given-before-new: the effects of discourse on …
Post on 09-Jun-2022
1 Views
Preview:
TRANSCRIPT
GIVEN-BEFORE-NEW:
THE EFFECTS OF DISCOURSE ON ARGUMENT STRUCTURE IN
EARLY CHILD LANGUAGE
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF LINGUISTICS
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Nola M. Stephens
May 2010
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/ns375tr4692
© 2010 by Nola Marie Stephens. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Eve Clark, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Herbert Clark
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Beth Levin
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Thomas Wasow
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
iv
Abstract
Child language researchers, particularly those studying the acquisition of syntax,
have often viewed young children’s utterances as isolated and self-contained units
of analysis. But “language does not exist in a vacuum” (Clark & Clark 1978:227),
and utterances don’t either. This dissertation explores the influence of conversational
context on early word order. Specifically, I consider how discourse givenness affects
the order of postverbal arguments in the speech of preschool children.
In three elicited production studies, I systematically varied the structure of the
discourse children heard just before they were asked to describe a filmed vignette.
Study 1 targeted verbs of locative transfer, both alternating locative verbs (cf. She
squirted the hotdog with the ketchup. vs. She squirted the ketchup on the hotdog.) and
non-alternating ones (cf. She filled the cup with sand. vs. *She filled sand into the
cup.). Studies 2 and 3 targeted alternating dative verbs (cf. She gave the man the
hat. vs. She gave the hat to the man.). These studies provide converging evidence
that (i) givenness has a robust effect on early argument ordering—like adults, children
tend to use given-before-new ordering, (ii) this discourse effect can be largely, but not
fully, attributed to the effect of discourse on referring expressions (viz. that given
arguments tend to be pronominal and new ones tend to be lexical), (iii) givenness
does not influence all verbs and all arguments equally. I argue that several factors are
needed to explain the asymmetrical effects of givenness across verbs and arguments.
v
These include patterns of distribution in the input, conceptual biases, and semantic
and pragmatic properties of the verbs.
I also evaluate several mechanisms that might drive early given-before-new order-
ing: those that are addressee-based, speaker-based, or experience-based. My data do
not decide between these, but they do offer preliminary evidence in favor of a speaker-
based account. Ultimately, my dissertation highlights the importance of approaching
syntactic acquisition from several directions simultaneously. Children must learn to
attend to cues from form, function, and discourse and use their limited processing ca-
pacities to integrate these cues into a larger model of language production. Linguists
must do likewise.
vi
Acknowledgments
Many thanks first to my committee. To Eve Clark, my advisor: Five years ago, you
called me in my dorm room in Indiana and said, “I hope this one works.” I’m so glad it
did, and that’s thanks in large part to you. You have shaped my thoughts and shared
so many of your own with me. For all the wonderful guidance, advice, encouragement,
warmth, and laughs (both with me and at me), thank you! To Beth Levin: Thank
you for the many invaluable insights, for carefully reading and rereading essentially
everything I’ve written, for keeping me on track and well-read, for your book on verbs,
for always filling the water-boiler, and for much more! To Herb Clark: Thank you
for all the thought-provoking comments, and especially for demonstrating to me how
someone can be fabulously kind and intellectually terrifying at the same time. Just
thinking about you keeps me on my toes. And to Tom Wasow: So many thanks for
stepping in at the last minute and for all of your help and kindness from the start.
I also owe a great debt of gratitude to Joan Bresnan, who has taught, challenged,
helped, and supported me all along the way. In meetings with you, I felt like my
brain was running the gauntlet. My brain and my thesis both are better off for
having been near you. And interacting with you has always been deeply encouraging
to me. Thank you for that. Warm thanks also to Peter Sells for his part in guiding
me through my first few years here. And thanks, Dan Jurafsky, for all the friendly
hellos and for purchasing the water-boiler.
vii
Victor Kuperman: You were God’s gift to me. At my point of greatest despair and
least confidence, you came along—exactly the person I needed, my statistical hero,
and my friend. Thank you for the countless hours you gave so freely and patiently.
For sharing your skills, your advice, your food, your liquor, your stories, and your
very dear family, thank you! May the blessing return to you and Katya and Boris
and Joseph many times over!
I’m thankful also for the staff and children at Bing. Thank you for working with
me and playing with me! You made my research delightful. Special thanks to little
“Bat Junior” (Batman’s son)—for the priceless reminder not to worry, when I was, in
fact, worried, and for doing such a beautiful job of “protecting” me. You were right,
of course: “All [you] had to do was stand there, and they all just fell down dead.”
And thanks to Aya Inamori, whose help transcribing and coding was solid gold.
And to all my friends and colleagues and even a couple of strangers who starred in
my “filmed vignettes.” And to my friend May Lin Au Yong who filmed and did most
of the editing for my first vignettes and who taught me how to survive the rest.
Many thanks also to the linguistics (and psychology) graduate students and post-
docs who have been here with me. A big thank you to my cohort—Anubha Kothari,
Olga Dmitrieva, and Stacy Lewis. We’ve heard that seven is better than four; but
three’s company, and what excellent company you’ve been! Special thanks also to
Anubha for sitting next to me year after year—it’s been grand having you there. And
thanks to Andrew Koontz-Garboden, Bruno Estigarribia, Lauren Hall-Lew, and Inbal
Arnon who made the path before me better. And to Marie-Catherine de Marneffe,
Uriel Cohen Priva, Scott (Oscar) Grimm, Matt Adams, Adriana Weisleder, Seung
Kyung Kim, Middy Pineda, and Chigusa Kurumada, Patricia Amaral, and Susan
Bobb for your help and friendship.
viii
Thank you also to my professors, mentors, and friends at Indiana University who
got me to Stanford and have cheered me on ever since. Extra thanks to Natusko
Tsujimura, Dan Dinnsen, and Rania Rizk.
And thank you to my friends at IVGrad (and nearby). I adore you. And I’ll miss
you even more than I’ll miss the weather. It seems unfair that I’ve had such a wealth
of incredible friends here. You’ve carried me though and amazed me on a daily basis—
it really would have been worth it just to spend the last five years with you. Special
thanks to a few (roughly in order of appearance): Rachel (bc, ag, gfm), Lena and
Zach, Eun-Mee, Alissa and Todd, Diane Schouten, Jacob Mattingley, Olivia, Kristyn
and Jonathan, May Lin, Nani-Nani, Xianne, Sadie, Nate and Amy, Ashley Wellman,
Neth, Kassahun, Rachel Gong, Heather, Kierstin, Liz Milner, Susan, and the Wu.
And above all, to Pete and Maria: From the beginning to the end, you have been
my pastors, counselors, therapists, teachers, co-conspirators, and my friends. You are
Stanford to me.
My very best thanks go to my family (on both sides of the Red River). Thank
you for your unconditional and self-evident love, for your wisdom, for all the best
memories, and for every prayer. I’m especially thankful to my parents—all of them—
and to my brother, Branson. Mom and Barry and Branson, you raised me to believe
that learning is about as important as breathing. I can’t imagine how any of this
would have been possible without you—it certainly wouldn’t have been as much fun.
Dad, I’ve always been inspired by your unwavering confidence in me. Thank you also
for every loving reminder to enjoy my days here. (I’m so happy you found someone
to enjoy yours with along the way!) Mutti and Du, perhaps everybody needs another
set of parents. I did, and I got the best set out there. Thank you!
And one last thing before I thank the NSF: Mom, for the war you waged so that
I would finally learn to read, thank you. Let’s call this document a tribute to that
effort!
ix
My dissertation is based in part upon work supported by the National Science
Foundation under Grant Number IIS-0624345 to Stanford University for the research
project “The Dynamics of Probabilistic Grammar” (PI Joan Bresnan). The opinions,
findings, and conclusions expressed here are my own and do not necessarily reflect
the views of the National Science Foundation.
x
Contents
Abstract v
Acknowledgments vii
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The alternations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Semantic properties . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Syntactic properties . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Discourse properties . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Acquisition of argument structure . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Locative alternation . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Dative alternation . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Acquisition of information structure . . . . . . . . . . . . . . . . . . . 16
1.4.1 Referring expressions . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.2 Word order . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5 Accounting for given-before-new ordering . . . . . . . . . . . . . . . . 34
1.5.1 Addressee-based accounts . . . . . . . . . . . . . . . . . . . . 34
1.5.2 Speaker-based accounts . . . . . . . . . . . . . . . . . . . . . . 37
1.5.3 Statistical-learning accounts . . . . . . . . . . . . . . . . . . . 38
xi
1.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2 Study 1: Locative alternation 42
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Study 1A: Child locatives . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3 Study 1B: Adult locatives . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3 Study 2: Dative alternation 81
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2 Study 2A: Child datives . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3 Study 2B: Adult datives . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.3.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4 Study 3: Dative alternation revisited 125
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
xii
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5 Conclusion 161
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.2 Do children use given-first or new-first ordering? . . . . . . . . . . . . 162
5.3 Are all verbs equally affected by givenness? . . . . . . . . . . . . . . . 164
5.4 Are all arguments equally affected by givenness? . . . . . . . . . . . . 165
5.5 Are the effects due to referring expressions? . . . . . . . . . . . . . . 166
5.6 Why do children use given-before-new ordering? . . . . . . . . . . . . 170
5.7 Adding new information to the discourse . . . . . . . . . . . . . . . . 173
xiii
List of Tables
1.1 CHILDES: Age of first use of double-object datives and prepositional
datives in previous corpus studies. . . . . . . . . . . . . . . . . . . . . 15
1.2 Bates 1976: Summary of Claudio and Francesco’s early SO and OS
productions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1 Study 1: List of locatums and locations. . . . . . . . . . . . . . . 45
2.2 Study 1A: Results of model predicting locatum-object constructions. 52
2.3 Study 1A: Proportion of responses with locatum-objects. . . . . . . 53
2.4 CHILDES: Distribution of verbs in parental speech to 7 children. . . 64
2.5 APW: Distribution of verbs in the English Gigaword Third Edition
corpus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.1 Study 2A: Distribution of responses with one postverbal argument. . 90
3.2 CHILDES: Distribution of verbs with both postverbal arguments in
parental speech to 7 children. . . . . . . . . . . . . . . . . . . . . . . 101
3.3 Study 2B: Distribution of responses with one postverbal argument. . 108
3.4 Study 2B: Results of model predicting prepositional datives in re-
sponses with two postverbal arguments. . . . . . . . . . . . . . . . . . 110
3.5 Study 2B: Results of model predicting prepositional datives in re-
sponses with two postverbal arguments and lexical themes. . . . . . 116
3.6 Study 2B: Distribution of responses with two postverbal arguments. . 118
xiv
4.1 Study 3: List of verbs and events. . . . . . . . . . . . . . . . . . . . . 130
4.2 Study 3: Patterns of verb use and substitutions. . . . . . . . . . . . . 135
4.3 Study 3: Distribution of responses with one postverbal argument. . . 137
4.4 Study 3: Proportion of responses with one vs. two postverbal arguments.137
4.5 Study 3: Proportion of prepositional datives in responses with two
postverbal arguments. . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.6 Study 3: Results of model predicting prepositional datives in responses
with two postverbal arguments. . . . . . . . . . . . . . . . . . . . . . 139
4.7 Study 3: Results of model predicting prepositional datives in responses
with two postverbal arguments and lexical themes. . . . . . . . . . . 147
4.8 Study 3: Proportion of prepositional datives in responses with two
lexical postverbal arguments. . . . . . . . . . . . . . . . . . . . . . . 148
4.9 Study 3: Frequency and length of pauses in responses with two postver-
bal arguments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.10 Study 3: Distribution of disfluencies in responses with two postverbal
arguments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.11 Study 3: Results of model predicting the presence of disfluencies. . . 153
xv
List of Figures
2.1 Study 1A: Proportion of locatum-object constructions by age. . . . 53
2.2 Study 1A: Proportion of locatum-object constructions by verb. . . . 54
2.3 Study 1A: Distribution of referring expression types across given vs.
new locatums (above) and locations (below). . . . . . . . . . . . 56
2.4 Study 1A: Proportion of locatum- vs. location-object construc-
tions for each type of locatum (above) and location (below). . . . 59
2.5 Study 1A: Proportion of locatum-object constructions in responses
with lexical postverbal arguments. . . . . . . . . . . . . . . . . . . . . 62
2.6 Study 1B: Proportion of locatum-object constructions in responses
with two postverbal arguments. . . . . . . . . . . . . . . . . . . . . . 72
2.7 Study 1B: Distribution of referring expression types across given vs.
new locatums (above) and locations (below). . . . . . . . . . . . 75
3.1 Study 2A: Proportion of prepositional datives. . . . . . . . . . . . . . 91
3.2 Study 2A: Distribution of information structure patterns. . . . . . . . 93
3.3 Study 2A: Distribution of referring expression types across given vs.
new themes (above) and recipients (below). . . . . . . . . . . . . 95
3.4 Study 2A: Proportion of prepositional (theme-first) vs. double-object
datives (recipient-first) for each type of theme (above) and recip-
ient (below). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
xvi
3.5 Study 2B: Proportion of prepositional datives in responses with two
postverbal arguments. . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.6 Study 2B: Distribution of information structure patterns. . . . . . . . 111
3.7 Study 2B: Distribution of referring expression types across given vs.
new themes (above) and recipients (below). . . . . . . . . . . . . 113
3.8 Study 2B: Proportion of prepositional (theme-first) vs. double-object
datives (recipient-first) for each type of theme (above) and recip-
ient (below). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.1 Study 3: Proportion of prepositional datives. . . . . . . . . . . . . . . 140
4.2 Study 3: Distribution of information structure patterns. . . . . . . . . 141
4.3 Study 3: Distribution of referring expression types across given vs. new
themes (above) and recipients (below). . . . . . . . . . . . . . . . 143
4.4 Study 3: Proportion of prepositional (theme-first) vs. double-object
datives (recipient-first) for each type of theme (above) and recip-
ient (below). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
xvii
xviii
Chapter 1
Introduction
1.1 Introduction
As children acquire language, they must learn to harmonize a large and complex
collection of knowledge. Children need to learn not only what is grammatical and
ungrammatical, but also which linguistic expressions to use in a given context. This
dissertation is primarily concerned with how children employ linguistic context in
deciding how to order words in a sentence. Consider the sentences in (1).
