F(w1)), and the bigram’s second word (F(w2)). Items in the Low Frequency/
High MD bin are restricted to those having a relatively low bigram frequency
(otherwise, obviously they would be classified as ‘High Frequency’). Thus the
only way for these items to surpass the ‘High Mutual Dependency’ threshold,
 Clay Beckner

while maintaining low frequency for the overall sequence, is for the bigram to
consist of lower-frequency component words.
With respect to the experimental task, there are in fact intuitive reasons why
lower-frequency component words prompt more affix errors. First, frequent
verbs would be expected to be produced more accurately in their inflected form,
assuming the base + inflection may be retrieved as a unit or as a well-practiced
sequence. Thus, high-frequency inflected units7 may be characterized by easier,
error-free production (see Stemberger and MacWhinney 1986b). Secondly,
note that the experimental task requires recognition of the bigram’s first and
second words as a precursor to inserting an inflection. That is, the stimulus
sentence must be parsed into component words before inserting the -s; it is
reasonable to expect that lower-frequency words will be less readily identified
as independent units.
I thus reanalyze the experiment data, to incorporate component word
frequencies more directly. In this reanalysis, Mutual Dependency is not included
explicitly, but I include all the (logged) component frequencies that define
MD for a given bigram (see Equation 1): frequency of the bigram (F(w1w2)),
frequency of the first word F(w1), and frequency of the second word F(w2).
This selection of variables yields a regression model with significant effects
for all three frequencies. Moreover, the results are rather interesting: once
individual word frequencies are included explicitly in regressions, the apparent
backward effect from bigram frequency vanishes. In the model, higher token
frequency for the bigram (F(w1w2)) now results in an increase in affix errors
(β =  0.49, p = 0.007). This result obtains, apparently, because the reversed
effects associated with bigram frequency are already better accounted for by
the other variables in the model. As expected, the component-word frequencies
have negative regression coefficients, indicating that lower frequencies are
associated with higher error rates. For F(w1) (the verb), β = −0.44 (p = 0.002),
and for F(w2), β = −0.60 (p < 0.0001). The concordance index for this model is
good, with C = 0.87.
In sum, then, a reanalysis of the data incorporating component-word
frequencies yields expected effects for all three variables included in the model:
frequency of the bigram, frequency of the first word, and frequency of the second

. Along these lines, it is also sensible to analyze the frequency of the -s inflected verb, e.g.,
the corpus frequency of 3rd person singular forms like makes, gains, and settles, since the ease
of producing such forms should be related to their frequency. However, the frequencies of the
base and inflected form are linked, and parallel analyses of the inflected-form frequency yields
results similar to the ones discussed below.
 The evidence add ups 

word. Lower frequencies of individual words within each bigram are associated
with an increase in affix errors. Once these component frequencies are expressly
included in the model, we can see that higher bigram frequency is also associated
with an increase in affix errors. Note again that the current post hoc model includes
all the elements used in the definition of Mutual Dependency (or other relative
frequency measures) – incorporated directly as frequency counts, rather than as a
summary ratio. When seen in this light, it becomes apparent that the distribution
of errors is indeed as predicted by the theory: errors evidencing holistic processing
increase when the whole unit is more accessible, or when its component parts are
less accessible.

4. Discussion and conclusion

This chapter has investigated a new experimental methodology for the quanti-
tative study of multi-word units. The findings are encouraging, insofar as affix
shifts and double-marked inflections are induced on approximately one out of a
hundred attempts – orders of magnitude more frequent than what we observe in
casual speech. Moreover, this initial foray into studying affix positioning errors
gives promising evidence that bigrams are more prone to being retrieved as a unit
when the two words frequently co-occur.
First, high-MD bigrams are overwhelmingly more likely to prompt shift
errors that indicate the sequence is activated as a unit, including full affix shifts
(gain weights) and double-marked errors (wraps ups). These positive findings
are concordant with earlier studies by Ellis et al. (2008) and Ellis and Simpson-
Vlach (2009), which provide empirical support for the related relative frequency
measure, Mutual Information. Like Mutual Information, Mutual Dependency
provides a mathematical representation of competition between the activation of
whole units (the measure’s numerator) and the activation of component units (the
denominator). The current findings suggest that MD is a useful summary statistic
worthy of further investigation.
The findings with respect to bigram frequency turn out to be rather more
complicated. The experiment does not provide evidence that bigram frequency
alone leads to an increase in errors. Initial quantitative analyses indicate that,
contrary to expectations, errors are more likely on bigrams that are low in
frequency. Followup analyses suggest that the anomalous frequency pattern is in
part an artifact of the experiment design. In the experimental stimuli, frequency
of the bigram is confounded with frequencies of the component words, and these
component word frequencies turn out to be essential to the distribution of affix
positioning errors. Indeed, follow-up analysis indicates that h ­ igher-frequency
 Clay Beckner

bigrams are more likely to prompt affix errors – but this finding is only observable if
we take into account the effects of component word frequencies. In the experimental
task, bigrams containing infrequent words (an effect involving both the first and
second words) are more likely to result in affix errors indicative of holistic retrieval.
The importance of component-word frequencies in the experimental task lends
support to relative frequency accounts of processing, in which units containing
lower-frequency components are more likely to be processed holistically.
It is worth acknowledging that the affix errors induced under the present
shadowing methodology give a glimpse into the joint effects of comprehension
and production. The shadowing methodology requires participants to segment
speech into words, and almost immediately echo it back, while monitoring
continuously for the appropriate site to insert a verbal inflection. Since partici-
pants’ production occurs quickly after comprehension (within approximately 2
seconds), it is reasonable that segmentation errors might result in syntagmatic
production errors. Of course, speech comprehension (including segmentation)
is still relevant to the study of prefabricated units; indeed, such processes are
at the core of other experimental investigations of prefabs (e.g., Kapatsinski &
Radicke 2009). Nevertheless, future work in this area may benefit from a revised
methodology that focuses more exclusively on speech production, by allowing
participants to hear the target sentence in its entirety before repeating it back
from memory.
The view emerging from current research into language processing and change
is that – alongside more abstract grammatical generalizations – there must exist a
complex system that is influenced by the frequencies of various units (Bybee 2007,
2010). The current study provides evidence that the frequencies of complex units,
in addition to the frequencies of their component parts, are registered in cogni-
tion. It is reasonable to maintain that absolute and relative frequencies both have
effects on the processing and retrieval of multi-word sequences. Some of these
effects may be overlapping, given that relative frequencies can be said to arise from
the competition between absolute frequencies of different units. This study sup-
ports the insight (Bybee 2001, 2006) that language structure and language usage
are intertwined, and this interrelationship includes the gradual development of
new multi-word units.

References

Alegre, Maria & Gordon, Peter. 1999. Frequency effects and the representational status of regu-
lar inflections. Journal of Memory and Language 40: 41–61. doi: 10.1006/jmla.1998.2607
Arnon, Inbal & Snider, Neal. 2010. More than words: Frequency effects for multi-word phrases.
Journal of Memory and Language 62: 67–82. doi: 10.1016/j.jml.2009.09.005
 The evidence add ups 

Baars, Bernard J. 1980. The Competing Plans Hypothesis: An heuristic viewpoint on the causes
of errors in speech. In Temporal variables in speech: Studies in honour of Frieda Goldman-
Eisler, Hans W. Dechert & Manfred Raupach (eds). The Hague: Mouton.
doi: 10.1515/9783110816570.39
Baars, Bernard J. 1992. A dozen competing-plans techniques for inducing predictable slips in
speech and action. In Experimental Slips and Human Error: Exploring the Architecture of
Volition, Bernard J. Baars (ed.), 195–215. New York NY: Plenum Press.
doi: 10.1007/978-1-4899-1164-3_6
Beckner, Clay. 2013. Quantitative Determinants of Prefabs: A Corpus-based, Experimental
Study of Multiword Units in the Lexicon. PhD dissertation, University of New Mexico.
Beckner, Clay & Bybee, Joan L. 2009. A usage-based account of constituency and reanalysis.
Language Learning 59 (Supplement 1): 27–46. doi: 10.1111/j.1467-9922.2009.00534.x
Bolinger, Dwight. 1976. Meaning and memory. Forum Linguisticum 1(1): 1–14.
Bouma, Gerlof. 2009. Normalized (pointwise) mutual information in collocation extraction. In
From Form to Meaning: Processing Texts Automatically, Proceedings of the Biennial GSCL
Conference 2009, Christian Chiarcos, Richard Eckart de Castilho & Manfred Stede (eds),
31–40, Tübingen: Gunter Narr.
Bybee, Joan L. 2001. Phonology and Language Use. Cambridge: CUP.
doi: 10.1017/CBO9780511612886
Bybee, Joan L. 2002a. Phonological evidence for exemplar storage of multiword sequences. Stud-
ies in Second Language Acquisition, 24(02), 215–221. doi: 10.1017/S0272263102002061
Bybee, Joan L. 2002b. Sequentiality as the basis of constituent structure. In The Emergence of
Language out of Pre-language [Typological Studies in Language 53], T. Givón & Bertram
F. Malle (eds), 107–132. Amsterdam: John Benjamins. Reprinted in Bybee. 2007, 313–335.
doi: 10.1075/tsl.53.07byb
Bybee, Joan L. 2006. From usage to grammar: The mind’s response to repetition. Language 82:
711–733. doi: 10.1353/lan.2006.0186
Bybee, Joan L. 2007. Frequency of Use and the Organization of Language. Oxford: OUP.
doi: 10.1093/acprof:oso/9780195301571.001.0001
Bybee, Joan L. 2010. Language, Usage, and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan L. & McClelland, James L. 2005. Alternatives to the combinatorial paradigm of
linguistic theory based on domain general principles of human cognition. In The Role of
Linguistics in Cognitive Science, Nancy A. Ritter (ed.). Special issue of The Linguistic Review
22(2–4): 381–410.
Bybee, Joan L. & Scheibman, Joanne. 1999. The effect of usage on degree of constituency: The
reduction of don’t in English. Linguistics 37: 575–596. Reprinted in Bybee. 2007, 294–312.
doi: 10.1515/ling.37.4.575
Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspec-
tive. Oxford: OUP. doi: 10.1093/acprof:oso/9780198299554.001.0001
Davies, Mark. 2008–. The Corpus of Contemporary American English (COCA): 450 million words,
1990–present. <http://www.americancorpus.org>
Deese, James. 1984. Thought into Speech: The Psychology of Language. Englewood Cliffs NJ:
Prentice-Hall.
Ellis, Nick. 2002. Frequency effects in language processing: A review with implications
for theories of implicit and explicit language acquisition. Studies in Second Language
Acquisition 24: 143–188. doi: 10.1017/S0272263102002024
 Clay Beckner