(1) a. She gave the hat to the man.
b. She gave the man the hat.
Children acquiring English need to learn that both of these sentences are grammatical
and that both can describe the same event. But they also need to learn how to
choose between these two orderings in production. Previous research has shown
that adult construction choice reflects a tendency to order given information (i.e.,
information previously mentioned in the discourse) before new information when this
option is available in the grammar (e.g., Firbas 1964; Halliday 1967a; Bock & Irwin
1980; Prince 1981; Gundel 1988; Levelt 1989; Birner & Ward 1998; Arnold et al.
1
2 CHAPTER 1. INTRODUCTION
2000; Ward et al. 2002; Wasow 2002; Ferreira & Yoshita 2003; Krifka 2004; Bresnan
& Nikitina 2009). There is little consensus, however, about whether child speech
reflects this same tendency, and many questions remain regarding how and when
children package information in a way that coheres well with prior discourse.
The primary goals of this dissertation are to (1) assess how discourse status (viz.
givenness) affects word order in the speech of English-speaking preschool children, (2)
consider which mechanisms underlie discourse effects (e.g., communicative pressures,
processing pressures, or distributional learning), and (3) explore how discourse effects
interact with other influences on syntactic choices (e.g., frequency, semantic proper-
ties, types of referring expressions). These research questions lie at the intersection
of information structure and argument structure and require attention to both. In
Section 1.2, I introduce the syntactic phenomena under investigation (locative and
dative alternations) and motivate these choices. Section 1.3 discusses a few influential
ideas about how children acquire verb argument structure and reviews previous work
on the acquisition of locative and dative alternations. Section 1.4 then turns to the
acquisition of information structure, and Section 1.5 outlines three potential expla-
nations for given-before-new ordering. Finally, Section 1.6 provides a brief preview
of the ensuing chapters.
1.2 The alternations
The English locative and dative alternations form the empirical focus of this research.
Verbs that participate in these alternations take three arguments, an agent-subject
and two postverbal arguments, and the alternation takes place within the verb phrase.
Locative alternation verbs allow alternative ordering of locatum and location
1.2. THE ALTERNATIONS 3
arguments, and dative alternation verbs allow alternative ordering of theme and
recipient arguments:1
(2) Locative alternation
a. She squirted the ketchup on the hotdog.
(locatum-first: locatum-object variant)
b. She squirted the hotdog with the ketchup.
(location-first: location-object variant)
(3) Dative alternation
a. She gave the hat to the man. (theme-first: prepositional dative variant)
b. She gave the man the hat. (recipient-first: double-object dative variant)
Importantly, not all locative and dative verbs alternate, and of the non-alternating
verbs, some take one ordering and some the other (see Levin 1993). Among loca-
tive verbs, non-alternating verbs (e.g., put, pour) are more frequent in child-directed
speech (CDS) than alternating verbs (e.g., stuff, squirt). Conversely, the most fre-
quent dative verbs in CDS (e.g., give, show) allow alternation, and non-alternating
dative verbs (e.g., donate, recommend, explain) are very rare in CDS. So for alter-
nating verbs, children need to learn that two constructions are acceptable and the
conditions under which one is preferred over the other. For non-alternating verbs,
they must learn that only one construction is acceptable (regardless of the context)
and which structure that is. The learning problem is then further complicated by the
fact that children receive more input for certain types of verbs than for others.1Other labels have been given to these arguments. The locatum (object transferred) is alterna-
tively referred to as the figure or content, and the location (reference point) as the ground,
container, or goal. For datives, grammatical labels are often used: The theme is called the
“direct object” and the recipient the “indirect object.” Other semantic labels for the recipient
are goal and possessor.
4 CHAPTER 1. INTRODUCTION
Beyond the complexity of the learning problem, the following factors render these
two alternations particularly well-suited to present purposes: The two word order
variants (i) can generally be used to describe the same event, (ii) have very similar or
identical surface syntactic structures, and (iii) allow for an investigation of discourse
effects beyond the subject position. These three advantages are discussed below.
1.2.1 Semantic properties
Although the two variants of the locative and dative alternations are not identical in
meaning, both can generally be used to describe the same event (see Clark 1990; Fisher
et al. 1994). Subtle meaning differences between the variants have been explored at
length in the semantics literature (Green 1974; Oehrle 1976; Rappaport & Levin
1988; Gropen et al. 1989; Pinker 1989; Levin 1993; Krifka 2004; Rappaport Hovav &
Levin 2008, and references therein). Consider again the locative alternation in (2),
which is repeated below:
(4) a. She squirted the ketchup on the hotdog. (locatum-object variant)
b. She squirted the hotdog with the ketchup. (location-object variant)
While both variants of the locative alternation describe a transfer event, the location-
object construction tends to highlight the endstate of the location (e.g., full),
whereas the locatum-object construction is more likely to highlight the manner
in which the locatum is transferred (e.g., by squirting) (see Gropen et al. 1991b).
Meaning differences between the two variants have also been characterized in
terms of object affectedness (see Givón 1984; Gropen 1989; Pinker 1989; Gropen et al.
1991b; Arad 2006). Specifically, the location is understood to be the more affected
entity in the location-object construction, and the locatum is understood to be
the more affected entity in the locatum-object construction. This reasoning has
1.2. THE ALTERNATIONS 5
been used to explain the intuition that location-object constructions have a “holis-
tic” reading, and locatum-object constructions a “partitive” reading (S.R. Anderson
1971; Schwartz-Norman 1976; J.M. Anderson 1977; Jeffries & Willis 1984; Lumsden
1994; see Goldberg 1995:§7.5 for discussion). In other words, using the location as
the direct object suggests that the location is fully affected by the event (e.g., fully
covered with ketchup). When the locatum is the direct object and, therefore, inter-
preted as the more affected entity, the location can (but need not) be interpreted
as only partially affected.
Similar ideas have been proposed for the dative alternation. Like the locative
alternation, datives express a kind of transfer event (sometimes broadly construed).
Prepositional datives (theme-first) as in (5a) are associated with a cause to go mean-
ing, while double-object datives (recipient-first) as in (5b) are associated with a
cause to possess meaning (e.g., Goldberg 1995; Viau 2006; but see Rappaport Hovav
& Levin 2008).
(5) a. She gave the hat to the man. (prepositional dative variant)
b. She gave the man the hat. (double-object dative variant)
This difference can also be attributed to object affectedness (see Gropen et al. 1989;
M. Anderson 2006): Cause to go emphasizes the displacement of the theme, and
cause to possess emphasizes the endstate of the recipient. Some researchers have
also pointed out subtle meaning differences with certain dative verbs (see Green 1974).
For example, the double-object dative in (6a) suggests that Molly caught the ball (i.e.,
was affected by the throwing event/became a possessor), whereas the prepositional
dative in (6b), need not yield the same interpretation (cf. Rappaport Hovav & Levin
2008:§5). Similarly, as Green (1974:158) argued, (7a) “implies that Fido perceived
the bone, while [(7b)] leaves this open.”
(6) a. Sam threw Molly the ball.
6 CHAPTER 1. INTRODUCTION
b. Sam threw the ball to Molly.
(7) a. John showed Fido a bone. (Green 1974:158 (6a))
b. John showed a bone to Fido. (Green 1974:158 (6b))
Importantly, though, this interpretation of successful transfer with double-object da-
tives is defeasible (Oehrle 1977; see also Rappaport Hovav & Levin 2008):
(8) a. Max handed her a cigarette, but she wouldn’t take it.
(Oehrle 1977:206 (4a))
b. When I took him his mail, I found that he had disappeared.
(Oehrle 1977:206 (4c))
Idioms with dative verbs have been also given as evidence for the firm association
of one semantic interpretation for each construction (e.g., Pinker 1989:110-111; Krifka
2004:12). For example, Krifka (2004) argued that (9a) is acceptable because it conveys
the meaning that Ann possesses the idea, and (9b) is unacceptable because there is
no motion of the idea from Ann to Beth.
(9) a. Ann’s behavior gave Beth this idea. (Krifka 2004:12 (66a))
b. * Ann’s behavior gave this idea to Beth. (Krifka 2004:12 (66b))
Nevertheless, Green (1971) observed that even idioms like these can occur with both
construction types with certain types of arguments (viz. with pronominal themes
or heavy NPs; see also Bresnan & Nikitina 2009). In these cases, the meaning of the
idiom is unchanged by the change in construction:
(10) a. Walt gave Dick the finger, and he gave it to Ted, too.
(Green 1971:87 (19a))
b. That gave an idea to all of us who had read the assignments faithfully.
(Green 1971:90 (39d))
1.2. THE ALTERNATIONS 7
In essence, the semantic biases of these constructions can be neutralized under certain
conditions, and meaning does not fully determine the syntactic choice (see discussions
in Rappaport Hovav & Levin 2008; Bresnan & Nikitina 2009).
While semantic differences should be taken into account in any study of how
children acquire alternations, I will limit my consideration of them to the following
question: When describing identical events, under what circumstances is discourse
pressure able to override semantic biases that encourage the speaker to choose one
construction over the other?
1.2.2 Syntactic properties
Locative and dative alternations are also well suited to research on discourse effects
because of the structural similarity of the two variants. For the locative alternation,
there is no potential confound between complexity of the surface syntactic structure
and choice of construction. Both variants take the form in (11a). The two alternates
of the dative alternation are also similar in syntactic structure, one having the form
in (11a) and the other the form in (11b).2
(11) a. verb + NP + PP
b. verb + NP + NP
Children produce both of these structures very early. Based on data from CHILDES
(MacWhinney 2000), Gropen et al. (1989:212) reported that both prepositional da-
tives (verb + NP + PP) and double-object datives (verb + NP + NP) are frequent2Scholars disagree on whether the variants have underlying structures distinct from the surface
structures (see Waryas & Stremel 1974; Goldberg et al. 2005; Emonds & Whitney 2006; Rappa-
port Hovav & Levin 2008, among many others). Derivational accounts have implications for the
difficulty of each variant for child speakers; but for this dissertation, I take a non-derivational ap-
proach to the syntax of these constructions and limit my attention to whether children have learned
the given surface structures.
8 CHAPTER 1. INTRODUCTION
in child language and emerge as early as 1;8 (even excluding formulaic responses).
Children in Gropen’s study used both of these dative constructions in spontaneous
speech by age 2 or 3, and there was no clear evidence that one structure emerged
before the other (see §1.3.2 for examples and further discussion).
These facts differ somewhat from those for active/passive alternation. Researchers
generally agree that children learning English master the syntax of active voice before
that of passive voice (e.g., Roeper et al. 1981; but see Clark & Carpenter 1989a;
1989b for evidence that children use functional equivalents of adult-like passives by
age 2; see also Demuth 1989; Budwig 1990). At this point, we cannot conclude (as
some narrative studies have done, see §1.4) that a delay in the use of passive voice
to mark discourse status reflects a general delay in the ability to mark discourse
status with word order. Children may simply avoid passives because they prefer an
easier syntactic structure. By studying behavior with identical or very similar surface
structures, this research project largely avoids the confound of structural complexity.
1.2.3 Discourse properties
In adult language, the ordering of postverbal arguments tends to follow the given-
before-new pattern in cases where multiple argument realizations are possible. For
example, the choice of construction with alternating dative verbs is affected by the
status of the referent as given or new in the discourse (see Halliday 1967a; Halliday
1970; Erteschik-Shir 1979; Givón 1988; Primus 1998; Rappaport Hovav & Levin 2008).
This has been shown in a variety of experimental and corpus studies on several differ-
ent languages (English: Ransom 1979; Smyth et al. 1979; Bock & Irwin 1980; Givón
1984; Thompson 1990; Collins 1995; Arnold et al. 2000; Wasow 2002; Bresnan et al.
2007; Bresnan & Nikitina 2009; see also Waryas & Stremel 1974; Ozón 2006; Japanese:
Ferreira & Yoshita 2003; Korean: Choi 2008; Tahitian and Tahitian French: Snyder
2003). Work on syntactic discourse effects in acquisition, however, has mostly focused
1.3. ACQUISITION OF ARGUMENT STRUCTURE 9
on alternations involving the grammatical subject (see §1.4.2), but these findings
might not be generalizable to other syntactic positions. As mentioned above, chil-
dren might recognize that the subject position has a privileged status with respect
to the discourse (e.g., “when possible, make a given referent the subject”), but do not
yet have a more general notion that given information should be ordered before new
information.