Ellis, Nick C., Simpson-Vlach, Rita & Maynard, Carson. 2008. Formulaic language in native
and second language speakers: Psycholinguistics, corpus linguistics, and TESOL. TESOL
Quarterly 42(3): 375–396. doi: 10.1002/j.1545-7249.2008.tb00137.x
Ellis, Nick C. & Simpson-Vlach, Rita. 2009. Formulaic language in native speakers: Triangulat-
ing psycholinguistics, corpus linguistics, and education. Corpus Linguistics and Linguistic
Theory 5(1): 61–78. doi: 10.1515/CLLT.2009.003
Erman, Britt & Warren, Beatrice. 2000. The idiom principle and the open choice principle. Text
20(1): 29–62.
Firth, David. 1993. Bias reduction of maximum likelihood estimates. Biometrika 80: 27–38.
doi: 10.1093/biomet/80.1.27
Fontenelle, Thierry, Bruls, Walter, Thomas, Luc, Vanallemeersch, Tom & Jansen, Jacques. 1994.
Survey of collocation extraction tools. Technical report, University of Liege, DECIDE
MLAP-project 93–19.
Fromkin, Victoria A. 1973. The non-anomalous nature of anomalous utterances. Language
47(1): 27–52. Reprinted in Victoria Fromkin (ed.), Speech Errors as Linguistic Evidence,
215–242. Paris: Mouton & Co.
Garnham, Alan, Shillcock, Richard C., Brown, Gordon D. A., Mill, Andrew I. D. & Cutler, Anne.
1981. Slips of the tongue in the London-Lund corpus of spontaneous conversation. Linguis-
tics 19(7–8): 805–817.
Garrett, M. F. 1980. Levels of processing in sentence production. In Language Production, Vol. 1
Brian Butterworth (ed.), 177–220. New York NY: Academic Press.
Green, Lisa J. 2002. African American English: A Linguistic Introduction. Cambridge: CUP.
doi: 10.1017/CBO9780511800306
Gregory, Michelle, Raymond, William D., Bell, Alan, Fosler-Lussier, Eric & Jurafsky, Daniel.
1999. The effects of collocational strength and contextual predictability in lexical produc-
tion. Chicago Linguistic Society 35: 151–166.
Harrell, Jr., Frank E. 2013. Regression Modeling Strategies: With Applications to Linear Models,
Logistic Regression, and Survival Analysis. New York NY: Springer Science & Business
Media.
Haspelmath, Martin. 1993. The diachronic externalization of inflection. Linguistics 31: 279–309.
doi: 10.1515/ling.1993.31.2.279
Hay, Jennifer. 2001. Lexical frequency in morphology: Is everything relative? Linguistics 39(6):
1041–1070. doi: 10.1515/ling.2001.041
Hay, Jennifer & Baayen, R. Harald. 2005. Shifting paradigms: Gradient structure in morphology.
Trends in Cognitive Sciences 9(7): 342–348. doi: 10.1016/j.tics.2005.04.002
Heinze, Georg & Ploner, Meinhard. 2003. Fixing the nonconvergence bug in logistic regression
with SPLUS and SAS. Computer Methods and Programs in Biomedicine 71: 181–187.
doi: 10.1016/S0169-2607(02)00088-3
Heinze, Georg, Ploner, Meinhard, Dunkler, Daniela & Southworth, Harry. 2013. Logistf: Firth’s
bias reduced logistic regression. R package version 1.21.
Heinze, Georg & Schemper, Michael. 2002. A solution to the problem of separation in logistic
regression. Statistics in Medicine 21(16): 2409–2419. doi: 10.1002/sim.1047
Heinze, Georg, Ploner, Meinhard, Dunkler, Daniela & Southworth, Harry. 2013. Logistf: Firth’s
bias reduced logistic regression. R package version 1.21.
Hermer-Vasquez, Linda, Spelke, Elizabeth S. & Katsnelson, Alla S. 1999. Sources of flexibility in
human cognition: Dual-task studies of space and language. Cognitive Psychology 39: 3–36.
doi: 10.1006/cogp.1998.0713
 The evidence add ups 

Jaeger, T. Florian. 2008. Categorical data analysis: Away from ANOVAs (transformation or not)
and toward logit mixed models. Journal of Memory and Language 59: 434–446.
doi: 10.1016/j.jml.2007.11.007
Kapatsinski, Vsevolod & Radicke, Joshua. 2009. Frequency and the emergence of prefabs: Evi-
dence from monitoring. In Formulaic Language, Vol. II: Acquisition, Loss, Psychological
Reality, Functional Explanations [Typological Studies in Language 83], Roberta Corrigan,
Edith Moravcsik, Hamid Ouali & Kathleen Wheatley (eds), 499–520. Amsterdam: John
Benjamins. doi: 10.1075/tsl.83.14kap
King, Gary & Zeng, Langche. 2001. Logistic regression in rare events data. Political Analysis
9(2): 137–163. doi: 10.1093/oxfordjournals.pan.a004868
Klein, Krystal & Yu, Chen. 2009. Joint or conditional probability: Why decide? Paper presented
at COGSCI 2009: The Annual Meeting of the Cognitive Science Society, VU University,
Amsterdam, July 29–August 1.
Krug, Manfred. 2003. Frequency as a determinant in grammatical variation and change. In
Determinants of Grammatical Variation in English, Gunter Rodenburg & Britta Mondorf
(eds), 7–67. The Hague: Mouton de Gruyter. doi: 10.1515/9783110900019.7
Labov, William. 1969. Contraction, deletion, and inherent variability of the English copula.
­Language 45: 715–762. doi: 10.2307/412333
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, Vol. 1. Stanford CA: Stanford
University Press.
Levelt, Willem J. M. 1989. Speaking: From Intention to Articulation. Cambridge MA: The MIT
Press.
MacKay, Donald G. 1979. Lexical insertion, inflection, and derivation: Creative processes in
word production. Journal of Psycholinguistic Research 8(5): 477–498.
Manning, Christopher & Schütze, Hinrich. 1999. Foundations of Statistical Natural Language
Processing. Cambridge MA: The MIT Press.
Marslen-Wilson, W. D. 1973. Linguistic structure and speech shadowing at very short latencies.
Nature 244: 522–523. doi: 10.1038/244522a0
Pawley, Andrew & Hodgetts Syder, Frances. 1983. Two puzzles for linguistic theory: Native-
like selection and nativelike fluency. In Language and Communication, Jack C. Richards &
Richard W. Schmidt (eds), 191–226. New York NY: Longman.
Pinker, Steven. 1991. Rules of language. Science 253: 530–535. doi: 10.1126/science.1857983
Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A Concise Gram-
mar of Contemporary English. New York NY: Harcourt Brace Jovanovich.
Reali, Florencia & Christiansen, Morten. 2007. Word chunk frequencies affect the processing
of pronominal object-relative clauses. Quarterly Journal of Experimental Psychology 60(2):
161–170. doi: 10.1080/17470210600971469
Sinclair, John. 1991. Corpus, Concordance, Collocation. Oxford: OUP.
Stemberger, Joseph P. 1984. Structural errors in normal and agrammatic speech. Cognitive Neu-
ropsychology 1(4): 281–313. doi: 10.1080/02643298408252855
Stemberger, Joseph P. 1985. The Lexicon in a Model of Language Production [Garland Outstand-
ing Dissertations Series]. New York NY: Garland. (Ph.D. dissertation, University of Cali-
fornia, San Diego, 1982).
Stemberger, Joseph P. & MacWhinney, Brian. 1986a. Form-oriented errors in inflectional
processing. Cognitive Psychology 18: 329–354. doi: 10.1016/0010-0285(86)90003-4
Stemberger, Joseph P. & MacWhinney, Brian. 1986b. Frequency and the lexical storage of
regularly inflected forms. Memory and Cognition 14(1): 17–26. doi: 10.3758/BF03209225
 Clay Beckner

Thanopoulos, Aristomenis, Fakotakis, Nikos & Kokkinakis, George. 2002. Comparative evalua-
tion of collocation extraction metrics. In Proceedings of the 3rd Language Resources Evalu-
ation Conference, 620–625. Las Palmas, Spain.
Tremblay, Antoine, Derwing, Bruce, Libben, Gary, & Westbury, Chris. 2011. Processing advan-
tages of lexical bundles: Evidence from self-paced reading and sentence recall tasks.
­Language Learning 61(2): 569–613. doi: 10.1111/j.1467-9922.2010.00622.x
Vogel Sosa, Anna & MacFarlane, James. 2002. Evidence for frequency-based constituents in the
mental lexicon: Collocations involving the word of. Brain and Language 83: 227–236.
doi: 10.1016/S0093-934X(02)00032-9
Wray, Alison. 2002. Formulaic Language and the Lexicon. Cambridge: CUP.
doi: 10.1017/CBO9780511519772
Wray, Alison. 2006. Formulaic language. In Encyclopedia of Language and Linguistics, Vol. 4,
Keith Brown (ed.), 590–597. Oxford: Elsevier. doi: 10.1016/B0-08-044854-2/04777-5
Wray, Alison. 2008. Formulaic Language: Pushing the Boundaries. Oxford: OUP.
 look up about
Usage-based processes in lexicalization

Dawn Nordquist
University of New Mexico

This chapter investigates the emergence of a new English phrasal-prepositional

verb, look up about, which is found predominantly in online discourse and is
largely synonymous with “to google”. It is argued that the verb emerged as the
result of a reanalysis of a source syntagm that eroded an internal constituency
boundary and resulted in a new lexical item. In order to understand how
reanalysis took place in the lexicalization of look up about, usage-based processes
such as chunking and holistic and heuristic-based processing are examined,
and Bybee’s (2002) Linear Fusion Hypothesis is also invoked to explain how this
multi-word expression has entered the lexicon of many users.

1. Introduction

This chapter investigates the lexicalization of what appears to be a new ‘phrasal-

prepositional verb’1 in English: look up about, illustrated in (1):
“When I got home today, I looked up about Slash and I also looked up about
(1) 
your life.”

The above quote comes from a 2014 young adult epistolary novel entitled Love
Letters to the Dead by Ava Dellaira in which the teen protagonist writes letters to
famous individuals who are no longer living.2 The quote is instructive in that it
contains two instances of look up about as a phrasal-prepositional verb, and the
repetition of the three word string suggests that the expression is an acceptable

. See Section 2.1 below for a discussion of this term.

. The example in (1) appears on page 46 in a letter written to Janis Joplin in which the teen
protagonist reveals that she has just learned about a 1980s band Guns N Roses. She wants to
learn more about the lead guitarist of the band (Slash) and also wants to learn more about
Janis Joplin.

doi 10.1075/slcs.192.10nor
© 2018 John Benjamins Publishing Company
 Dawn Nordquist

usage and not a typo that went undetected in the copy editing stages of publica-
tion. However, although it appears from (1) that look up about might be a new
usage, at least for the author of the novel, the Oxford English Dictionary (OED)
does not contain an entry for the usage. Moreover, the string is not found among
the OED citations for the well-established and related phrasal verb look up nor
are there any annotations to reflect the possible usage (compare, for example, the
OED entry for find out,3 which contains the annotation “also with prep. about”
alongside ­relevant citation examples demonstrating the usage). Similarly, a search
of the Contemporary Corpus of American English (COCA, Davies 2008-) reveals
only five tokens of the sequence look up about, none of which represents an
instance of the phrasal-prepositional verb usage shown in (1) above. The absence
of the phrasal-prepositional verb in these databases suggests that its use is rela-
tively new and that it might be found more frequently outside the domains/genres
represented in COCA, which are identified as Fiction, Magazine, Newspaper, Aca-
demic, and Spoken. Indeed, the first attestation in this study’s data set occurs as
part of an internet forum discussion post, shown in (2):
…I need to look up about ghost possessing people. We all know that older
(2) 
students like to tease the younger ones. I mean, Flint does that all the time.
 (September, 1991)

The example in (2) comes from the online Pottermore site for Harry Potter fans,
where ghosts are a common topic, and is dated September, 1991. It is not until 2000
that another four tokens are found in a Google search for the string, but the phrasal-
prepositional verb’s use grows from there. In fact, as we will see in ­Section 3.0, use
of this verb has increased steadily since 1991, nearly tripling in use over the last few
years, especially in online contexts, suggesting that look up about represents a lexi-
cal innovation in English. Before looking at the data in more detail, though, we will
consider topics relevant to the emergence of multi-word verbs: previous analyses
of ‘phrasal-prepositional verbs’, relevant lexicalization research, and usage-based
mechanisms that support the emergence of new lexical forms.

2. Background

2.1 ‘Phrasal-Prepositional Verb’ interpretations

look up about contains two ‘P-words’, a term O’Dowd (1998) coined to label those
words (e.g., up and about) that function as prepositions and/or particles or that func-
tion gradiently. Given that P-words do not behave categorically, it is perhaps not a

. OED Online. (2014). find, v. Oxford University Press.

 look up about 

surprise that multi-word expressions containing P-words (e.g., ‘prepositional verbs’

or ‘phrasal prepositional verbs’) receive different structural interpretations in the
literature. Huddleston (2002: 273), for instance, in his discussion of verb + P-word
combinations, defines the ‘prepositional verb’ as a verb which specifies a preposi-
tional complement. For Huddleston, prepositional verbs are not unitary items, and
he argues that the separateness of V and P is borne out through various syntactic
tests (p. 277): PP coordination (She should refer to this article and to your previous
work);4 adverbial positioned between V and P (She referred frequently to the work of
Johnston); and, fronting of the PP (To which article did she refer?). Although Hud-
dleston (ibid.) recognizes the possibility of the prepositional passive (e.g., the article
is referred to on your site as a “Comprehensive testing report”),5 he does not address
the fact that the prepositional passive suggests that the P-word is syntactically gra-
dient between verbal particle and prepositional status.
Citing these same syntactic tests, Huddleston makes a similar argument for
verbs with two following P-words (i.e., ‘phrasal-prepositional verbs’), arguing
that these verbs take two prepositional complements (e.g., look forward to): “As
with prepositional verbs, we take the transitive preposition to belong syntactically
with the following NP, so that look has two PP complements, one consisting of
the intransitive preposition alone, the other of the transitive preposition together
with its object” (p. 287). In other words, I am looking [forward] [to my summer
vacation] contains a verb + intransitive preposition + transitive preposition.
Again, Huddleston does recognize the possibility of the prepositional passive, but
he restricts his discussion to highlighting the questionable acceptability for many
double prepositional complement-taking verbs to undergo passivization, and he
does not discuss the prepositional passive in terms of evidence of alternative struc-
tural boundaries for those verbs which do felicitously undergo passivization (e.g.,
“The star player, Kyle Lowry, was given up on twice by his first two NBA teams,
thought to be sullen and difficult and uncoachable, and really given up on a third
time by this Raptors team that had verbally traded his rights away.”)6
In contrast to Huddleston, Claridge (2000) identifies the prepositional verb as
a unit. She argues that V + P-word form a unit both syntactically and semantically.
Semantically, Claridge argues that the P-word forms a greater cohesive bond with
the preceding V than with the following NP. Synonym tests are offered as evidence of
semantic unity (e.g., refer to ~ cite), but single verb synonyms are not always ­available

. Examples are from Huddleston (2002) unless otherwise indicated.

. Posted by hightower99 on 11–27–2007, 03:29 AM at <http://www.ultimatecarpage.com/
forum/archive/index.php/t-2958-p-4.html> (1 December 2016).
. <http://news.nationalpost.com/sports/toronto-raptors-strength-comes-from-having-to-
fight-for-everything-theyve-got> (1 November 2016).
 Dawn Nordquist

and therefore perhaps not a reliable test of V + P-word semantic unity. Claridge
points out that there is nonetheless an idiomaticity found among V + P-word strings
(which encompass ‘prepositional verbs’) that underlies their cohesion. Indeed, the
collocational fixity of V + P-word in signaling non-compositional meaning is widely
recognized as a feature of such two word verbs (e.g., Biber et al. 1999: 404).
Commenting specifically on ‘phrasal-prepositional’ verbs, i.e., multi-word
verbs which seemingly contain a phrasal verb plus a preposition (e.g., face up to),
Claridge writes:

My approach to phrasal-prepositional verbs is much more semantically, or rath-

er idiomatically, determined than in the case of the preceding two categories
(phrasal verbs and prepositional verbs – DN). With three words co-occurring,
two of which are high-frequency functional items, one has to make especially
sure that they really form a lexical unit, and do not just happen to be used side
by side. After all, almost any phrasal verb can happen to be followed by a prepo-
sitional phrase (cf. Carstensen 1964: 318f; Palmer 1974: 238). Unitary mean-
ing seems to be the right way of establishing lexical-unit status in this case…If
one downplayed semantic considerations here, given the frequency of phrasal
verbs and the ubiquity of prepositions, one might end up with a large class of
phrasal-prepositional verbs, most of them having very little internal cohesion.
 (2000: 64–65)

Because ‘phrasal-prepositional’ verbs exhibit variability with respect to struc-

tural tests for identifying syntactic boundaries, and because grammatical tests
of word class are generally idiosyncratic with respect to their applicability and/
or criterial ranking (Croft 1991: p. 6ff; Hoffmann 2005: 33), idiomaticity, as
discussed above, may be one of the stronger arguments for claiming that look
up about is a new multi-word verb. While this study does not aim to close the
debate on the ultimate structure of these multi-word strings, based on the analy-
sis in Section 4.1, look up about will be viewed, in line with Claridge (2000), as
a semantically cohesive phrasal-prepositional lexical item, consisting of a verb
and two P-words.

2.2 Lexicalization
For the purposes of analyzing the lexicalization of look up about, the following
characterization will be adopted:

Lexicalization is the change whereby in certain linguistic contexts speakers use a

syntactic construction or word formation as a new contentful form with formal
and semantic properties that are not completely derivable or predictable from
the constituents of the construction of the word formation pattern. Over time,
there may be further loss of internal constituency and the item may become more
­lexical. (Brinton & Traugott [2005: 96])
 look up about 

According to the above quote, a key feature of lexicalization is whether or not a

form-meaning correspondence requires learning and memorial storage in the lex-
icon. In other words, lexicalization refers to those instances in which a form comes
to have new semantics, underivable from its individual constituents’ semantic
properties.7 As will be shown in Section 4.1, look up about represents such an
example of a new form-meaning pairing that has been adopted into the lexicon of
many English language users.
One of the advantages of this approach is that it does not restrict lexicaliza-
tion to a specific process that produces a new contentful form. However, many
researchers do define lexicalization in terms of the responsible mechanism. In fact,
depending on the theoretical framework adopted, lexicalization may refer to one
of the following categories (Brinton & Traugott 2005: 32):
1. Word formation processes (e.g., compounding, backformation, etc.);
2. Promotion of a previously bound form, leading to autonomy of the new form; or,
3. Syntactic fusion across independent forms with a resultant loss of composi-
tionality
With respect to the first category, the emergence of look up about cannot reliably
be attributed to productive English word formation processes. It does not appear
to be the result of compounding, for example. The second category is not involved
in the addition of look up about to the lexicon since the verb is clearly not the result
of an increase in autonomy of a previously bound form. Instead, it will be argued
that look up about is best analyzed within the third category as an instance of a syn-
tacticogenetic process: “when a phrase or a syntactically-determined lexical item
becomes a full-fledged lexical item in itself (Moreno Cabrera 1998: 214). Because
Brinton and Traugott (2005) focus on the result of, and not the mechanisms that
lead to lexicalization, their approach is broad enough to allow for this third research
focus, which can also include reanalysis or analogy as concomitant mechanisms in
the lexicalization process, allowing for the syntactic to become lexical.