Also, even if children understand that subjects are discourse prominent relative
to objects, they may not understand that the same asymmetry holds between the
first and second arguments in the VP. As discussed in Givón (1984), the discourse
prominence (‘topicality’) of the first object after the verb is secondary to that of
the subject; and across languages, there is less consistency in the marking of the
special discourse status of the first object in comparison to the subject. In fact,
Givón (1984:178) suggested that maintaining two topics in a discourse is universally
less common than maintaining one topic, and that, “This seeming limitation may
ultimately reflect some neurological limits on the processing complexity of the human
organism.” If children show a given-before-new ordering preference for postverbal
arguments, this will provide compelling evidence that children use word order to
mark discourse status more generally.
1.3 Acquisition of argument structure
The literature on the acquisition of verb argument structure is extensive (for a concise
review see Behrens 2007). Broadly speaking, theories of argument structure acqui-
sition can be characterized as either ‘top-down’ or ‘bottom-up’. Top-down accounts
tend to focus on the link between syntax and semantics and argue that (innate)
linking rules provide a way for children to learn syntax from the semantics (seman-
tic bootstrapping, e.g., Pinker 1989, 1994) or to learn semantics from the syntax
10 CHAPTER 1. INTRODUCTION
(syntactic bootstrapping; e.g., Gleitman 1990). Proponents of these theories have ar-
gued that children at an early age are sensitive to the (subtle) semantic differences
among different syntactic constructions (e.g., Gropen et al. 1989,1991b; Fisher et al.
1994; Brinkmann 1995; Kim et al. 1999; but see Bowerman 1989).
Another type of top-down theory posits that children approach syntax from a
conceptual starting point (Slobin 1985). Very early in development, children be-
gin to categorize events, and they subsequently discover how the event types (e.g.,
agent acts on theme) are encoded in the ambient language (e.g., as transitive sen-
tences). Along these lines, some researchers have considered whether children encode
prototypical events (events they are exposed to more) earlier than non-prototypical
ones. Although children seem to have more difficulty comprehending descriptions
of atypical events than typical events (see 12-13), spontaneous production data do
not support the hypothesis that prototypical events are actually encoded earlier than
non-prototypical events (see Bowerman 1985; Uziel-Karl 2002; Stephens 2008).
(12) Transitive sentences (see Chapman & Miller 1975)
a. Typical event: The boy is carrying the truck.
b. Atypical event: The truck is carrying the boy.
(13) Dative sentences (see Osgood & Zehler 1981)
a. Typical event: The brother gives the sister the block.
b. Atypical event: The tiger gives the cat the puppy.
As Behrens (2007:196) concluded, “To date, there is no evidence for a privileged
conceptual starting point for the acquisition of verb syntax and semantics.”
More recently, bottom-up approaches to argument structure have gained cur-
rency. These accounts argue that knowledge of argument structure can be built
up in a piecemeal fashion by observing distributional patterns in the input (see
1.3. ACQUISITION OF ARGUMENT STRUCTURE 11
Tomasello 2003; Goldberg 2006, and references in §1.5.3). For example, several stud-
ies have shown that item-based input frequency plays an important role in the ac-
quisition of syntactic constructions (e.g., de Villiers 1985; Naigles & Hoff-Ginsberg
1998; Cameron-Faulkner et al. 2003; Demuth et al. 2003; Demuth et al. 2005).
One important notion from this literature is Braine & Brooks’s (1995) entrench-
ment hypothesis (see also Brooks et al. 1999; Theakston et al. 2004; Matthews et al.
2005). As summarized in Ambridge et al. (2008:97), entrenchment means that “re-
peated presentation of a verb (e.g., disappear) in one (or more) attested construction
(such as the intransitive construction; e.g., The rabbit disappeared) causes the learner
to gradually form a probabilistic inference that adult speakers do not use that par-
ticular verb in non-attested constructions.” If this is true, argument structure errors
should be more likely with verbs that are less frequent in CDS.
In sum, previous work on the acquisition of argument structure has uncovered a
range of factors that influence early syntax. Some key factors are given in (14).
(14) a. Semantic/conceptual: event type (e.g., caused motion vs. caused posses-
sion), participant type (e.g., agent vs. patient), animacy, affectedness,
prototypicality, plausibility, salience
b. Distributional: input frequency and distribution of verbs, constructions,
and argument types (e.g., noun vs. pronoun; human vs. inanimate)
My dissertation incorporates insights from this literature and attempts to further our
understanding of the factors that influence argument structure acquisition. In the
next subsections, I provide a summary of previous work on the argument structure
of locative and dative verbs in child language acquisition. Despite the interest in
alternations within the argument structure literature, there has been relatively little
work on these alternations (cf. work on active/passive and causative alternations);
and with a few exceptions, none of this work takes discourse into account.
12 CHAPTER 1. INTRODUCTION
1.3.1 Locative alternation
Using diary data, Bowerman (1982) drew attention to the fact that children some-
times make errors with non-alternating locative verbs, and that these errors are not
restricted to a specific locative construction (examples from Bowerman 1982:338):
(15) a. Locatum-first error: Can I fill some salt into the bear?
b. Location-first error: I spilled it of orange juice.
Bowerman argued that these errors result from overgeneralizing the argument struc-
ture of some verbs to semantically similar verbs. Most subsequent work on the ac-
quisition of locative alternation has focused on semantic aspects of the verbs and
constructions involved.
In a cross-linguistic production study, Kim et al. (1999) elicited locative construc-
tions from children (aged 3 and 4) for 14 locative verbs using both partitive and
holistic events. There was no effect of the partitive/holistic manipulation. Con-
struction choices from the adult control group were very similar to the children’s
construction choices, except with the verb fill. English-speaking adults never used
fill (an arguably non-alternating location-object verb) with locatum-objects, but
children did so 56% of the time. The other location-object verbs, however, were
produced with high accuracy, and the verb-by-verb correlations between mother and
child uses were very high. Thus, the errors with fill were probably not due to the fact
that locatum-object constructions are generally more common across locative verbs.
Kim et al. speculated that children make errors with fill because this verb exhibits
a large amount of cross-linguistic diversity in its argument expression, and English-
speaking children have not yet learned the ‘idiosyncratic rule” that fill requires a
location-object.
The most extensive research on the acquisition of English locative alternation is
that of Jess Gropen and his colleagues (Gropen 1989; Gropen et al. 1991a,1991b).
1.3. ACQUISITION OF ARGUMENT STRUCTURE 13
These researchers argued that children use an innate and universal rule which links
affected arguments to the object position. They elicited locative event descriptions
for familiar verbs (pour, fill, dump, empty, stuff, splash) (Gropen 1989: Experiments
1-2; Gropen et al. 1991b) and for novel verbs (Gropen 1989: Experiments 3-6; Gropen
et al. 1991a). The children in these studies ranged in age from 2;6 to 9;1, and the pri-
mary focus was on how the manner and endstate of the event influenced construction
choice (see §1.2.1). Overall, Gropen and his colleagues argued that more locatum-
object responses were produced when the locatum was affected (i.e., when the man-
ner was highlighted), and more location-object responses were produced when the
location was affected (i.e., when the endstate was highlighted). Gropen’s work also
tangentially addresses issues of information structure. I discuss his findings related
to discourse status in Section 1.4.
1.3.2 Dative alternation
Early work on the acquisition of the dative alternation focused primarily on the order
in which children acquire the two dative constructions. Several comprehension studies
found that children (aged 3-10) had more difficulty acting out double-object datives
than prepositional datives (e.g., Cook 1976; Roeper et al. 1981; Osgood & Zehler
1981). These studies manipulated the animacy of the postverbal arguments (theme
and recipient). This manipulation influenced the interpretation of double-object
datives, but not necessarily that of prepositional datives. In particular, children had
trouble acting out sentences like those in (16), which deviate from the prototypical
pattern of animate recipient and inanimate theme (e.g., Give the man the car ; see
also Fraser et al. 1963; Fischer 1971).
(16) a. Give the car the man. (inanimate recipient, animate theme)
b. Give the cow the horse. (animate recipient, animate theme)
14 CHAPTER 1. INTRODUCTION
But as White (1987) noted, adults might also struggle to interpret these sentences,
perhaps because they describe highly atypical events. In fact, in a preliminary study of
adult comprehension of dative constructions, Steedman & Johnson-Laird (1978:188)
found that adults were slower to read double-object datives with two animate postver-
bal arguments than to read corresponding prepositional datives. When the theme
was inanimate and the recipient animate, latencies were essentially the same for
double-object and prepositional datives. White (1987) tested children (aged 3;8-5;8)
on their comprehension of plausible double-object and prepositional datives (ani-
mate recipients, inanimate themes). She found that accuracy was not related to
construction type, and children never misinterpreted the thematic roles of the two
postverbal arguments. Using a novel verb paradigm in a production study, Conwell
& Demuth (2007) also found that three-year-olds have productive knowledge of the
dative alternation and, under certain experimental conditions, can generalize a verb
they learned in one construction to the other (see also Shimpi et al. 2007).
Corpus studies with familiar verbs further indicate that young children use both
prepositional and double-object datives. As mentioned in Section 1.2.2, Gropen et al.
(1989) found no compelling evidence that one construction is acquired before the
other. On the other hand, Viau (2006) argued that double-object datives emerge
“significantly” earlier than prepositional datives (see also Snyder & Stromswold 1997).
But in Viau’s corpus study, the average gap between the first use of each was only
3 months. This difference may be statistically significant, but it need not mean that
the syntax of prepositional datives is more difficult to acquire than that of double-
object datives or that the gap has important implications for language development.3
Despite some controversy over the order of acquisition, studies of spontaneous child
speech consistently show that children use both constructions by age 3. Table 1.13Like Snyder & Stromswold (1997), Viau’s results are based on age of first mention. For a critical
analysis of this measure see Abbot-Smith & Behrens 2006.
1.3. ACQUISITION OF ARGUMENT STRUCTURE 15
provides a summary of these studies, all of which included a broad range of verbs.4
Some examples of early datives (including non-adult-like uses) are given below.
(17) Examples of prepositional datives (from Snyder 2003:53)
a. She bringed the tiny little baby pony to Daddy. (Terrance M., 4;11)
b. I gave a piece of sugar candy to Jonah yesterday. (Billy B., 7;2)
(18) Examples of double-object datives (from Gropen et al. 1989)
a. Show Fraser horsie. (Eve, 1;9)
b. You can write me a lady on that page. (Eve, 2;3)
c. Give doggie paper. (Adam, 2;3)
d. Jay said me no. (Ross, 2;8)
Study Double-object datives Prepositional dativesGropen et al. (1989) 1;8-2;11 1;11-3;0Snyder & Stromswold (1997) 1;8-2;11 2;0-3;4Campbell & Tomasello (2001) 1;6-2;9 1;10-3;2Viau (2006) 1;7-2;10 1;10-3;4
Table 1.1: CHILDES: Age of first use of double-object datives and prepositionaldatives in previous corpus studies.
These experimental and corpus results clearly demonstrate that children are able
to produce both variants of the dative alternation by age 4 (the age tested in my dative
studies). Prior research has also suggested that the following factors also influence
the acquisition of dative constructions: (a) item-based input frequency of the verb
in a particular construction (Campbell & Tomasello 2001), (b) structural persistence4Except for Campbell & Tomasello (2001), all of the authors mentioned in Table 1.1 specified that
they excluded formulaic utterances for their analyses. For lists of verbs found in these spontaneous
child datives, see Gropen et al. (1989:213ff) and Campbell & Tomasello (2001:260).
16 CHAPTER 1. INTRODUCTION
(Shimpi et al. 2007; Thothathiri & Snedeker 2008; de Marneffe et al. Submitted; see
also Conwell & Demuth 2007), (c) semantic properties of the event (Mazurkewich
& White 1984; Gropen et al. 1989; cf. Campbell & Tomasello 2001; Goldberg et al.
2005; Conwell & Demuth 2007), and (d) morphological characteristics of the verb (see
Mazurkewich & White 1984; Gropen et al. 1989). In Section 1.4.2 below, I discuss
three studies that also address information structure in the acquisition of English
dative alternation.
1.4 Acquisition of information structure
Most discussions of information structure and notions like “given” and “new” informa-
tion begin with complaints about the terminological and conceptual chaos surrounding
these ideas. In light of this confusion, a careful discussion of terminology is in order
before I review relevant findings in the acquisition literature. Kruijff-Korbayová &
Steedman (2003:250) defined information structure broadly as:
the utterance-internal structural and semantic properties reflecting the
relation of an utterance to the discourse context, in terms of the discourse
status of its content, the actual and attributed attentional states of the
discourse participants, and the participants’ prior and changing attitudes
(knowledge, beliefs, intentions, expectations, etc.).
This definition encompasses a host of different notions that have been proposed in
the literature to account for discourse-related phenomena. As Kruijff-Korbayová &
Steedman (2003) pointed out, theories about information structure, though diverse,
generally draw from one or both of two (often overlapping) dimensions.
One dimension corresponds to Gundel’s category of “relational givenness” and
subsumes notions such as “topic/comment,” “theme/rheme,” and “topic/focus” (see
1.4. ACQUISITION OF INFORMATION STRUCTURE 17
Gundel 1988; Gundel & Fretheim 2004). This dimension distinguishes between two
main parts of an utterance. The first part relates the utterance to the discourse
purpose and can be characterized roughly as what the utterance is about (see, e.g.,
Gundel 1988; Halliday 1967b:211). The second part further advances the discourse
and generally provides new information associated with the first part. The other
dimension corresponds to Gundel’s “referential givenness” and is tied to notions such
as “given/new,” “background/kontrast,” and “contextually bound/nonbound.” This
dimension does not necessarily break the sentence into two parts, but rather dis-
tinguishes between elements (e.g., referring expressions) in an utterance in terms of
their informational status in the discourse. For overviews and comparisons of these
ideas, see Bates & MacWhinney (1979), Vallduví (1992:Ch. 3), Arnold (1998), and
Kruijff-Korbayová & Steedman (2003), among others.
This dissertation essentially concerns the second dimension mentioned above (viz.
referential givenness). I am interested, in particular, in the relative ordering of two NP
arguments within the verb phrase and how this order depends on their informational
status in the discourse. Information structure notions that divide a sentence only
into two parts, like “topic/comment,” do not straightforwardly capture informational
differences between multiple elements or arguments within the predicate (see Givón
1984:154; Vallduví 1992:34). Also, I will not be concerned with the discourse status of
the grammatical subject, though it might be considered more ‘given’ or ‘topical’ than
either of the postverbal arguments. In my studies, the status of the subject is held
constant across discourse conditions. For this research, I need a notion of discourse
status that can apply to sentence elements that are not necessarily the ‘primary topic’
(see Givón 1984:154).
Restricting the focus to referential givenness, however, still leaves open the ques-
tion of how exactly givenness is defined. Here, two questions are relevant:
(19) a. How is givenness conceived of generally?
18 CHAPTER 1. INTRODUCTION
b. Is givenness a binary distinction?
Regarding the first question (19a), I will use the term givenness in the sense of what
Clark & Marshall (1981) referred to as linguistic copresence—a referent is given if
it has been mentioned in the conversation, and new otherwise (i.e., if it is textually
evoked in the sense of Prince 1981, or discourse-old information as in Prince 1992
and Ward et al. 2002). Bock & Irwin (1980:467) noted that, “a definite referring
expression [. . . ] with a coreferential antecedent in an immediately preceding sentence
is an uncontroversial example of given information.” For the experiments discussed
below, all given entities satisfy this description.
Importantly, this type of givenness (viz. linguistic copresence) may or may not
overlap with givenness due to what Clark & Marshall (1981) called physical copres-
ence (i.e., when the referent is perceptually available and jointly attended to by the
speaker and addressee—a situationally evoked referent in Prince’s taxonomy). In my
studies, all of the referents for the arguments of the verb are physically copresent, but
only some of them are linguistically copresent, as will be made clear in the methods
sections. Crucially, my use of the term ‘given’ will only refer to entities that are lin-
guistically copresent, regardless of their psychical copresence. Though these types of
mutual knowledge can be difficult to tease apart in corpus data, experimental settings
allow for a careful specification of both.
Givenness in terms of linguistic copresence should also be distinguished from given-
ness that is defined with reference to knowledge states. Prince (1981) summarized sev-
eral ways in which givenness has been equated with knowledge states (see also O’Neill
2005; Birner & Ward 2009): information is considered ‘given’ (or ‘old’, ‘known’, etc.)
if (i) the speaker assumes the hearer can predict its occurrence in the ensuing dis-
course or recover it from the preceding discourse (e.g., Kuno 1972; Halliday 1967b),
(ii) the speaker assumes the hearer has that information in his or her consciousness
at the time of the utterance (e.g., Chafe 1976), or (iii) the speaker assumes the hearer
1.4. ACQUISITION OF INFORMATION STRUCTURE 19
knows or can infer the information even if the hearer is not thinking about it (e.g.,
Clark & Haviland 1977).
One aspect that unites these three notions of givenness is that they make reference
to something the speaker assumes about the knowledge state of the hearer. While
this assumption on the part of the speaker may well be what drives speakers to
treat various pieces of information differently in a discourse, a concept of givenness
based on speaker assumptions somewhat unwieldy for linguistic analysis (cf. Collins
1995:41). The linguist labeling experimental or corpus data with respect to these
categories of givenness must try to estimate what the speaker was assuming about
the hearer’s knowledge state at the time of the utterance. In other words, they must
make an assumption about what the speaker was assuming about what the hearer
was thinking.
With respect to my studies, when I refer to an entity as ‘given’, I mean the
following: a referent that has been mentioned explicitly (and repeatedly)
in the immediately preceding discourse. I do not mean to assert that the
speakers assume anything about the hearers’ thinking with respect to that entity. As
discussed in Section 1.5, it is theoretically important whether given (i.e., linguistically
copresent) entities have a special status in the mind of the speaker and whether the
speaker assumes they have such a status in the mind of the addressee. For this
reason, it is especially important to separate givenness in terms of cognitive status
from givenness in terms of linguistic (or even physical) copresence.
Orthogonal to the question of what “givenness” means is the question of whether
givenness is a binary distinction. Some argue that it is (e.g., Chafe 1976), others
use three or more categories of givenness (e.g., Prince 1992; Birner 2006), and still
others posit a givenness/accessibility hierarchy (e.g., Givón 1984; Gundel et al. 1993;
Ariel 2001) or other gradable notions of givenness. For example, within the Prague
School framework and the theory of Functional Sentence Perspective, Jan Firbas (e.g.,
20 CHAPTER 1. INTRODUCTION
1964, 1966, 1983) defended the idea that sentence elements carry (and are ordered
according to) a degree of “communicative dynamism,” which is “the extent to which
the sentence element contributes to the development of the communication, to which,
as it were, it ‘pushes’ the communication forward” (Firbas 1966:240). This notion can
be thought of as resulting in a type of discourse status hierarchy. As Firbas pointed
out, information that is new to the discourse carries a higher degree of communicative
dynamism than information that has already been introduced (e.g., Firbas 1964:270).
The claims made in this dissertation do not depend crucially on whether or not
givenness is gradable. Because the discourse context is carefully controlled, either the
referent for one of the postverbal arguments is mentioned the same number of times
and in the same way within a given study or not at all. Nevertheless, the number
of mentions does vary somewhat between studies. The degree of givenness, then, in
terms of the number of mentions might be relevant to the results.
Though there is some child language research on information structure and word
order, most work on the effects of givenness on childhood linguistic choices has an-
alyzed the types of referring expressions children use (e.g., pronoun, definite NP,
indefinite NP). This work largely ignores the syntactic context in which these refer-
ring expressions occur or analyzes the syntactic ordering separately. Overall, there
is very little work that systematically considers associations between word order and
referring expressions (for a few exceptions, see Karmiloff-Smith 1980; Hickmann et al.
1996; Prat-Sala & Hahn 2007; de Marneffe et al. Submitted). The next few subsec-
tions provide an overview of studies that address the effects of givenness on child
referring expressions and word order.
1.4.1 Referring expressions
The influence of givenness on referring expressions is well known. Definite NPs or
pronouns tend to refer to given information and indefinite NPs tend to refer to new
1.4. ACQUISITION OF INFORMATION STRUCTURE 21
information (e.g.,Chafe 1976; Clark & Haviland 1977; Levelt 1989; Ariel 2001). But
the precise conditions under which a particular type of referring expression is chosen
are complex and involve many more factors than simply whether the information was
mentioned in the prior discourse (see discussions in Brown 1973:340-356; Hickmann
2003:Ch. 5). The age at which children master this system is a matter of debate.
Emslie & Stevenson (1981) argued that this mastery is evidenced by age 3 (see also
MacWhinney & Bates 1978), but Kail & Hickmann (1992) and Hickmann et al. (1995)
found errors in the choice of referring expressions as late as age 9. As discussed at
length in Hickmann (2003), the disparity across studies regarding age of acquisition
can be largely attributed to methodological differences, such as the type of stimuli
used (e.g., filmed vignettes vs. picture books) and whether both interlocutors were
jointly attending to the stimuli (see also Kail & Hickmann 1992; Hickmann et al.
1995; Hickmann 2004).
Across studies, however, one recurring question is whether children take the ad-
dressee’s perspective into account when choosing referring expressions. Brown (1973)
observed that young children are prone to making ‘egocentric’ errors with referring
expressions. When a referent is familiar to the child but unfamiliar to the addressee,
the child will often use a definite referring expression (signaling familiarity to the
addressee) rather than an indefinite one as in the examples in (20) taken from Brown
(1973:353).
(20) a. Sarah: The cat’s dead.
Mother: What cat?
b. Sarah: I want to open the door.
Mother: What door?
22 CHAPTER 1. INTRODUCTION
This type of error has also been reported in numerous elicited production studies
(e.g., Maratsos 1974; Warden 1976; Power & Dal Martello 1986; but cf. Emslie &
Stevenson 1981).
While reports of ‘egocentric’ errors cast some doubt on the ability of young children
to take the perspective of the addressee into account when choosing between multiple
means of expressing the same referential content, they do not warrant the conclusion
that discourse fails to influence children’s choice of referring expressions. The children
in these studies showed appropriate (although not error-free) differentiation in their
use of definiteness marking (see Power & Dal Martello 1986). For example, when told
a story introducing a particular character (e.g., “Once there was someone who wanted
to have an animal. . . . He saw two animals, a frog and a turtle. . . ”), children as young
as 3 subsequently referred to the given character using a definite NP (rather than an
indefinite one) 83% of the time (Maratsos 1974). And Campbell et al. (2000) found
that young children (aged 2;6 and 3;6) were more likely to use pronouns or ellipsis
when the referent was mentioned in prior discourse. Similar findings were reported
by Matthews et al. (2006) and Serratrice (2008) for preschool children (aged 3-5).
Thus, while the work of Maya Hickmann and her colleagues has repeatedly shown
that adult-like use of referring expressions in connected discourse is not fully attained
until well after preschool, there is converging evidence that preschool children are sen-
sitive to givenness in discourse and exploit that sensitivity in their choice of referring
expressions.
1.4.2 Word order
The effects of givenness on word order are less well reported and even less clear than
those for referring expressions. Most of this work concerns the choice of grammatical
subject, often in elicited narratives. For example, in a story elicitation task, Karmiloff-
Smith (1980) found that children did not choose the subject of their sentences based
1.4. ACQUISITION OF INFORMATION STRUCTURE 23
on prior discourse until around age 5. Work on lengthier child narratives claims an
absence of discourse-motivated word order choices until age 7 or later (Hickmann
et al. 1996; Hickmann 1997; Hickmann & Hendriks 1999). And in a cross-linguistic
developmental study of givenness, MacWhinney & Bates (1978) found that “initial-
ization” of nouns during a picture description task was not related to givenness in the
speech of children (or adults!).
Other work, however, indicates that children are sensitive to properties of the dis-
course and that this knowledge affects their comprehension and production. Song &
Fisher (2005, 2007) found that children as young as 2;6 expect discourse prominent
referents to be sentential subjects (but cf. Arnold et al. 2007). Also, Snow (1979:173)
noted that early multiword constructions by children learning Dutch can be charac-
terized as sentences in which the most topical information (e.g., given information)
proceeds less topical information.
A few elicited production studies have also shown effects of discourse on word
order (Braine et al. 1990; Brooks et al. 1999; Prat-Sala & Hahn 2007, see also Turner
& Rommetveit 1968). Braine et al. (1990) tested English-speaking children (aged 2
and 4) and adults on their use of three verb types: (a) fixed transitivity verbs (either
transitive or intransitive), (b) optionally transitive verbs, and (c) novel verbs. The
experimenter (or the child) acted out the target event, and event descriptions were
elicited via the prompts in (21).5
(21) Example stimulus: Experimenter flips a spoon by hitting it on one end.
a. Agent-given: What am I doing?
b. Patient-given: What is the spoon doing?
c. Open question: What’s happening?
5The agent and patient labels used here are the labels used by Braine and his colleagues.
24 CHAPTER 1. INTRODUCTION
Braine and his colleagues found that children (of both ages) made discourse-motivated
syntactic choices. For all three verb types, children were most likely to use an agent-
subject (causative) sentence when the agent was given (e.g., You’re VERB-ing the
spoon.) and most likely to use a patient-subject (inchoative) sentence when the
patient was given (e.g., The spoon is VERB-ing.). The results for the adult data
were similar, except that adults (as expected) did not show any alternations with
fixed transitivity verbs. Braine et al. (1990:338) concluded that:
[D]iscourse pressure can cause children to use canonical sentence schemas
to supply argument structures that fit the needs of the discourse frame,
even when the verbs are very familiar and the argument structures are
not sanctioned by the lexical entry. [. . . ] In effect, there is competition
for control of the response between the lexical entry and the canonical
sentence schema that fits the needed discourse frame.
This study and a similar study by Brooks et al. (1999) provide convincing evidence
that discourse can influence word order before age 5. But do the effects of discourse
in these studies reflect a general preference to order given information before new
information or a more specific preference to use prominent discourse participants as
sentential subjects?