. Grammaticalization shares much with the lexicalization process as described here.

However, while lexicalization involves a move from the already lexical or the phrasal into
the lexicon, grammaticalization moves the lexical into the grammar. Blank (2001: 1597),
citing and translating Lehmann (1989), sums up the difference as follows: “grammaticaliza-
tion transfers linguistic entities to grammar and thus changes a syntagm or word-form into
a rule of grammar (e.g., the grammaticalization of latin inf + habeo, habes, etc. as ‘future’),
whereas lexicalization brings linguistic entities into the lexicon and thus ‘individualizes’ and
‘deregulates’ them, e.g., by combining two morphemes”. (See also Wischer 2000: 364–365 for
a similar point of view.) While a continuum between lexical and grammatical is recognized,
it is believed that most readers will agree that the semantics of the new phrasal-prepositional
verb look up about places it on the lexical end of the continuum, as its semantics do not signal
traditional grammatical categories such as tense, aspect or modality.
 Dawn Nordquist

For example, idioms are the result of creating lexical items from syntactic
units. Particularly striking examples are those wherein “a whole utterance is trans-
formed into a more or less complex word expressing a contiguous concept” (Blank
2001: 1602 “delocutivity”). Italian non so che ‘ I don’t know’, for instance, has given
rise to nonsoche ‘something which is difficult to explain’ (Blank 2001: 1604; see,
too, the analogous French je-ne-sais-quoi). Here we find that the delocutive lexical
item has a meaning that is not strictly derivable from its component parts and is
no longer a syntactic unit consisting of freely combining component lexical items.
Blank (2001: 1604) comments that delocutive lexicalization is a marginal phenom-
enon. However, delocutivity is simply a specific instance of univerbation, a more
widely recognized result of fusing syntactic units into a lexical item. Brinton and
Traugott (2005: 48) write that “one of the commonest conceptions of lexicaliza-
tion is the unification, or univerbation of a syntactic phrase or construction into
a single word”. For example, nickname comes from Middle English an eke-name8
(“additional name”). In this case, the compound noun nickname originated in a
full noun phrase which fused phonetically and lost its semantic compositionality.
Not only has the erstwhile eke undergone phonetic reduction and fused with the
nasal final allomorph of the indefinite article, its independent meaning is no lon-
ger transparent; though nicknames are technically additional names, the lexical-
ized compound denotes a familiar, humorous or affectionate moniker and as such
warrants adoption into the lexicon.
Hagège (1993) also recognizes (quasi) sententially-derived complex words
and remarks on how they may or may not reflect the syntactic relations of the
original expression’s independent words. For instance, the Hungarian complex
word (CW) form in (3) does not retain a “-t marker of the definite object, although
in the complete sentence corresponding to this CW, such a marker is obligatory in
Hungarian” (Hagège 1993: 176).
(3) dolg-a-vegez-etlen
affair-poss-settle-priv.suff
‘without having settled his/her affairs’

An even starker example of a lexicalized counterpart losing its source syntagm’s

relational markings comes from Spanish metonymic nicknames such as el bigotes
(‘the one wearing a moustache’) or la rizos (‘the one having curly hair’). These
lexicalizations, according to Moreno Cabrera (1998: 215–216), are the reduced
versions of the syntactic phrases el hombre de los bigotes or la mujer de los rizos
whereby the loss of the head noun of these original phrases resulted in a ­lexicalized

. “nickname, n.” OED Online. Oxford University Press, June 2016. Web. 5 August 2016.
 look up about 

expression that does not faithfully reflect the original syntactic relations since the
article does not have the expected gender and/or number agreement: bigotes is
plural but in the lexicalized expression the plural noun collocates with a singular
definite article, and while rizos is masculine plural, it co-occurs with a feminine
singular article in the lexicalized la rizos.
These examples suggest that the above-mentioned lexicalized expressions may
have come from a syntactic string rather than resulting from a word formation
process, although historical data would be needed to confirm this assumption.
Look up about, though, is similar in that it could owe its provenance to a syntagm.
Specifically, it will be suggested that look up about could have emerged as a lexi-
cal item from a source construction which contains a major syntactic boundary
between the constituent words. It is further suggested that the string could have
been reanalyzed such that the major syntactic boundary was eroded. In order to
understand this proposed reanalysis of look up about, this study will invoke some
usage-based processes: chunking and holistic and heuristic-based processing.

2.3  Usage-based processes

2.3.1 Chunking
Chunking is a cognitively-general, memorial reflex whereby an experiencer
records recurrent, contiguous experiences as whole units in storage (Bybee 2010).
With respect to linguistic experiences, users will encounter repeated strings of
lexical and grammatical items. These recurring units are chunked and stored
whole (Bybee 2002b). For example, phrases such as salt and pepper or cause a
problem are semi-idiomatic expressions that are experienced as regular units and
are theorized to be stored holistically, as evidenced by the oddity of saying pepper
and salt or make a problem for expressing the same semantics of the established
expressions. Although, the individual words of larger chunks are independently
stored units themselves (see more in Section 2.3.2 below), it is the frequent recur-
rence of cause a problem (Ellis & Frey 2009), for example, which has led to its own
independent chunked storage.
Because conceptually-related items tend to co-occur in proximity to one
another in discourse, it typically follows that recurrent, semantically cohesive
strings will become chunked items that can be analyzed in terms of traditional
constituent structure (Bybee & Scheibman 1999: 593); recurrent salt and pepper
and cause a problem, for example, are a NP and a VP respectively. However, align-
ment with syntactic constituency is not a prerequisite for unit status. Chunking can
occur across traditional constituent boundaries as Bybee (2002b) and Krug (1998)
show for fused subject pronoun auxiliary strings (e.g., I’m), as Beckner and Bybee
(2009) and Hoffmann (2005) show for complex English prepositions (e.g., in spite
 Dawn Nordquist

of), and as Dąbrowska (2004) shows for German preposition and article fusion
(e.g, zu ‘to’ + der > zur). It is therefore the case that the same chunking mecha-
nism that produces traditional constituents also results in non-traditional units.
In fact, Bybee (2002b) argues for a Linear Fusion Hypothesis that states “sequen-
tial co-occurrence is more basic than constituency” (p. 120). Bybee explains that
“sequentiality is basic to language and that constituent structure emerges from
sequentiality because elements that frequently get used together bind together into
constituents” (p. 109). As we will see, chunking of the recurring sequence look up
about is yet another example of the Linear Fusion Hypothesis at work in which
sequential co-occurrence is more basic than constituency.

2.3.2 Holistic and shallow processing

The storage of chunked units and the parallel storage of individual constituents in
the usage-based lexicon is in opposition to a basic assumption within generative
frameworks that the lexicon, as a repository for non-productive, arbitrary units, is
redundancy-free. However, in the case of expressions such as cause a problem, the
chunk must be stored alongside its component words because there are no rules to
generate the conventionalized, semi-idiomatic form. Therefore, at a minimum, in
a usage-based model which espouses storage of chunked language, there are four
stored items (a, problem, cause, and cause a problem), and redundancy is directly
built into the lexicon. Given this redundancy in memorial storage, the question
naturally arises as to how the string cause a problem is processed: holistically or
algorithmically? Indeed, an implication of chunked storage is that processing can
proceed along lines other than strictly structural ones (Hay 2002; Sinclair 1991).
That is, speakers may access and produce stored chunks of linguistic material with-
out resorting to structural rules. A holistic type of language processing necessarily
implies that speakers are not always parsing the internal structure of the chunked
unit – an implication supported by production data (see, e.g., Bybee, 2002a for an
overview of the latter). Moreover, according to Dąbrowska (2004), “there is grow-
ing evidence that the processing involved in ordinary language comprehension
is, in fact, fairly shallow: that is to say, people may integrate the meanings of the
most salient lexical items, some easily accessible grammatical cues, and contextual
information to arrive at a semantic interpretation without fully parsing the sen-
tence” (p. 23). In other words, language users can process language holistically and
shallowly, so that they arrive at an interpretation that is “good enough”, relying on
approximate representations to parse the input (Sanford 2002). In fact, a variety of
experimental studies have shown that, when presented with non-canonical, com-
plex syntax, language users’ comprehension does not always conform to the fully-
specified structure identified via application of syntactic rules/algorithms (e.g.,
 look up about 

Christianson, Hollingworth, Halliwell & Ferreira 2001; Dąbrowska 1997; Ferreira

2003). Such studies imply that default processing may proceed along holistic and
heuristic-based mechanisms.
Because humans typically process language in relatively noisy contexts with
relatively slow brain processing power (Dąbrowska 2004), it is not surprising that
humans may apply heuristics to process complex linguistic input, only relying on
rule-based processing when the communicative task calls for it or when ample
time is available (Morford 2003). Given that heuristic processing may be the
default mechanism for linguistic decoding, especially for the processing of non-
canonical or less familiar syntax, and given that one of the relevant look up about
contexts represents a complex syntactic environment (see Section 4.2), it is con-
ceivable that heuristically-based syntactic processing may have played a role in the
emergence of look up about.

3. Data collection and method

Because look up about is a relatively new development in the English lexicon which
is favored in online genres, the data for this study come from internet language
retrieved using the Google search engine. Although there are methodological
problems with using Google as a corpus (Fletcher 2007; Leech 2007), Diemer
(2008) argues that “if the research is primarily qualitative and not quantitative, no
separate corpus matches the web for providing examples”, especially as the research
pertains to “. . . documentation of lexical or phraseological innovation”. For exam-
ple, Diemer (2008) shows how internet blogs contain innovative, semi-productive,
non-standard uses of prefixed verbs such as intake, oncome and forespeak (see, too,
Baayen 2003: 7–9; and Kuzmack 2010 for other online lexical innovations). There-
fore, given that the distribution of look up about is largely restricted to internet
contexts, where lexical innovation can often occur, it is a reasonable first step to
use Google to investigate the meaning and possible development paths of look up
about. Indeed, Google hits for the string look up about contain examples that point
to the range of constructions that the string appears in and therefore provide data
for the potential sources for the newly emergent lexical item.
The current method employed Google’s custom tools to conduct lemmatized
searches, by year, for look up about in quotations (e.g., “looked up about” for 1980–
1981). This searching process was repeated for the years 1980 to 2006. Because
no new usage categories were identified between 2000–2006, and because hand
coding was time intensive, years 2007 through 2014 were not searched. However,
in order to gauge more current usage, the first three months of 2015 were also
analyzed to include contemporary tokens of look up about.
 Dawn Nordquist

All returned Google hits for these time periods were coded for the seman-
tics of look up as marking either a ‘visual’ sense (e.g., Look up about 13 slides!
May, 2002), an ‘improvement’ sense (e.g., things began to look up about 10
years ago, February, 2002), or a ‘search’ sense; the ‘visual’ and ‘improvement’
senses are not included in this analysis because their semantics are not related
to the new lexical form’s semantics and are therefore less likely sources for the
new verb’s provenance. Additionally, ‘search’ senses of look up that collocated
with approximate “about” in a prepositional phrase were also excluded as a
possible source of the new phrasal-prepositional verb because of their relative
infrequency (e.g., I really only had to look up about two answers b/c the oth-
ers were easy…, January, 2015). The remaining tokens include two kinds of
strings of look up about: those that represent the new multi-word expression,
illustrated de novo in (4), and strings that represent the proposed source for
the new verb, illustrated in (5), where a syntactic boundary exists between
look up and about.
I just thought I would look up about this by chance and found there is
(4) 
actually a name for it and lots of other poor people are suffering from this. 
 (November, 2002)
The book covers just about everything you’d want to look up | about Excel
(5) 
 (July 2000)

Because examples like (5) are assumed to be a possible source construction for the
new verb, they were retained for analysis. I will use the term ‘extraposed source
construction’ when referring to instances like those in (5) (see Section 4.2 for more
discussion).
Table 1 shows the frequency of look up about strings that function as the new
phrasal-prepositional verb, starting from the first identified instance that appears
in 1991. Table 1 also indicates that look up about has been gaining traction in the
language, as evidenced by the fact that the token count in the first three months of
internet usage in 2015 is almost triple the number of tokens found in the entire set
of Google hits for the year of 2006.9

. The Wikipedia website was launched in 2001, and by 2007, “over 70% of the visits to
Wikipedia in the week ending March 17 came from search engines” (Rainie & Tancer 2007),
suggesting that a good portion of internet activity, at least at the time, was spent researching
information online. To the extent that posting to and reading Wikipedia pages represented a
new cultural activity, it is perhaps not a surprise that a new lexical item emerged at the same
time. In fact, tokens of look up about which denote online searching increase in frequency as
online searching activity increases.
 look up about 

Table 1. Frequency of look up about as a Phrasal-Prepositional Verb

1991 1 token
2000 4 tokens
2001 17 tokens
2002 20 tokens
2003 19 tokens
2004 45 tokens
2005 59 tokens
2006 87 tokens
2015 237 tokens
(01/01/2015 – 03/01/2015)

In the next section, we turn to an analysis of the new phrasal-prepositional verb.

First, we consider its meaning and argue for its lexical status. Second, we consider
how the proposed source construction, which contains a major syntactic bound-
ary, could have been reanalyzed as a multi-word lexical item.

4. Analysis

4.1 ‘Look up about’ as a lexical item

In Section 2.1, it was argued that semantic cohesion is critical to identifying multi-
word strings as having lexical status or not. In fact, Claridge (2000: 64–65) warns,
“almost any phrasal verb can happen to be followed by a prepositional phrase”, so
it must be semantics which guides us in identifying which of those strings func-
tion as lexical items as opposed to representing happenstance co-occurrences. In
the case of the string look up about, when it is acting as a chunked lexical item, it
expresses idiomatic, unitary meaning. As is evident in examples (6) and (7), look
up about is more or less substitutable with the converted and generalized verb “to
google”, meaning to (re)search topics on the internet.
(6) If you want to look up about the vintage Ibanez’s go
to……………….www.comcat.com/~alnico5:cool:  (August, 2001)
(7) I have to look up about the laser printer …
http://inventors.about.com/library/inventors/blanderson.htm

There’s information about the windshield wipers ….

Information about the fire escape is here:

http://www.answers.com/topic/1887
(February, 2002)
 Dawn Nordquist

Given that look up about appears most often in online forum discussion web
pages (for example, 81% of the 2015 tokens in Table 1 above were found on discus-
sion forum sites), it is perhaps not surprising that the multi-word expression regu-
larly appears in contexts related to information gathering that takes place through
internet searches. Other examples that invite an inference of online searching are
provided below. The underlined language in the examples in (8) either explicitly or
implicitly suggests that the act of “looking up about” is occurring online.
(8) a. I looked up about Robins on the internet  (February, 2002)
b.  i’ve been looking up about when ovulation happens, but most websites say
14 days before or after your cycle  (February, 2002)
c.  oh btw the link is to a hoax.. go look up about it outside of the thing you
posted, you will find it’s fake  (April 2002)
d.  I also looked up about the play and watched several videos of a few
different versions of the play.  (February 2015)

These examples show that look up about has an idiomatic meaning that is not
derivable from the sum of its parts since it is not predictable that the phrasal verb
look up, in its search sense, when collocated with the P-word about would result in
the meaning ‘to search on the internet’. As a result, it is prudent to assume that the
three words represent a new form-meaning pairing that is stored in the lexicon of
many English speakers. In the next section, we will address how the three words
may have come to be associated as a multi-word lexical item.

4.2 Development of Multi-Word ‘look up about’

Before turning to an analysis of how modern-day look up about might have
emerged, it is worthwhile to consider how multi-word phrasal verbs emerged
in English generally. According to Denison (1985), English prepositional verbs
and phrasal-prepositional verbs emerged between 1300 and 1700. Denison dis-
cusses a number of period-specific language change conditions that increased the
opportunity for V + P collocations.10 For instance, Denison points out that, as
Middle English syntax moved towards SVO word order and experienced a con-
comitant loss of nominal declensions, prepositional marking of the grammatical
status of post-verbal NPs increased. As prepositional phrase placement after the V
increased, multi-word verbs became a structural possibility, especially given that
P-word stranding was a stylistic discursive feature of Old English that continued
into Middle English. As a result, conditions were such that V + P-word sequences

. Denison (1985: 202–203) argues that the same conditions are observable for the ‘complex
prepositional verb’, his term for the ‘phrasal-prepositional’ verb.
 look up about 

were more frequently encountered than any given P NP string. This sequential-
ity favored the emergence of multi-word phrasal verbs since the P-word would
have co-occurred more often with the same V than a following NP (Bybee’s Linear
Fusion Hypothesis). Moreover, V + P-word strings were encountered in contexts
in which the NP was not always necessarily adjacent to the P-word (in so-called
‘stranding constructions’). In turn, especially within the appropriate discourse
context, the stranding of the P-word could have invited an inference that the V
and P-word structurally formed a unit. For example, Denison (1985) argues that
elliptical coordinate constructions created conditions that encouraged interpret-
ing the P-word as belonging with the V, as shown in (9).
(9) To lufe and lok on Þe
 ‘to love (you) and look on you’ 
 (a. 1450 Rolle, EDormio (CMB) 72.359 [Denison (1985: 191])

In (9) the object NP is shared across a coordinate verbal structure in which the
first V slot contains a transitive V and the second slot, a V + P-word combination.
In this context, Denison argues that it is possible that, as a hearer/reader processes
the coordinate structure, the NP is isolated in parsing its object status vis-à-vis the
first verb slot, and that such isolation encourages the NP to also be parsed as sepa-
rate from the P-word in the second slot. A commensurate analysis is that speakers,
defaulting to shallow processing, might have applied a [V and V N] heuristic in
these contexts. Such a processing scenario leaves the P-word “unparsed” and free
to be affiliated with the V, especially given the Linear Fusion Hypothesis, which
can result in new lexical items. Under either interpretation, the P-word would
have been interpreted to have a stronger affinity with the preceding V in repeated
V + P-word collocations.
A similar coordinate structure with ellipsis exists in the data for look up about,
shown in (10).
I frequently look up about and take advantage of free events in my city of
(10) 
Boston, like donation days at the art museum.  (January, 2015)

However, because (10) was produced in 2015, several decades after the first appear-
ance of look up about, (10) is probably better seen as a reflection of the already chun-
ked lexical item than as a mechanism for the verb’s emergence (although tokens
such as [10] can reinforce the mental lexical representation of look up about). It
therefore still remains to be explained how the new verb might have emerged.
While the specific grammatical changes that conditioned the emergence of
Middle English phrasal verbs are not at play in the case of Present Day English
look up about, the same cognitive processes that Denison discusses are still rele-
vant today, as they represent usage-based mechanisms that impact language users’
 Dawn Nordquist

l­ exicons. As a result, d
­ ifferent constructional contexts are involved in the develop-
ment of lexical look up about, but chunking and usage-based processing are argued
to still be relevant here. One such context that may have allowed for the emergence
of lexical look up about is the ‘extraposed source construction’ schematized in (11)
whereby a PP headed by about is extraposed after the phrasal verb look up. The
PP in (11) modifies an NP that is also modified by an intervening relative clause
containing the verb look up.11
(11) head npi [relative clause: np look up ___i ] [about np]extra-posed pp