A few additional studies have argued that discourse/pragmatics plays a role in
child word order. But these studies, mostly on the acquisition of Italian, argue for
an early new-before-given strategy. Fava (1978) analyzed longitudinal tape-recorded
data from six children learning a dialect of Italian. The children were recorded once a
week for a year starting at the age of 1;6. Fava reported that children used VS orders
(VS, VOS, VSO) in the first stages of their development and SV orders (inclucing
SV and SVO) later. She attributed this finding to an early tendency to position new
information before given information, though she offered no systematic analysis of the
1.4. ACQUISITION OF INFORMATION STRUCTURE 25
information structure of these utterances. According to Fava, “Very often the word, or
words conveying the most information and which are firstly selected [=mentioned first]
by the child are not part of the background [the] child and adult have in common” (p.
518). Fava also suggested that a tendency to use “comment-topic” order (new-before-
given) explains early choices between VSO and VOS structures, and that children used
“topic-comment” order (given-before-new) as a repair strategy when the comment-
topic utterances were not understood.
In another longitudinal study of children learning Italian, Bates (1976) also an-
alyzed early orderings of the subject, verb, and object. For the two children Bates
studied, Claudia and Francesco, she found that in the earlier transcripts (viz. Claudia:
1;3-1;9; Francesco 1;4-2;0), “there is actually a statistical preference for subject-final
constructions” (p. 188).6 Bates provided very little information about the types
of arguments the children used as subjects and objects during this period, though
she mentioned that both children referred to the speaker and addressee with proper
names, rather than first and second person pronouns (see Bates 1976:196). Conse-
quently, most of the subjects and objects were probably proper names and lexical
NPs, though the children may have also used some third person pronouns.
Bates also attributed this early subject-final strategy to an early tendency to or-
der new information before old information. One explanation offered for this is that
children “[extend] the old figure-ground strategy by blurting out the new or most in-
teresting information first, and adding other information on as a sort of afterthought,
until [they run] out of processing space” (p. 160). In other words, young children
may use an attention-based “first-things-first” strategy (see Gundel 1988:229). Ac-
cording to Bates, around MLU 3.0, both children began to alternate more between
a subject-final and a subject-first strategy. This was also about the time when they6The following utterance types were included in this analysis: SV, VS, OV, VO, SO, OS, V only,
SVO, SOV, OVS, OSV, VSO, VOS. See Bates (1976:185–200) for details.
26 CHAPTER 1. INTRODUCTION
Child Subject-before-object Object-before-subject
Claudia 9 SVO 6 OSV6 VSO 1 VOS
Subotal 15 7
Francesco 8 SVO 10 OSV4 VSO 11 VOS
Subotal 12 21
Total 27 28
Table 1.2: Bates 1976: Summary of Claudio and Francesco’s early SO and OS pro-ductions.
began using first and second person pronouns, so the types of arguments the children
produced may have played a role in these early syntactic choices.
A closer look at Bates’ data from the period when the children purportedly pre-
ferred a subject-final strategy, however, reveals that this pattern is not clearly ev-
idenced when it comes to the relative ordering of arguments. A summary of the
data that involve a verb and two nominal arguments from the periods mentioned
above is given in Table 1.2. Claudia produced more SO utterances and Francesco
more OS utterances. Also, the overall count of these utterances is very low, so it is
difficult to draw firm conclusions from these data. While these children may have
more subject-final utterances in this stage of development than later, the data do not
provide compelling evidence for a new-before-given strategy for the relative ordering
of sentential arguments. Also, though it is generally fair to assume that the subject
will be given information relative to the object, Bates did not provide any informa-
tion about the actual discourse status of these arguments. Overall, then, evidence
from spontaneous speech data for a new-before-given strategy in early multi-argument
productions is inconclusive at best.
1.4. ACQUISITION OF INFORMATION STRUCTURE 27
To my knowledge, the only experimental work that was specifically designed to test
whether children observe given-before-new ordering is Narasimhan & Dimroth (2008).
These researchers investigated conjunct ordering by German-speaking children (aged
3 and 5) and adults. Participants saw an object (e.g., an egg) in a transparent,
round container, and an experimenter who could not see the object asked, Was ist da
drin? (‘What’s in there?’). After participants labeled the object, they were shown
two objects in the container, the previous object (the egg) and a new object (e.g., a
bed). The experimenter then repeated the question (Was ist da drin? ). Narasimhan
& Dimroth found that children tended to label the two objects with new-before-
given ordering (bed and egg), while adults preferred given-before-new ordering (egg
and bed). Narasimhan & Dimroth speculated that this difference in child and adult
responses might reflect a bias that young children have to focus on novel entities and
events (see Bates 1976). Despite the fact that adults still preferred given-before-new
ordering, it is questionable whether givenness should affect conjunct ordering in the
same way it affects argument ordering (see Bock & Warren 1985). I will return to
these results in more detail in Chapter 5.
Locative alternation
Most work on the acquisition of locative alternation has focused on the lexical seman-
tic properties that purportedly drive the alternation (see §1.2.1). Very little has been
said about potential effects of information status on this alternation, though Bower-
man (1982) argued that word-order errors as in (15) are not motivated by discourse
pressure (i.e., pressure to use given-before-new ordering) until age 5 or later.
Nevertheless, Gropen’s (1989) data suggest that information structure may, in
fact, play a role in early construction choices with locative verbs. Gropen systemati-
cally varied discourse status by mentioning either the locatum or the location in
the prompt (see Gropen 1989:36; italics indicate prosodic stress):
28 CHAPTER 1. INTRODUCTION
(22) Example stimulus: Three panel drawing of a woman filling a glass with water
a. Locatum-given: Point to the glass ! . . . Say glass ! . . . Say filling ! . . .What
is the woman doing to the glass?
b. Location-given: Point to the water ! . . . Say water ! . . . Say filling !
. . .What is the woman doing to the water?
Gropen was not interested in the development of discourse sensitivity per se, but “used
the pragmatics of the query in order to flush out the range of possible forms that a
verb can take” [emphasis in original] (Gropen 1989:36). Gropen essentially assumed
that children would use given-before-new as long as the child’s grammar permitted
the relevant construction.
Overall, discourse seems to have played some role in Gropen’s data, but the results
were far from conclusive. Gropen used ANOVAs to test the effect of question type
on each verb individually. For the non-alternating verbs (pour, dump, fill), only the
‘incorrect’ forms were analyzed (see 23-24).7
(23) location-objects with locatum-object verbs
a. *He poured the bucket with paint.
b. *She dumped the table with play-dough.
(24) locatum-object with location-object verb
*She filled paint into the bucket.
These analyses only showed a significant effect of question type for dump: There were
more erroneous location-object responses (23b) when the location was given than
when the locatum was given (i.e., given-before-new ordering). For the alternating
verbs (empty, stuff, splash), all responses (including adult responses) were included
in the analyses. For each of these verbs, there was a main effect of question type.
7These examples were constructed based on Gropen’s experimental stimuli.
1.4. ACQUISITION OF INFORMATION STRUCTURE 29
In each case, the given entity was more likely to be the direct object (again, given-
before-new).
Gropen also tested for interactions between question type and age. Two verbs,
splash and fill, showed this interaction. For splash, question type influenced syntactic
choices (again, given-before-new) for the older children (4;6-5;11) and adults but not
for the younger children (2;6-4;5). For fill, the older children, but not the younger
ones, unexpectedly produced more locatum-objects when the location was given
(new-before-given). Gropen (1989:43) concluded,
We have no explanation for why the oldest children flouted the discourse
function of locative forms; however, we note that, for every group of sub-
jects except the oldest child group, the query which treated the identity
of the content [= locatum] as old information was more successful in
eliciting content-locative forms (though not significantly so according to
post-hoc t-tests). [emphasis in original]
In three studies with novel verbs (Experiments 3-6), Gropen found the predicted
effect of question type, but did not report any age related differences or interactions
between age and question type. Also, there was no control condition with respect to
discourse (all prompts were either locatum- or location-given). Thus, the effect of
discourse in one condition over another is impossible to assess. While Gropen’s results
seem promising, they leave open the question of how and when discourse sensitivity
is reflected in word order.8
8Gropen’s results are also problematic because the form of the prompt (viz. What was the woman
doing to the glass? ) highlights the affectedness of the mentioned object. Recall that Gropen’s
primary goal was to assess how object affectedness influences construction choice. The question is
whether Gropen’s findings reflect discourse sensitivity or an attempt to realize the most affected
entity as the object.
30 CHAPTER 1. INTRODUCTION
Dative alternation
There are a few production studies that test the effects of discourse on children’s
ordering of postverbal arguments in dative constructions (viz. Gropen et al. 1989;
Snyder 2003; de Marneffe et al. Submitted). For example, in a novel verb experiment
with children aged 6;3-8;6, Gropen et al. (1989) varied the discourse status of the
theme and recipient in a manipulation similar to the one in the locative alternation
research mentioned above. Children were taught a novel verb of transfer in either
a prepositional or a double-object dative and asked questions that mentioned one
postverbal argument or the other (cf. 22):
(25) Example stimulus: Experimenter transferred a ball to a toy mouse.
a. Theme-given: Can you tell me what I’m doing with the ball?
b. Recipient-given: Can you tell me what I’m doing with the mouse?
Question type significantly affected construction choice, but only when the child used
a construction that was not modeled by the experimenter. When prepositional datives
were modeled (e.g., I’m mooping the ball to the mouse.), children produced double-
object datives (e.g., You’re mooping the mouse the ball.) more for recipient-given
questions (44%) than for theme-given questions (17%). When double-object datives
were modeled (e.g., I’m mooping the mouse the ball.), children produced prepositional
datives (e.g., You’re mooping the ball to the mouse.) more for theme-given questions
(83%) than for recipient-given questions (55%). In other words, productive uses
of dative constructions reflected a bias for given-before-new ordering. When children
used the construction that was modeled, there was no effect of question type (i.e., no
effect of givenness). Gropen et al. concluded that, “the salience of the construction in
which the verb was taught overshadowed any pragmatic effects of what was focused
in the query” (p. 232).
1.4. ACQUISITION OF INFORMATION STRUCTURE 31
Gropen’s results with novel dative verbs indicate that givenness can, under cer-
tain circumstances, influence syntactic choices in the speech of school-aged children.
Whether these results would generalize to familiar verbs and/or to younger children is
unclear. And as with the locative alternation studies, because there was not a neutral
discourse condition, there is no baseline with which to compare the discourse effects.
Again, Gropen et al. were not primarily interested in discourse factors, and they did
not consider the implications of discourse effects for the semantic and morphological
constraints they proposed.
In addition to Gropen’s work, two corpus studies have considered the role of in-
formation structure in dative alternations in child speech. Snyder (2003) analyzed
spontaneous uses of three alternating dative verbs (viz. give, take, bring) and bene-
factive uses of the verb make in the speech of children aged 3;3-8;1. These data were
extracted from a corpus of recorded caregiver-child and child-child interactions. Sny-
der excluded constructions with pronominal postverbal arguments to “avoid pesky
questions of cliticization” (p. 52). Her dataset consisted of 304 dative/benefactive
constructions involving two lexical postverbal arguments. Snyder found that heavier
postverbal arguments (as measured in syllables) tended to appear in final position.
Across verbs, age groups did not differ significantly with respect to this, though for
the verb give, older children were apparently more likely to place heavier arguments
last (though perhaps not statistically so). Since heaviness tends to be correlated with
information status (e.g., Ariel 2001), this finding may reflect a general tendency to
order given information before new information.
In a separate analysis, Snyder tested whether hearer-new recipients were (cate-
gorically) preferred in final position—a pattern she found in corpus data from adult
English speakers (see Snyder 2003:24). Snyder defined hearer-newness as “whether
the hearer [can] plausibly be assumed to be familiar with the NP” (p. 23), but she
offered no details about her coding of hearer-new information. This contrasts with
32 CHAPTER 1. INTRODUCTION
discourse-newness, which relates strictly to whether the entity is mentioned in the
discourse (Prince 1992; see discussion above). Snyder found that children aged 3-4
positioned hearer-new recipients in final position (NP PP) only about 20% of the
time, while children aged 5-6 did so about 70% of the time and those aged 7-8 about
80% of the time.
From this analysis, Snyder concluded that “children as old as seven or eight are
beginning to use information status in exactly the same way that adults do, while
very young children do not appear to use it at all.” Snyder did not report the choice
of construction based on the relative newness of the recipient with respect to the
theme, or the frequency of hearer-new recipients by age group, nor did she pro-
vide any statistical analysis of these data. Given these missing elements, we cannot
take these data as strong evidence against the notion that children aged 3-4 use
given-before-new ordering for postverbal arguments of alternating dative verbs. Also,
because Snyder only considered hearer-newness (and did not offer criteria for coding
this factor), these data do not allow us to assess the effects of discourse status on
early syntactic choices.
Using data from the childes database, de Marneffe et al. (Submitted) also con-
sidered whether discourse influences dative syntax. Based on a model of adult dative
alternation from Bresnan et al. (2007), they developed a multi-factor analysis of da-
tive alternation in both child speech (children aged 2-5) and CDS. In Bresnan et al.’s
(2007) model, givenness of the theme and the recipient were significant predictors
of adult construction choice, as were the related factors of pronominality, length, and
definiteness. These factors each made a significant independent contribution to the
model and were harmonically aligned (Aissen 1999)—referents that were discourse
given, pronominal, shorter, and definite tended to precede referents that were dis-
course new, lexical, longer, and indefinite. De Marneffe et al.’s work yielded a very
similar model for adult speech to children. In this model, pronominality and length
1.4. ACQUISITION OF INFORMATION STRUCTURE 33
of the theme and recipient were significant predictors as was givenness of the
theme, though not givenness of the recipient. They did not discuss why recip-
ient givenness was not significant in the CDS model, but perhaps there was not
enough variation in this feature to reliably predict a difference in construction choice
(e.g., because recipients were mostly given).
Importantly, the child and CDS models were also quite similar, except that theme
givenness (the least significant factor in the CDS model) was non-significant in the
child model. Can we conclude from this that givenness does not affect child dative
constructions? The child model did show that pronominality of the theme and of the
recipient were important predictors of construction choice. Children mostly used
pronouns for given entities, though they occasionally used them for new referents
as well. De Marneffe et al. suggested that, “the effect of givenness on children’s
dative choices may be masked by the larger proportion of cases where the influence
of givenness and pronominality lead towards different constructions” (p. 20).
One reason that givenness may not have been significant in the child model is
that givenness was coded in the same way in the child and adult data. A referent was
counted as given if there was a coreferential expression in the previous 10 turns of
the dialogue. But because children have a more limited memory and attention span
than adults, this criterion may be too inclusive. Perhaps givenness would indepen-
dently contribute to the model if fewer turns were considered. Another possibility
is that most themes and recipients in child speech are given. So again, there
may not be enough variability in givenness for the model to detect a robust effect.
In essence, while previous work has furthered our understanding of early use of the
dative alternation, the extent to which givenness influences child datives remains an
open question.
34 CHAPTER 1. INTRODUCTION
1.5 Accounting for given-before-new ordering
As mentioned above, there is little agreement as to whether children use word order to
mark givenness. There is also little consensus about why we should (or should not)
expect children to prefer given-before-new ordering. This question is an especially
thorny one because the potential sources for this ordering phenomenon may well
be interrelated. In this section, I outline a few potential explanations for differential
ordering of given and new information in child speech: (i) addressee needs, (ii) speaker
needs, and (iii) distribution in the input. It is surely no accident that these three
notions might each provide a suitable account for the selfsame phenomenon. For
example, are some patterns more frequent because they are easier to produce (or
comprehend), or are they easier to produce (or comprehend) because children have
more experience with them, or both? We know that children recruit a wide range of
knowledge to learn a language; perhaps considerations of communication, processing
and frequency all contribute to given-before-new order. While it may be impossible
to fully tease apart the types of accounts presented here, this dissertation should yield
insights into the sources of discourse-motivated word order.
1.5.1 Addressee-based accounts
Children might use given-before-new ordering to be cooperative communication part-
ners. As Wasow (2002:62) noted, “Just as the punch line naturally comes at the end
of a joke, it makes sense for speakers to lay the groundwork first and save the high-
light of an utterance for last.” By placing given information before new information,
speakers link what has been said to what will be said and, thus, make the discourse
more coherent and easier to comprehend (see Arnold et al. 2000). When speakers
facilitate comprehension in this way, they are abiding by what Clark & Haviland
(1977:9) referred to as the Given–New Contract:
1.5. ACCOUNTING FOR GIVEN-BEFORE-NEW ORDERING 35
(26) Given–New Contract: Try to construct the given and new information of each
utterance in context (a) so that the addressee is able to compute from memory
the unique antecedent that was intended for the given information, and (b) so
that he will not already have the new information attached to that antecedent.
Because addressees need to retrieve the given information from memory (e.g., find the
referent of a pronouns) and then attach the new information to it, “it is optimal to take
in the given information before the new information; otherwise, they have to hold the
new information temporarily while they search for the antecedent to which it is to be
attached” (Clark & Haviland 1977:259). So on an addressee-based account, children
should use given-before-new ordering to facilitate the addressee’s comprehension.
Determining whether speakers adjust their production to aid comprehension or
for more egoistic reasons is notoriously difficult, but recent studies of both adult
and child production provide compelling evidence that the background knowledge
or perspective of the addressee influences language production. For example, Galati
& Brennan (2010) found that adults attenuate their utterances more when retelling
a story to someone who they told it to previously than when retelling it to a new
addressee. Also, in a communication task, Nadig & Sedivy (2002) found that five-
and six-year-olds use adjectival modification differently depending on whether their
interlocutor could see the same set of items in the child’s view. Children provide ad-
jectival modification more often when it is necessary to uniquely identify a referent in
common ground than when it is necessary only from their own privileged perspective.
Whether preschool children have the cognitive and/or linguistic skills needed to
abide by the Given–New Contract is a matter of debate (see Saylor et al. 2006). Yet
young children are apparently aware of what information is new (or at least relevant)
for their interlocutor, and they exploit this knowledge for successful communication
(see discussion and references in Clark 2009). O’Neill (1996) found that when two-
year-olds request a hidden object, they provide more information when the parent
36 CHAPTER 1. INTRODUCTION
was absent during the hiding event than when the parent was present (see also O’Neill
2005; Saylor et al. 2006). Also, research on children in the one-word stage suggests
that early one-word utterances express new information (e.g., Bates 1976; Greenfield
1979). Clark & Bernicot (2008) also argued that young children (aged 2;3 and 3;6)
are sensitive to the status of information in the discourse. They showed that two-
year-olds repeated new information from the previous utterance and signaled (via
this repetition) that they had attended to what was new and now regard it as shared
information (i.e., information in common ground). They also found that three-year-
olds, after repeating the new information, tended to add new information of their own
to the discourse (see also Allen 2000). In effect, these studies suggest that by age 2 or
3, children are capable of adjusting their utterances to better suit the needs of their
interlocutor. Thus, we might reasonably expect preschoolers to use given-before-new
ordering for the benefit of the addressee.
Another way to conceptualize an addressee-based account is to think of given-
before-new ordering as a strategy to convey a particular ‘perspective’ on an event.
As Clark (1990:1203) noted, “children must learn the form-meaning combinations
and also how each can be used to convey a specific perspective.” Consider again the
sentences in (1), repeated here:
(27) a. She gave the hat to the man.
b. She gave the man the hat.
While these two sentences might be used to describe the same event, they arguably
convey two different perspectives on the event. If asked, What did Rachel do with the
hat?, (27a) would be a more natural response than (27b). This is because (27a), which
highlights the fate of the hat, is more consistent with the addressee’s focus of atten-
tion. Previous work has shown that preschool children are able to convey multiple
perspectives on an event (e.g., Clark 1990; Clark 1997; Narasimhan & Gullberg 2006),
1.5. ACCOUNTING FOR GIVEN-BEFORE-NEW ORDERING 37
but the full range of linguistic devices used to convey perspective may not develop
until quite late (see Hornby 1971; Karmiloff-Smith 1980; Hickmann 1997; Hickmann
& Hendriks 1999; Arnold et al. 2007).
1.5.2 Speaker-based accounts
Alternatively, speakers might prefer given-before-new ordering, not for the sake of the
addressee, but because this sequence is easier to produce. Since language production
proceeds incrementally and speakers construct sentences on the fly (e.g., Levelt 1989;
Clark & Wasow 1998), speakers might delay the production of certain parts of a
sentence because those parts require extra time to plan. The idea is that given
information should take less time to plan (and produce) than new information and
should, therefore, be mentioned earlier. Two related notions are important here—
accessibility and length. Given referents are more accessible than new referents. In
other words, when a referent has just been mentioned, it is easier to retrieve from
memory (see Bock & Irwin 1980). Items that are retrieved faster can then be processed
faster, and thus, produced earlier, all the while buying time to plan the more difficult
parts (Bock & Irwin 1980; Bock & Warren 1985; Ferreira & Yoshita 2003). Given
referents also tend to be shorter than new referents (see, among others, Givón 1983;
Gundel et al. 1993; Arnold 1998; Ariel 2001). And shorter forms (e.g., pronouns) are
also generally mentioned before longer ones, again presumably because they require
less cognitive/articulatory effort or planning time on the part of the speaker (see
Hawkins 1994; Wasow 1997; Temperley 2007).
If given-before-new ordering is a speaker-based process, do speakers then prefer
this ordering because given information is more accessible, or because it is shorter,
or both? A few studies on adult production have tried to disentangle these highly
correlated features. Arnold et al. (2000) and Wasow (2002) argued that length and
accessibility both independently influence word order (see also Bresnan et al. 2007).
38 CHAPTER 1. INTRODUCTION
Arnold et al. (2000:50) concluded that “variation in constituent ordering cannot be
explained by a single factor, and that the speaker’s choice may be driven by different
considerations on different occasions, depending on which factors are strongest or
which needs are most pressing.” In general, speakers apparently prefer to save the
hardest parts of the sentence until the end, whether those bits are hardest because of
accessibility or length or some combination of these factors.
If adult given-before-new ordering is an effort to make production easier, we might
well expect to find similar effects in young children. Indeed, several researchers have
shown that there is continuity between the way children and adults process language
(e.g., Nadig & Sedivy 2002; Song & Fisher 2005; Song & Fisher 2007; Kidd et al.
2007; Bannard & Matthews 2008; McDaniel et al. 2010; Arnon 2010; de Marneffe
et al. Submitted).
1.5.3 Statistical-learning accounts
Yet another possibility is that children use given-before-new ordering because they
have learned this from the input. We know that distributional patterns in the input
play a crucial role in linguistic development (e.g., Saffran et al. 1996; Rowland et al.
2003; Diessel 2007; Smith et al. 2007). Children might observe, for example, that
postverbal arguments are generally arranged according to givenness. This seems
unlikely, though, because many verbs with two arguments never (or only rarely)
alternate, and for some of the alternating verbs, the alternation may be driven more
by semantic features than discourse ones (see §1.2). Overall, the data that provide
evidence for discourse-driven alternation of postverbal arguments are limited to a
relatively small set of verbs.
Alternatively, children might notice that subjects tend to refer to given infor-
mation (but see Arnold et al. 2007). They might then develop a general rule like
place given information as early as possible. In other words, children would notice
1.6. SUMMARY 39
this probabilistic tendency for subjects and generalize it broadly across grammatical
categories. As just noted, however, children receive considerable evidence from sen-
tences with multiple postverbal arguments that this rule does not hold, and any such
generalization might be quickly abandoned.
Some of the strongest evidence for distributional learning comes from findings that
child data mirror adult data on a verb-by-verb basis (e.g., de Villiers 1985; Naigles
& Hoff-Ginsberg 1998; Campbell & Tomasello 2001; Matthews et al. 2005; Chenu &
Jisa 2006). From this work, we might expect that children only use given-before-new
ordering with verbs that show sufficient evidence of this pattern in the input. Under
such an item-based account, verbs that do not alternate in the input would not be
expected to show variation in the child data.
Another way to learn given-before-new ordering from input might be to learn
(perhaps again on a verb-by-verb basis) that pronouns are often mentioned before
lexical nouns. If so, we should only see effects of givenness in the cases where one
postverbal argument is pronominal and the other is lexical. For the most part, these
distributional learning accounts would predict that older children will show more
effects of given-before-new ordering than younger children, because they have had
more experience with these patterns in the input.
1.6 Summary
In essence, this dissertation integrates two major lines of research—work on the acqui-
sition of argument structure and work on the acquisition of information structure—
and furthers our understanding of both by considering the relative order of postverbal
arguments in the speech of preschool children in contexts where both the discourse
context and situational context are highly controlled. Unlike previous studies, I also
carefully consider the interaction between word order and the form of the referring
40 CHAPTER 1. INTRODUCTION
expressions as well as other potential mediating effects like verb class and item-based
input patterns.
In Chapter 2, I report the results of two studies, one on preschooler’s use of
locative verbs, and a control study on adult usage. Although adults in this study
showed no tendency for given-before-new ordering, child syntactic choices were ro-
bustly influenced by the discourse. I also argue that type of referring expression,
semantic/conceptual biases, and patterns of distribution in the input are needed to
fully account for the children’s patterns of alternation.
In Chapter 3, I then turn to dative alternation verbs, reporting the results of an
elicited production study with four-year-old children and adults. In these studies, I
found that givenness influenced child and adult construction choices. Both groups
of speakers preferred to order given information before new information, though,
patterns in the input and/or semantic verb classes may have reduced this effect. The
child data also show particularly strong associations between givenness, word order,
and choice of referring expression.
In Chapter 4, I present a follow up study to this work on child datives, which
further assesses the roles of referring expressions, verb frequency, and semantic verb
class. Again, this study replicates the finding that givenness influences early syntactic
choices, but that a complex of other factors moderate this influence. This final study
also tests one of the underlying assumptions of the speaker-based explanation for
given-before-new ordering and suggests, preliminarily, that using given-before-new
ordering eases the processing load for preschool children.
Finally, Chapter 5 provides a summary of the major findings and and takes up
five major questions that I address throughout this dissertation:
(28) a. Do children use given-first or new-first ordering?
b. Are all verbs equally affected by givenness?
1.6. SUMMARY 41
c. Are all arguments equally affected by givenness?
d. Are givenness effects attributable to the type of referring expressions?
e. Why do children use given-before-new ordering?