This source schema is illustrated with an example in (12a), which represents one
of the five instances of the string look up about in COCA (Davies, 2008-) and con-
tains an extraposed PP, about pregnancy.
(12a) he remembered some of the things he’d looked up about pregnancy. 
 (FICT, 1997)

The alternate syntax without extraposition is given in (12b):

(12b) he remembered some of the things about pregnancy (that) he’d looked up

As (12b) illustrates, (12a) is not an instance of the new phrasal-prepositional verb

because, in (12a), it is clear that the P-word forms a structural relationship with
the following NP (pregnancy) and, as a traditional PP, modifies the noun things;
therefore, while (12a) contains the string look up about, it also contains a syntactic
boundary between the two adjacent P-words: look up | about.
In order to understand how this extraposition source context may have con-
tributed to the emergence of look up about as a phrasal-prepositional verb, it is
important to consider the features of the extraposed source construction schema-
tized in (11):

–– a PP headed by about postmodifies a nominal unit;

–– the postmodified nominal unit is often an indefinite pronoun or an otherwise
low content noun (65% of the ‘extraposed source construction’ tokens have
this feature);
–– the PP is in a long distance dependency with the postmodified nominal unit;
–– the PP is positioned after the phrasal verb look up; and,
–– the phrasal verb look up is contained in a subordinate clause (specifically an
object-extracted relative clause).

. The term extraposition is used here as a shorthand for describing the pattern that is sche-
matized in (11). The use of the term should not be taken to imply any theoretical assumptions
about storage and processing relationships between the two structures.
 look up about 

Example (12a) illustrates all of these properties: about pregnancy has a long
distance dependency with the nominal unit it modifies – the low content noun
things, and the PP follows directly after the phrasal verb look up which is con-
tained in an object-extracted relative clause. Crucially, though, the structure
schematized in (11) and illustrated in (12a) puts the three words in sequence and
therefore puts them in a position to be subject to the Linear Fusion Hypothesis,
their sequentiality laying the groundwork for the three words to be chunked as
a single unit.
It has been much reported that object-extracted relative clauses are difficult to
parse compared to subject-extracted relative clauses (although see Gordon, Hen-
drick & Johnson 2001). Given that shallow parsing may be a default form of syn-
tactic parsing for structural complexity (see Section 2.3.2), it seems reasonable to
assume that a listener or reader might be predisposed to parse the object-extracted
relative clause in (11) as a stand-alone clause, rather than perform the necessary
syntactic parsing required to arrive at a fully-specified representation. Shallow
parsing may also be encouraged in this context given that the relativized noun is
a low content noun (e.g., things). As Roland, Dick and Elman (2007: 357) discuss,
the semantic content for a noun such as thing can often be specified in a relative
clause; as a result, a language user’s experience may already predispose her to rely
on the relative clause itself for providing the relevant semantic content, making a
full parsing of the structure unnecessary. Therefore, the language user does not
necessarily fully integrate the relativized low content noun into the syntactic gap
of the relative clause. In such a scenario, the relative clause is perhaps akin to a
stand-alone clause, and the adjacent PP contains the semantics that modifies the
low content NP. The two constituents – relative clause and extraposed PP – may,
in this case, be processed heuristically, thereby obscuring the syntactic boundary
between up and about.
Evidence for this claim comes from recent comprehension studies of syntac-
tically complex clauses. For example, Christianson, Hollingworth, Halliwell &
­Ferreira (2001) report that, for the sentence While Anna bathed the baby played
in the crib, subjects correctly parsed the main clause as evidenced by their affir-
mative answers to the question Did the baby play in the crib? Yet, a majority of
subjects did not correctly parse the dependent clause, as evidenced by their affir-
mative answers to the follow-up question, Did Anna bathe the baby?12 Based on
subjects’ responses to this last question, it would appear that the stimulus sen-
tence was processed as two stand-alone clauses that both contain the NP, the

. Inversion of the clauses or introduction of a comma improved subjects’ performance on

the task.
 Dawn Nordquist

baby, ­ostensibly ignoring a major syntactic boundary. In light of these data, it

seems plausible that (12a) could have been parsed in a similar fashion. More to
the point, the relative clause and adjacent PP could potentially be pseudo-parsed
or scanned using a NVN pattern that maps onto a subject/agent-verb-object/
patient pattern (Townsend & Bever 2001). Indeed, Ferreira (2003) reports on a
series of experiments which shows that language users may comprehend experi-
mental stimuli in this way, such that Noun Verb Noun sequences are interpreted
via application of a general Agent Verb Patient semantic role pattern, irrespective
of the full syntactic parse. Therefore, if the relativized noun of the ‘extraposed
source construction’ is not semantically salient (e.g., things), and language users
are applying a NVN heuristic parsing strategy, it is possible that they may find
Subject Verb and Object in what could present as, more or less, a stand-alone
clause, shown in (13), with the original syntactic boundary unparsed, leading to
its possible erosion.
(13) N V N
he’d looked up about pregnancy

The unassigned P-word in (13) is now available to be integrated with the V.

Moreover, the shallowly processed version is not semantically incompatible
with the fully integrated and processed he’d looked up things about pregnancy
because the shallow version (he’d looked up about pregnancy) semantically
encompasses the fully-specified version. Therefore, failing to arrive at the fully-
specified syntactic structure does not prevent the language reader/listener from
comprehending the expression’s semantics (see Sandford & Sturt 2002 for a
discussion of how shallow processing does not limit language users’ compre-
hension in other areas of grammar). As a result, the role of the low content
NP in this process should probably not be underestimated. As stated earlier,
low content nouns13 regularly find their content in other clauses. In a sense,
the low content noun directs the language user to focus on the relative clause
or “downstream” syntax. Thus, while low content nouns can help structure the
discourse (Schmid 1999), they may contribute very little in terms of the seman-
tics of a sentence or utterance. As a result, low content nouns may not be fully
integrated into the processing of syntactic arguments if shallow processing is in
place where heuristics map over more contentful units to arrive at a parse that
is “good enough”.
In summary, the argument presented here suggests that shallow processing
of the extra-posed PP structure encourages the erosion of a syntactic boundary.

. Schmid (1999) uses the terms “shell noun”

 look up about 

Specifically, application of a NVN heuristic identifies a likely V and potential argu-

ment NPs, notably selecting the NP in a source PP structure as a possible object. In
turn, the “abandoned” P is available to be incorporated into a multi-word expres-
sion, further sanctioned by other multi-word verbs (e.g., find out about). There-
fore, in this extraposition context, erosion of the original syntactic boundary gives
way to a new verbal chunk. With enough repeat experiences of this reanalysis,
lexicalization can take place.

5. Conclusions

I have argued that a new multi-word verb has entered the English language,
especially in online domains, and offered an account for how the verb came
to be part of English language users’ lexicons. The argument presented here is
that the development of look up about is the result of usage-based processes
that have been observed and documented for other linguistic innovations. In
particular, I have suggested that look up about first appeared as a string in a
source construction that postposed a prepositional phrase headed by about after
the phrasal verb look up. Once the sequence was established in the ‘extraposed
source construction’, a new lexical item could emerge based on the cognitive
chunking ability of humans, the driving force behind constituency (Bybee 2010).
It was further argued that the analysis is especially motivated in light of the
fact that language users may not necessarily fully parse syntactic input but may
process it more shallowly, unless the communicative context requires otherwise.
Online discourse, where the bulk of the look up about tokens appear, is written,
for instance, and may encourage shallower processing since a user can always go
back and re-read the text for clarification. In any case, shallow processing means
that long distance dependencies (such as the postposed PP headed by about)
may not need to be fully resolved for communication to take place, especially
when the original PP is often modifying a low content noun. In this context, it
seems likely that a reanalysis could take place. The emergent multi-word verb is
further sanctioned by the fact that the lexicon contains other similar phrasal-
prepositional verbs (e.g., find out about) and associated schemas that look up
about can map onto (Bybee 2006). As a result, it is perhaps not a surprise that we
find examples like the one shown in (14a).

(14a) Why do you want the numbers in F anyway? It does not relate to
anything else, as any comparisons or information you try to look up
about about the temperatures you experience will all be in C. 
 (December, 2011)
 Dawn Nordquist

The discussion board poster in this case has slotted the phrasal-prepositional
verb look up about into the very source construction schematized in (11), leading
to a “double” about usage:14

head npi [relative clause: np look up ___i ] [about np]extra-posed pp

information [you try to look up about] [about the temperatures you
experience]

Compare (14a) with the alternate syntax in (14b), an invented example to high-
light the structure in (14a) where a new phrasal-prepositional verb is apparently
used.

(14b) Why do you want the numbers in F anyway? It does not relate to anything
else, as any comparisons or information about the temperatures you
experience that you try to look up about will all be in C.

Although we cannot be sure of the verb’s provenance, with examples like (14a),
we can be fairly certain of its new status in the language. Furthermore, exam-
ples such as (14a) strongly suggest that look up about could be the result of a
syntacticogenetic process in which a recurrent string of words was reanalyzed
as a multi-word verb, eroding a syntactic boundary in the process. As with all
levels of structure, lexical items’ emergence, too, is sensitive to usage-based
pressures.

Acknowledgements

I would like to thank two anonymous reviewers for their helpful input. Any errors or omissions
remain my sole responsibility.