In the end, this dissertation demonstrates that young children are able to incorporate
discourse cues into their production strategies, and it emphasizes the necessity of
taking multiple factors into account simultaneously in any study of child language
development.
Chapter 2
Study 1: Locative alternation
2.1 Introduction
Study 1 assesses the effects of discourse status on the realization of the locatum
and location arguments of locative verbs (see 29).
(29) a. She squirted the ketchup on the hotdog. (locatum-object variant)
b. She squirted the hotdog with ketchup. (location-object variant)
In particular, I address the following questions: (i) Do children (and adults) use given-
before-new ordering for these arguments? (ii) Can early argument structure errors
with non-alternating locative verbs be attributed to discourse pressure? and (iii) Do
age, verb frequency, and choice of referring expression moderate discourse effects?
I predict that children and adults will show sensitivity to discourse-givenness. I
expect this sensitivity to be manifested not only in their choice of referring expres-
sions (indefinite NP vs. definite NP vs. pronoun), but also in their choice of syntac-
tic constructions (locatum-object construction vs. location-object construction).
Specifically, I expect speakers of all ages to prefer given-before-new ordering; though
other pressures may, in some cases, override this preference. Adults should only show
42
2.2. STUDY 1A: CHILD LOCATIVES 43
given-before-new ordering with verbs that permit alternation. Similarly, if children
have learned that a particular verb never alternates (i.e., the verb’s argument struc-
ture is entrenched; see §1.3), givenness should not influence their syntactic choices
with that verb. But if children are unsure of a verb’s argument structure or have
learned that both constructions are acceptable, their patterns of alternation should
reflect the structure of the discourse (given-before-new). Thus, in the child data, error
patterns with non-alternating verbs and patterns of alternation with alternating verbs
should show given-before-new ordering. And as long as speakers use word order to
mark discourse status, I further predict that syntactic choices will align with choices
of referring expressions.
I first discuss the child data and then turn to the adult control study. Although the
methods for the child and adult studies were the same, I analyze the results separately
because of heterogeneity of variance in the two data sets. In the end, I argue that
preschoolers and adults show sensitivity to the discourse in their production choices.
In the child data, this sensitivity is reflected in both word order and choice of referring
expression. In the adult data, only referring expressions show discourse sensitivity.
Factors beyond discourse play a role in construction choice for both age groups, and
these factors likely nullified any potential for discourse-driven word order variation in
the adult data.
2.2 Study 1A: Child locatives
2.2.1 Methods
Participants. Participants for this study were 14 three-year-olds (2;10-3;11, M=3;6,
6 male/8 female) and 14 five-year-olds (4;11-5;7, M=5;3, 6 male/8 female). Nine
three-year-olds (2 female/7 male) and seven five-year-olds (3 female/4 male) were
44 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
tested but replaced because they refused to complete the task (n=5), or they refused
to respond to instructions (n=2), or because of experimenter error (n=9). All of the
children were tested in a small quiet room at their nursery school.
Materials. The primary stimuli were 18 videotaped vignettes depicting transfer
events. There were three vignettes for each of the six verbs in (30).
(30) a. Alternating verbs: stuff, squirt
b. Locatum-object verbs: drop, pour
c. Location-object verbs: cover, fill
These verbs were chosen based on (i) their occurrence in young children’s spontaneous
speech in CHILDES, (ii) their membership in a given semantic verb class, (iii) how
clearly they could be depicted in filmed vignettes, and (iv) their use in previous
studies (especially Gropen et al. 1989).1 Unfortunately, the verbs that best met these
criteria were not well balanced for input frequency. I discuss the potential effects of
frequency below.
For each verb, the manner and endstate of the event were held constant across
the three vignettes to ensure that syntactic alternations were due to discourse effects
and not semantic factors (see §1.2.1). In the vignettes for the locatum-object verbs
(drop, pour) and the alternating verb squirt, the locations were only partially
affected. In those for the location-object verbs (cover, fill) and the alternating
verb stuff, the locations were holistically affected. Also, within each verb type, the
events for one verb involved a mass noun locatum (viz. for squirt, pour, fill) and1While fill is clearly regarded as a non-alternating locatum-object verb in the literature, adults
occasionally use fill with locatum-objects. Lumsden (1994:90) noted that this can happen in “cer-
tain special contexts.” To my knowledge, every study of English locative alternation in child speech
mentions that children make “errors” with this verb by using it in locatum-object constructions.
A large-scale corpus analysis of fill in adult speech might require us to revise the “error” label.
2.2. STUDY 1A: CHILD LOCATIVES 45
Vignette 1 Vignette 2 Vignette 3
Verb Locatum Location Locatum Location Locatum Location
stuff skirt box shirt bag sweater basketsquirt soap sponge ketchup hotdog water towel
drop spoon plate crayon paper toothbrush towelpour juice glass tea cup milk bowl
cover shirt shoe blanket pillow towel flowerfill sand cup dirt jar rice bowl
Table 2.1: Study 1: List of locatums and locations.
those for the other verb involved a count noun locatum (viz. for stuff, drop, cover ;
see Table 2.1).
Each critical-item vignette involved one human actor and four inanimate objects:
the locatum, the location, and an alternative locatum and location. The
alternative items were included to encourage complete responses (see Gropen et al.
1989; Gropen et al. 1991b). The actual locatum and location referents were
matched for number of syllables (e.g., shoe and shirt; towel and flower), so that
differences in length of lexical NPs would not influence construction choice (cf. §1.5.2).
The locatums and locations depicted in the vignettes are listed in Table 2.1. In
addition to these, I also constructed six filler vignettes. Each of these depicted a
human actor engaged in an intransitive activity (sitting, hiding, or running). In
total, there were 24 vignettes, and no individual actor or item appeared in more than
one vignette. Prior to viewing the vignettes, children saw a corresponding photograph
depicting the actor seated behind the items, which were arranged in random order.
46 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
Procedure. Each participant saw all 24 vignettes. To control for potential order
effects, 14 lists were created (one per subject in each age group). Each list contained
a unique pseudo-random order of vignettes.2
Discourse context was manipulated via three discourse conditions: one where the
locatum was mentioned in the prompt, one where the location was mentioned,
and one where neither of the postverbal arguments were mentioned (see 31). The
assignment of discourse condition to the three vignettes for each verb was random,
and new assignments were made for each of the 14 lists.
Before showing a vignette, I (the experimenter) showed the participant the cor-
responding photograph, labeled the items in the photograph, and primed the target
verb. After the participant repeated the verb, I played the vignette, established the
discourse context, and elicited the event description as in (31).
(31) Example prompts for Study 1
Vignette: female (agent) squirts ketchup (locatum) on hotdog (location)
E(xperimenter): Look, here’s a girl, and here’s a tart, and here’s a hotdog, and
here’s some mustard, and here’s some ketchup. Now I’m going to show you
a movie about squirting. Say squirting! [[Participant: squirting]] Yeah!
Now let’s watch! [[E shows vignette]] . . .
a. Control condition (agent-given):
E: Point to the girl! [[Participant points]] Yeah, there she is! What’s
the girl doing in that movie?2The order of vignettes per list was random with the restriction that the first vignette was always
a filler, and subsequent fillers were separated by at least two critical items. Vignettes depicting the
same verb were separated by at least two other vignettes, and at least one instance of each verb
occurred in each half of the list.
2.2. STUDY 1A: CHILD LOCATIVES 47
b. Locatum-given condition:
E: Point to the ketchup! [[Participant points]] Yeah, there it is! What’s
the girl doing with the ketchup?
c. Location-given condition:
E: Point to the hotdog! [[Participant points]] Yeah, there it is! What’s
the girl doing with the hotdog?
The pre-elicitation labeling procedure ensured that the participants could identify
each locatum and location. This procedure was carried out first (before the verb
prime and presentation of the vignette) so that the labeling would have the least
possible influence on the discourse status of the items. I elicited descriptions for the
six filler films in the same way as the critical trials in the control condition. Overall
all, 12 trials highlighted an animate actor, and 12 highlighted an inanimate object.
Finally, when children used pronouns to refer to the postverbal arguments, I
checked up on the referent of that pronoun in cases where only one postverbal argu-
ment was given. For example, if a child said, She squirted it, I then asked, Squirted
what?. In other words, I never assumed a priori that pronouns referred to given
information. To avoid giving children the impression that their utterances involving
pronouns were uninterpretable, I asked the question as though I simply had trouble
hearing what they said (e.g., by putting my hand behind my ear). When children
gave one pronominal postverbal argument and one lexical one (e.g. She squirted it
with ketchup.), I did not check up on the pronoun but considered it the referent of
the other postverbal argument.
Design. The study used a 2 x 3 x 6 design with (three-year-olds, five-year-olds), dis-
course condition (control, locatum-given, location-given), and verb (cover, drop,
fill, pour, squirt, stuff ) as independent factors.
48 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
Coding. Audio recordings were made of the experimental sessions and transcribed
for analysis. In the analysis, I only included initial responses to the elicitation prompt
that contained the target verb and at least one postverbal argument.3 The primary
coding was based on which postverbal argument was realized as the direct object.
For example, forms as in (32) were coded as locatum-object responses and those as
in (33) as location-object responses.
(32) a. She’s pouring some tea inside the cup. (S31, 5;5, control condition)
b. Pouring tea. (S10, 3;10, location-given condition)
(33) a. She’s covering the shoe with the shirt. (S27, 5;5, control condition)
b. She’s covering the shoe. (S21, 3;10, locatum-given condition)
Also, only the locatum and location were counted as postverbal arguments. In
two cases (see 34), the child mentioned one of these arguments plus an instrument.
(34) a. She’s dropping it [=crayon=locatum] with her hands like this.
(S24, 5;4, locatum-given condition)
b. Squirting it [=towel=location] with a squirt gun.
(S26, 5;7, location-given condition)
These were coded as responses with one postverbal argument. Specifically, (34a) was
coded as a locatum-object response with no location argument, and (34b) was
coded as a location-object response with no locatum argument.
I also coded for the form of the postverbal arguments. Noun phrases with definite
determiners (e.g., the cup, the milk) were coded as definite, and those with indefinite3When children failed to use the target verb, I prompted them to answer again (e.g., Yeah, but
remember this one is with filling! Can we do this one again and use the word filling? ). Also, when
children only produced the verb (e.g., Squirting.), I followed up by asking Squirting what?. The
responses to these additional prompts are discussed in §2.2.2.
2.2. STUDY 1A: CHILD LOCATIVES 49
determiners (e.g., a cup, some milk) or no determiner (e.g., milk) were coded as
indefinite. Recall, though, that half of the locatums were mass nouns. Definite and
indefinite determiners with mass nouns can be used to indicate quantity, rather than
discourse status (see Schwartz-Norman 1976:284–285; Jeffries & Willis 1984:720). For
example, she poured the milk may mean that she poured all of the milk, and may be
felicitous even if milk was not mentioned in prior discourse. In each of the vignettes
with mass noun locatums, only part of the mass was transferred to the location.
Potential complications due to this aspect of the vignettes are discussed below.
Analysis. I analyzed the data using mixed-effect logistic regression models with
the software package R (Baayen 2008). These regression models are well-suited for
language research, because they allow for the simultaneous evaluation of multiple
variables and permit generalization over individual speakers and items through the
inclusion of random effects for participants and items (Quené & Van den Bergh 2004;
Baayen 2008; Baayen et al. 2008; Jaeger 2008). As with ordinary logistic regression
models, mixed logit models predict the probability of a binary response and are,
therefore, appropriate for studying linguistic choices that involve two alternative ex-
pressions. These models have been used effectively to predict adult behavior with
dative alternation (Bresnan et al. 2007), genitive alternation (Hinrichs & Szmrec-
sanyi 2007), and the presence or absence of complementizers (Jaeger & Wasow 2006).
Recent work in acquisition has also benefited from mixed-effects logistic regression
modeling, including corpus work on dative alternation (de Marneffe et al. Submit-
ted) and experimental work on conjunct ordering (Narasimhan & Dimroth 2008) and
relative clause production and comprehension (Arnon 2010).
50 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
2.2.2 Results
Because each child (n=28) in this study saw all 18 items, the maximum number of
child responses was 504. Of these, only 252 (50%) met the inclusion criteria (viz.
initial responses that involve the target verb and at least one postverbal argument).
Of the 262 excluded responses, 245 (94%) involved the target verb and at least one
postverbal argument, but they were not the child’s initial response to the prompt. A
representative example of these non-initial (or prompted) data is given in (35).
(35) E: . . . There it is! What’s the boy doing with the shirt? (locatum-given)
Child (S32, 5;2): He’s putting it in a little container.
E: Uh-huh, great! Can you tell me with the word stuffing what he’s doing with
the shirt?
Child: He’s stuffing the shirt into a container.
The fact that only half of the responses met the inclusion criterion is not surprising
given that most of the events in this study could be felicitously described with the
verb put, which is one of the most frequent verbs in child speech (Laakso & Smith
2007). The decision to exclude the non-initial data responses from the final analysis
was based on two factors. First, the initial response to the prompt often altered
the discourse context. As in (35), children generally mentioned one or both of the
postverbal arguments in their initial responses, which caused those elements to be
given information, regardless of the discourse condition. Second, the syntactic struc-
ture used in the child’s initial response could have caused the child to persist in using
that same structure for the subsequent response.