References

Baayen, Harald. 2003. Probabilistic approaches to morphology. In Probabilistic Linguistics, Rens

Bod, Jennifer Hay & Stephanie Jannedy (eds), 229–287. Cambridge MA: The MIT Press.
Beckner, Clayton & Bybee, Joan L. 2009. A usage-based account of constituency and reanalysis.
Language Learning 59 (Supplement 1): 27–46. doi: 10.1111/j.1467-9922.2009.00534.x

. The example in (14a) could be the result of a typo – the writer could have mistak-
enly written about twice. However, if look up about has developed into a multi-word verb, as
argued for here, “double” about examples such as (14a) would be expected to occur.
 look up about 

Biber, Douglas, Johansson, Stig, Conrad, Susan, Leech, Geoffrey & Finegan, Edward. 1999.
Longman Grammar of Spoken and Written English. Harlow: Pearson Education.
Blank, Andreas. 2001. Pathways of lexicalization. In Language Typology and Language Univer-
sals, Vol. II [Handbücher zur Sprach- und Kommunikationswissenschaft 20.2], Martin
Haspelmath, Ekkehard König, Wulf Oesterreicher & Wolfgang Raible (eds), 1596–1608.
Berlin: Walter de Gruyter.
Brinton, Laurel J. & Traugott, Elizabeth Closs. 2005. Lexicalization and Language Change.
­Cambridge: CUP. doi: 10.1017/CBO9780511615962
Bybee, Joan L. 2010. Language, Usage and Cognition. Cambridge: CUP.
doi: 10.1017/CBO9780511750526
Bybee, Joan L. 2006. From usage to grammar: The mind’s response to repetition. Language
82(4): 711–733. doi: 10.1353/lan.2006.0186
Bybee, Joan L. 2002a. Phonological evidence for exemplar storage of multiword sequences.
­Studies in Second Language Acquisition 4: 215–221.
Bybee, Joan L. 2002b. Sequentiality as the basis of constituent structure. In The Evolution of
­Language out of Pre-language [Typological Studies in Language 53], T. Givón & Bertram
Malle (eds), 107–132. Amsterdam: John Benjamins. doi: 10.1075/tsl.53.07byb
Bybee, Joan L. & Scheibman, Joanne. 1999. The effect of usage on degree of constituency: The
reduction of don’t in English. Linguistics 37: 575–596. doi: 10.1515/ling.37.4.575
Christianson, Kiel, Hollingworth, Andrew, Halliwell, John F. & Ferreira, Fernanda. 2001.
­Thematic roles assigned along the garden path linger. Cognitive Psychology 42: 368–407.
doi: 10.1006/cogp.2001.0752
Claridge, Claudia. 2000. Multi-word Verbs in Early Modern English: A Corpus-based Study.
Amsterdam: Rodopi.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago IL: University of
Chicago Press.
Dąbrowska, Eva. 2004. Language, mind, and brain: Some psychological and neurological
­constraints on theories of grammar. Washington, D.C: Georgetown University Press.
Dąbrowska, Eva. 1997. The LAD goes to school: A cautionary tale for nativists. Linguistics 35:
735–766. doi: 10.1515/ling.1997.35.4.735
Davies, Mark. 2008– The Corpus of Contemporary American English: 450 million words, 1990–
present. <http://corpus.byu.edu/coca/>
Dellaira, Ava. 2014. Love Letters to the Dead. New York NY: Farrar Straus and Giroux.
Denison, David. 1985. Why Old English had no prepositional passive. English Studies 66(3):
189–204. doi: 10.1080/00138388508598384
Diemer, S. 2008. Corpus linguistics with Google? Paper presented at International Society for
the Linguistics of English, Boston.
Ellis, Nick C. & Frey, Eric. 2009. The psycholinguistic reality of collocation and semantic ­prosody
(2): Affective priming. In Formulaic Language, Vol. 2: Acquisition, Loss, Psychological Real-
ity, and Functional Explanations [Typological Studies in Language 83], Roberta Corrigan,
Edith Moravcsik, Hamid Ouali & Kathleen Wheatley (eds), 473–497. ­Amsterdam: John
Benjamins. doi: 10.1075/tsl.83.13ell
Ferreira, Fernanda. 2003. The misinterpretation of noncanonical sentences. Cognitive Psychol-
ogy 47: 164–203. doi: 10.1016/S0010-0285(03)00005-7
Fletcher, William H. 2007. Concordancing the web: Promise and problems, tools and tech-
niques. In Corpus Linguistics and the Web, Marianne Hundt, Nadja Nesselhauf & Carolin
Biewer (eds), 25–46. Amsterdam: Rodopi. doi: 10.1163/9789401203791_004
 Dawn Nordquist

Gordon, Peter C., Hendrick, Randall & Johnson, Marcus. 2001. Memory interference during
language processing. Journal of Experimental Psychology: Learning, Memory, and Cognition
27(6): 1411–23.
Hagège, Claude. 1993. Language Builder: An Essay on the Human Signature in Linguistic
­Morphogenesis. Amsterdam: John Benjamins. doi: 10.1075/cilt.94
Hay, Jennifer. 2002. From speech perception to morphology: Affix ordering revisited. Language
78(3): 527–555. doi: 10.1353/lan.2002.0159
Hoffmann, Sebastian. 2005. Grammaticalization and English Complex Prepositions: A Corpus-
based Study. London: Routledge.
Huddleston, Rodney. 2002. The clause: Complements. In The Cambridge Grammar of the English
Language, Rodney Huddleston & Geoffrey K. Pullum (eds), 213–322. Cambridge: CUP.
Krug, Manfred. 1998. Gotta – the tenth central modal in English? Social, stylistic and regional
variation in the British National Corpus as evidence of ongoing grammaticalization. In
The Major Varieties of English, Hans Lindquist, Staffan Klintborg, Magnus Levin & Maria
Estling (eds), 177–191. Växjö: Edman & Westerlunds Tryckeri AB.
Kuzmack, Stefanie. 2010. How medium shapes language development: The emergence of quota-
tive re online. In Studies in the History of the English Language, V: Variation and Change in
English Grammar and Lexicon: Contemporary Approaches, Robert A. Cloutier, Anne-Marie
Hamilton-Brehm & William A. Kretzschmar Jr. (eds) 293–310. Berlin: De Gruyter Mouton.
doi: 10.1515/9783110220339.2.293
Leech, Geoffrey. 2007. New resources, or just better old ones? The Holy Grail of representative-
ness. In Corpus Linguistics and the Web, Marianne Hundt, Nadja Nesselhauf & Carolin
Biewer (eds), 133–150. Amsterdam: Rodopi. doi: 10.1163/9789401203791_009
Lehmann, Christian. 1989. Grammatikalisierung und Lexikalisierung. Zeitschrift für Phonetik,
Sprachwissenschaft und Kommunikationsforschung 42:11–19. (Cited in Blank 2001).
Moreno Cabrera, Juan C. 1998. On the nature of grammaticalization and lexicalization. In The
Limits of Grammaticalization [Typological Studies in Language 37], Anna Giacalone Ramat
& Paul J. Hopper (eds), 211–227. Amsterdam: John Benjamins. doi: 10.1075/tsl.37.10mor
Morford, Jill P. 2003. Grammatical development in adolescent first-language learners. ­Linguistics
41(4): 681–721. doi: 10.1515/ling.2003.022
O’Dowd, Elizabeth M. 1998. Prepositions and Particles in English: A Discourse-functional
Account. Oxford: OUP.
Rainie, Lee & Tancer, Bill. 2007. Pew Internet and American Life Project. Data Memo, April.
<http://www.pewinternet.org/files/old-media/Files/Reports/2007/PIP_Wikipedia07.pdf.
pdf>
Roland, Douglas, Dick, Frederic & Elman, Jeffrey L. 2007. Frequency of basic English grammati-
cal structures: A corpus analysis. Journal of Memory and Language 57: 348–379.
doi: 10.1016/j.jml.2007.03.002
Sanford, Anthony J. 2002. Context, attention and depth of processing during interpretation.
Mind & Language 17(1–2): 188–206. doi: 10.1111/1468-0017.00195
Sanford, Anthony J. & Sturt, Patrick. 2002. Depth of processing in language comprehension: Not
noticing the evidence. Trends in Cognitive Sciences 6(9): 382–386.
doi: 10.1016/S1364-6613(02)01958-7
Schmid, Hans-Jörg. 1999. Cognitive effects of shell nouns. In Discourse Studies in Cognitive
­Linguistics [Current Issues in Linguistic Theory 176], Karen van Hoek, Andrej A. Kibrik
& Leo Noordman (eds), 111–132. Amsterdam: John Benjamins. doi: 10.1075/cilt.176.09sch
 look up about 

Sinclair, John McH. 1991. Corpus, Concordance, Collocation. Oxford: OUP.

Townsend, David J. & Bever, Thomas G. 2001. Sentence Comprehension: The Integration of Rules
and Habits. Cambridge MA: The MIT Press.
Wischer, Ilse. 2000. Grammaticalization versus lexicalization – “methinks” there is some confu-
sion. In Pathways of Change: Grammaticalization in English [Studies in Language Compan-
ion Series 53], Olga Fischer, Anette Rosenbach & Dieter Stein (eds), 355–370. Amsterdam:
John Benjamins. doi: 10.1075/slcs.53.17wis
 About the authors

Edith Bavin is an honorary professor in Psychology at La Trobe University in

­Melbourne, and an honorary researcher at the Murdoch Childrens Research
Institute (MCRI). She has published widely- some on East African languages but
mainly on child language, including the acquisition of Warlpiri, an indigenous
language of central Australia, typical and atypical language development, and lan-
guage processing. She is a member of the Australian Institute for Aboriginal and
Torres Strait Islander Studies (AITSIS) and a Fellow of the Association for Psy-
chological Science (APS). She served as editor of the Journal of Child Language,
the official journal of the International Association for the Study of Child Lan-
guage (IASCL) for six years and remains on the Editorial Board, and edited the
­Cambridge Handbook of Child Language.
Soteria Svorou has a BA in English Language and Literature from the University
of Athens, Greece, and MA and Ph.D degrees in Linguistics from the University
at Buffalo. She teaches syntax, semantics and historical linguistics at San José State
University where she is an Associate Professor. Working within functionalist and
cognitive approaches to language, she is the author of The Grammar of Space (John
Benjamins) and has published works on grammaticalization of spatial grams and
relational constructions. She is currently working on a usage-based approach to
conjunctive coordination constructions with a special focus on Modern Greek.
Terry Janzen is Professor and Head of the Department of Linguistics at the U
­ niversity
of Manitoba, Canada. His doctoral thesis, from the University of New Mexico, is
entitled Topicality in ASL: Information Ordering, Constituent Structure, and the
Function of Topic Marking. His research interests include the morpho-syntax of
verb structures in American Sign Language and in particular, perspective-­marking
in ASL clauses, discourse structure, grammaticalization in signed languages, and
intersubjectivity in interpreted interactions and interpreters’ use of language.
Rena Torres Cacoullos earned her degree in Spanish linguistics from the Uni-
versity of New Mexico. She is Professor of Spanish and Linguistics at Penn State
University and serves as editor of Language Variation and Change.
Earl K. Brown, PhD, works as an Associate Professor of Linguistics in the Depart-
ment of Linguistics and English Language at Brigham Young University, Provo,
Utah. His research centers on the quantification of language variation, especially
 Functionalist and Usage-based Approaches to the Study of Language

in Spanish. In most of this research, he employs the phonetics software Praat and
uses the corpus linguistics techniques of text searching and data manipulation
with the programming languages Python and R.
Esther L. Brown is an Associate Professor of Hispanic Linguistics in the Depart-
ment of Spanish and Portuguese at the University of Colorado Boulder. Her
research centers on the study of phonological variation from a usage-based per-
spective. Her work establishes significant correlations between variation and use;
specifically between phonological reduction and variables such as word frequency
and discourse context frequencies. This line of research has implications for
theories that attempt to explain the mechanisms behind language variation and
change, as well as for theories of lexical representation that attempt to account for
such patterns of variability.
K. Aaron Smith is professor of linguistics in the Department of English at Illinois
State University in Normal, Illinois, USA, where he teaches general linguistics, his-
tory of English and Latin. His scholarly interests are language change, functionalist
linguistic theory, history of English, verbal morphosyntax and comparative English
and Dutch grammar. His latest publications treat Ælric’s 10th century Latin-Old
English grammar within a history of linguistic thought and a chapter on teaching
about indigenous contact languages in the colonialization period of the history of
English (with Susan M. Kim) to appear in an MLA volume on the teaching of the
history of English. His book (co-authored with Susan M. Kim), This Language a
River: A History of English, was released in December 2017 by Broadview Press.
Damián Vergara Wilson is an Associate Professor in the Department of Spanish
and Portuguese at the University of New Mexico where he also coordinates the
Spanish as a Heritage Language program. His work focuses on applying notions
of usage-based analysis to sociolinguistic inquiries on bilingualism. He also pub-
lishes on Spanish as a Heritage Language.
Clay Beckner is a postdoctoral research fellow at the New Zealand Institute of
Language, Brain and Behaviour (NZILBB). His research interests include usage-
based linguistics, psycholinguistics, sociolinguistics, and morphosyntax.
Dawn Nordquist, PhD, is a Lecturer in Linguistics at the University of New
Mexico. She teaches a variety of courses, to support the cognitive, functional and
usage-based approaches of the department. Dawn also teaches in the Freshman
Learning Community program and was a UNM 2016–2017 Teaching Fellow. Her
research has focused on verbal collocational patterns and the role that colloca-
tions play in memorial schematic representation. She has also turned her attention
to narrative, both as a study in its own right and as an instructional resource for
teaching linguistics.
 Index

/d/ xxv, 127–148 coordination 17, 19, 21–22, H

/s/ xxv, 107–122 26–28, 39, 44, 48, 227 Hindi 3–6, 12
corpus linguistics 181 holistic processing xxvi, 30,
A 199–202, 219–220, 231–233
absolutive 3, 5–6, 10–12 D (see also shallow processing)
acquisition xvii–xxiv, 1–13 demonstrative xxiv, 89–92,
African American 94, 96, 102–103 I
English xxv, 155–156, diachronic xvii, xix–xxi, imperfective aspect 3, 5, 7,
158–160 xxvi, 25, 47, 49, 129, 156, 20–21, 23, 27–30, 42, 47–48
ain’t xxv, 155–170 159–160, 165, 175–177, 191,
analogy 2, 10, 177, 229 200–201 L
analogical extension 175–177, discourse marker 59–60, language change xviii, xix,
190, 192, 194 67–68, 70–72, 75–77, 83 xxii, 62, 236
argument 2–12, 34–35, 37, 41, duration xxv, 18, 24, 35, 44–45, lexical frequency xxv, 107–108,
68, 240–241 49, 59, 107, 111–121, 127, 147 110, 114–115, 118, 120–122
articulatory reduction lexical representation xix,
108–109, 128 E 128–130, 146, 237
ASL xxiv, 59–84 ellipsis 2, 6–7, 9, 237 Linear Fusion
aspect xx–xxi, xxiv–xxv, 3, episodic traces 127, 129–131, Hypothesis xxvi, 232,
5–6, 10–12, 17, 19–23, 134, 148 237, 239
26–30, 42, 45, 47–49, ergative xxiv, 1, 3–13, 90 linguistic complexity 168
157, 166–170 exemplar model 128, 146, 157,
auxiliary verb xxv, 31, 37–38, 169, 179 M
155–158, 161, 167–170, 231 Mayan 3–4, 10, 13
F mental lexicon xix, xxi, xxv,
C frequency xxii, xxiv–xxvi, 2, 49, 127–132, 134, 146–148,
case 2–10, 12–13, 18 5, 33, 41, 83, 92–93, 107–122, 199–200, 229–230, 232–233,
categorization 2, 175, 177, 128–131, 133–136, 140–141, 236, 238, 241
179, 181 143–144, 146, 148, 156, Mexican Spanish xxv, 107–122
center of gravity xxv, 107, 159–160, 163–166, 169–170, mixed effects linear
111–115, 118–121 175, 177, 179–191, 195, regression 107, 115–116, 118,
chunking xxii, 39, 175, 199–200, 202–210, 213–220, 120
199–201, 206, 231–232, 235, 228, 234–235 multi-word expression xxvi,
237–239, 241 199–207, 219–220, 226–228,
conjunctive coordination 17, G 234–237, 241–242
21, 26, 28 gender xxv, 5, 95, 97, 99–103, mutual information 203, 219
constructionalization xxiv, 110, 113, 118, 120, 133, 231
18, 25, 49 gradience xxi, xxv, 76, P
contact point 60, 66, 69, 75, 82 111–112, 116, 121, 145, 175–177, perfective aspect 3–6, 12,
contextual ratio 179, 181–182, 184, 192, 195, 20–21, 27–30, 42, 47–48, 62
frequency xxv, 107–122 199, 202, 226–227 phonetic reduction xxiv,
conventionalization xxv, grammaticalization xx, xxi, 62, 199, 230 (see also
xxvi, 19, 175–196 xxiv, 60–62, 64, 70, 76, 83, articulatory reduction and
conventionalization 89–103, 156–157, 229 sound reduction)
index 177, 181–186, 188–189, grammaticization xx prefab xxvi, 178, 182–183, 191,
191, 193, 195 Greek xxiv, 17–49 196, 199–203, 220
 Functionalist and Usage-based Approaches to the Study of Language

pseudo-coordination 19, source construction xxvi, 91, topic-comment 59, 61–64, 68

21–22 102, 231, 233–234, 235, 238, topicality xxv, 63, 89, 94–95,
240–242 97–98, 101, 103
R Southern-US English 159, 163
reference point 59, 63, Spanish xxiv, xxv, 89–103, U
79–80, 83 107–122, 127–148, 164, underdetermination 169
relative frequency xxvi, 183, 175–195, 230 undershoot xxiv, 60, 66, 75,
203–204, 219–220 speech errors xxvi, 199–206, 79, 83, 128, 130
reanalysis 229, 231, 241 211–220
retention xxiv, 91, 102–103, structural trace 31–39 V
109, 177, 190 subject pronoun xxiv, 68–69, variation 69, 74–76, 78,
79, 89–103 80–81, 83, 89–93, 96, 102–103,
S syntactic integration 25 109–110, 113, 127–130, 132–133,
semantic frames 41, 48 syntacticogenetic process 229, 136, 138–140, 143, 145–148,
shallow processing 232, 237, 242 161, 176–178, 190, 195, 207, 215
239–241 (see also holistic verbs of ‘becoming’ xxv,
processing) T 175–196
sociolinguistic identity xxv, third person/3rd Voiced stops 131–132, 145
155, 157 person xxiv–xxvi, 6–7, 11,
sound change 109–110 40, 49, 84, 89–90, 135, 202, W
sound reduction 107, 205, 212–213, 218 Warlpiri 4, 6–9, 12
121–122 (see also articulatory topic marking xxiv, 60,
reduction and phonetic 62–64, 76–77, 80, 83–84 Y
reduction) topic negotiation 72, 74, 79, 83 yes/no question 64–65, 71–72
The contributions to this volume honor Joan Bybee’s 2005 LSA Presidential
address “Grammar is Usage and Usage is Grammar,” as a cumulative
articulation of Professor Bybee’s long and inluential career in linguistics.
The volume begins with a functional examination of child language
acquisition of ergative languages. The next three contributions successively
investigate the grammaticalization of Greek postural verbs, Spanish third
person pronouns, and American Sign Language topicalization constructions.
The two following papers report on usage-based phonological studies of
Spanish /s/ and /d/, respectively. The book concludes with four papers that
address usage-based efects concerning the grammatical status of ain’t in
African American English, Spanish verbs of “becoming”, and English lexis
and prefabs. This volume will be of interest to a wide audience of functional
and cognitive linguistic researchers.

     

   