In addition to these non-initial data, four other responses were excluded because
the child never used the target verb (see 36). In these cases, the verb used in place
of the target verb was always put.
2.2. STUDY 1A: CHILD LOCATIVES 51
(36) E: Yeah, there she is! What’s she doing in that movie? (control condition)
Child (S14, 5;4): She’s putting clothes in a basket.
E: Yeah! Can you use the word stuffing to tell me what she’s doing?
Child: She’s doing something because she is. . . something. . . She’s packing up
and she’s putting something in a basket!
Finally, I also excluded responses with the form V + PP (see 37). Unlike V + NP
responses, these responses were very rare (n=3). Also, the analyses presented below
are based on which argument was realized as the direct object, but these data are not
easily categorized as locatum- or location-object responses.
(37) a. She was pouring in a cup. (S32, 5;2, control condition)
b. Pouring into a glass. (S10, 3;10, locatum-given condition)
c. Squirting with the ketchup. (S22, 2;10 location-given condition)
In the analyses below, I consider only the responses that met the inclusion criteria.
Word order. To test the effects of givenness on child locative constructions, I fitted
a regression model using construction type (locatum-object vs. location-object)
as the binary outcome variable and discourse condition (control vs. locatum-given
vs. location-given) and age (three vs. five) as the main predictor variables. I in-
cluded the logged trial number (1-24) and the number of postverbal arguments (1
vs. 2) as control variables, and participants and verbs as random effects. I also tested
for interactions between the variables and dropped non-significant interactions from
the model. Because age and the random effect for participants were not significant, I
removed them as well. The results of the regression are presented in Table 2.2. Posi-
tive coefficients indicate a preference for locatum-object constructions, and negative
coefficients a preference for location-object constructions. Overall, the model fit
was good (C-statistic=0.97, Dxy=0.94, R2=0.92).
52 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
Factors Coefficient SE Odds P-valueLocatum-given condition 2.6608 0.7723 14.3077 0.000571 ***Location-given condition 0.3135 0.5844 1.3682 0.591652 n.s.Order -1.7246 0.5554 0.1782 0.001903 **No. of arguments (=2) -4.3613 2.8064 0.0128 0.120176 n.s.Order x No. of args. (=2) 2.6546 1.0777 14.2193 0.013766 *
Table 2.2: Study 1A: Results of model predicting locatum-object constructions.
Crucially, there was a significant main effect of discourse condition. Locatum-
object constructions were significantly more likely when the locatum was given than
in the control condition. There was, however, no significant difference between the
location-given condition and the control condition. Recall also that there was no
significant effect of age nor any interaction between age and discourse. As expected,
both child age groups tended to realize given arguments as the first object: three-year-
olds and five-year-olds used more locatum-object constructions when the locatum
was given (Figure 2.1).
The pattern of responses with respect to the discourse condition (viz. more locatum-
object when the locatum was given) was also consistent across responses with one-
and two-postverbal arguments (see Table 2.3). Responses with two postverbal argu-
ments were generally more likely to have locatum-objects, which may be due (at
least in part) to patterns of usage in the input (see Table 2.4). Importantly, though,
there was no significant main effect of the number of arguments and no interaction
between the number of arguments and the discourse condition.
Additionally, there was a significant effect of order (i.e., the logged trial number).
As the negative coefficient for order reveals, locatum-object responses were less
likely later in the experiment. This may be a simple effect of the randomization
process; by chance, there were more trials with location-object verbs toward the
end of the sessions. There was also a significant interaction between order and number
2.2. STUDY 1A: CHILD LOCATIVES 53
condition
proportion
0.0
0.2
0.4
0.6
0.8
1.0
control
locatum
location
3 yrs
control
locatum
location
5 yrs
Figure 2.1: Study 1A: Proportion of locatum-object constructions by age.
Discourse condition
N. of args. Control locatum location Totalone-arg. 14/31 (45%) 21/36 (58%) 13/39 (33%) 48/106 (45%)two-args. 35/43 (81%) 47/50 (94%) 45/53 (85%) 127/146 (87%)Total 49/74 (66%) 68/86 (79%) 58/92 (63%) 175/252 (69%)
Table 2.3: Study 1A: Proportion of responses with locatum-objects.
54 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
of arguments. Order influenced responses with one postverbal argument (given-first),
but not those with two.
Finally, the verbs were not equally subject to discourse pressure (see Figure 2.2).
Alternations were mainly found with fill, squirt, and stuff. Pour, drop, and cover
rarely alternated and instead showed their required argument structure. Importantly,
these patterns do not align perfectly with verb type; the two location-object verbs
(cover, fill) differ markedly from each other. Children were at ceiling with cover, but
produced a high proportion of locatum-object responses with fill (cf. Kim et al.
1999). As predicted, these error patterns reflect discourse pressure: locatum-object
constructions with fill were produced primarily when the locatum was given.
condition
proportion
0.0
0.2
0.4
0.6
0.8
1.0
control
locatum
location
drop (locatum-obj)
control
locatum
location
pour (locatum-obj)
control
locatum
location
cover (location-obj)
stuff (alternating) squirt (alternating)
0.0
0.2
0.4
0.6
0.8
1.0fill (location-obj)
Figure 2.2: Study 1A: Proportion of locatum-object constructions by verb.
2.2. STUDY 1A: CHILD LOCATIVES 55
Word order and referring expressions. The data presented thus far do not
address whether locatum-givenness influences construction choice directly or indi-
rectly via the choice of referring expressions. As discussed in Chapter 1, an argument’s
givenness, length, and definiteness are often highly correlated. Compared to new ar-
guments, given arguments are more likely to be short and definite. Although the
stimuli for this study were designed so that lexical postverbal arguments would have
the same length in syllables (e.g., towel, flower), children often produced pronomi-
nal postverbal arguments (which are both short and definite). And when children
produced lexical postverbal arguments, those were sometimes definite and sometimes
indefinite. So can the effect of locatum-givenness be reduced to a pronoun-first
(short-first) and/or a definite-first strategy? Figure 2.3 gives the distribution of dif-
ferent types of referring expressions across given versus new arguments. Importantly,
givenness (given, new) and choice of referring expression (indefinite NP, definite NP,
pronoun) were highly correlated for locatums (Spearman’s r=0.57, p<0.0001) and
for locations (Spearman’s r=0.52, p<0.0001).4
4Because of the high correlations (collinearity) between givenness and choice of referring expres-
sion, building a statistical model with both factors is problematic. Such a model is also problematic
because the data set includes one- and two-argument responses. If we add the choice of referring
expression, then we must compare four “types” of referring expressions per postverbal argument:
indefinite NPs, definite NPs, pronouns, and non-overt arguments. But if one argument is non-overt,
then the other is always the direct object. So construction choice for that “type” of locatum or
location would be categorical, and categorical outcomes cannot be modeled in a probabilistic
analysis. We could instead recode the referring expressions, creating a factor with fewer levels (e.g.,
pronouns vs. others; definite expressions vs. others), but this would require grouping non-overt
arguments with another type (like lexical NPs or indefinite NPs). Given that argument ellipsis did
not pattern with the discourse, such a decision would be theoretically unmotivated. In this section,
I address the influence of referring expressions without appealing to a statistical model.
56 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
given locatums new locatums
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
given locations new locations
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
Figure 2.3: Study 1A: Distribution of referring expression types across given vs. newlocatums (above) and locations (below).
2.2. STUDY 1A: CHILD LOCATIVES 57
Altogether, the children produced 150 given arguments. Nearly all of these were
definite (either pronouns or definite NPs). Only 5% (n=7) were realized as indefinite
nouns (see 38).
(38) a. E: What’s he doing with the milk? (locatum-given condition)
Child (S34, 5;1): He’s pouring some milk into a bowl.
b. E: What’s she doing with the pillow? (location-given condition)
Child (S34, 5;1): She’s covering a pillow so she could hide it from us.
Of the seven indefinite NPs for given information, three referred to mass noun lo-
catums as in (38a). Such uses are not necessarily infelicitous, because only part
of the locatum was transferred to the container (see §2.1). Perhaps children used
indefinites in these cases to indicate the partial nature of the transfer.5 The only
truly infelicitous uses of referring expressions for given arguments were the other four
cases, all of which involved indefinite NPs for given locations as in (38b).
Of the discourse-new arguments (n=249), 28% were realized as indefinite NPs,
61% as definite NPs, and 11% as pronouns. But the child and experimenter were
jointly attending to the stimuli (so all referents were all physically co-present). Con-
sequently, uses of definite NPs for these arguments were pragmatically acceptable.
The only pragmatically infelicitous uses of referring expressions for new information
were the occasional pronominal ones (n=28).
As shown in Figure 2.4, indefinite lexical NPs were about as likely to occur in the
direct object position as definite ones. Definite and indefinite locatums occurred
in locatum-object constructions around 85% of the time. Definite and indefinite
locations occurred in location-object constructions around 30% of the time. This
suggests that NP definiteness in itself was not a strong determinant of construction5Even when the locatum was new, there were more indefinite NPs for mass locatums (n=33)
than for count locatums (n=12).
58 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
choice. This may seem surprising given the robust correlation between givenness and
type of referring expression, but recall that definite NPs were produced for given
arguments and for new ones. The same was true of indefinite NPs, though indefinite
NPs for new information were extremely rare (n=6, 2 locatums/4 locations).
If locatum-givenness influenced construction choice directly (and not just by
influencing the choice of referring expression), we might find that definite NP lo-
catums for given information were more likely to appear in locatum-object con-
structions than those for new information. This was the case, at least to some ex-
tent. Of the 13 definite NP locatums for given information, all but one occurred in
locatum-object constructions (93%). Of the 57 definite NPs for new information, 47
(82%) occurred in locatum-object constructions. Although the difference in these
proportions is small, it offers some preliminary evidence that givenness influenced
construction choice independent of referring expressions.6
Pronominal arguments, on the other hand, were more likely than lexical ones to
appear in the direct-object position (i.e., short-first). Of the pronominal locatums
(n=76), all but two appeared in locatum-object constructions. The two uses of
oblique pronominal locatums were both for the non-alternating location-object
verb cover in the locatum-given condition (see 39).
(39) a. E: What’s she doing with the shirt? (locatum-given condition)
Child (S11, 3;10): She’s covering her shoe with it.
b. E: What’s he doing with the towel? (locatum-given condition)
Child (S28, 5;3): He’s covering the flower with it.6This difference did not emerge for locations. Definite NP locations for given information
(n=33) occurred in location-object constructions 33% of the time, and those for new information
(n=95) occurred in location-object constructions 34% of the time. Recall that, unlike locatum-
givenness, location-givenness was not a significant predictor of construction choice.
2.2. STUDY 1A: CHILD LOCATIVES 59
indefinite NP locatums definite NP locatums pronominal locatums
proportion
0.0
0.2
0.4
0.6
0.8
1.0 locatum-obj.
location-obj.
indefinite NP locations definite NP locations pronominal locations
proportion
0.0
0.2
0.4
0.6
0.8
1.0 locatum-obj.
location-obj.
Figure 2.4: Study 1A: Proportion of locatum- vs. location-object constructionsfor each type of locatum (above) and location (below).
60 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
Although most pronominal locatums referred to given information, there were some
(n=22, 29%) that referred to new information. All of these occurred in locatum-
object constructions (see 40).
(40) a. E: What’s he doing in that movie? (control condition)
Child (S14, 3;6): He’s stuffing it into the box.
b. E: What’s he doing with the towel? (location-given condition)
Child (S31, 5;5): Dropping it on a towel.
c. E: What’s she doing with the hotdog? (location-given condition)
Child (S23, 3;5): Squirting it on the hotdog.
Thus, locatum-pronouns showed a very strong tendency to appear in the first
postverbal position, regardless of givenness.
But pronominality itself did not fully determine syntactic choice. Only about half
(56%) of the pronominal locations were produced in location-object responses.
Almost all pronominal locations referred to given information (88%). There were
only six uses of pronominal locations for new information (see 41)—all but one in
location-object constructions. None of the responses in the child data contained
two pronominal postverbal arguments.
(41) a. E: What’s he doing with the spoon? (locatum-given condition)
Child (S23, 3;5): He dropped the spoon to it.
b. E: What’s he doing in that movie? (control condition)
Child (S12, 3;3): Filling it with rice.
Importantly, when pronominal locations occurred in locatum-object construc-
tions (i.e., in second position), it was mostly with the non-alternating locatum-
object verbs drop (n=8) and pour (n=8), though they also occurred with the two alter-
nating verbs, stuff (n=2) and squirt (n=2), and once with fill. Pronominal locations
2.2. STUDY 1A: CHILD LOCATIVES 61
in location-object responses were mostly with the non-alternating location-object
verb cover (n=14), though also with fill (n=7), squirt (n=5) and stuff (n=1). In the
end, the distribution of locatum- and location-pronouns across constructions and
verbs suggests that there was, indeed, a strong bias to realize the pronoun as the di-
rect object, except in cases where this choice would have led to a highly dispreferred
construction (e.g., location-object constructions with drop and pour).
Another way to approach the question of whether the effect of locatum-givenness
is reducible to pronominality is to consider cases in which both postverbal arguments
were lexical (n=128). While a regression on these data shows no effect of discourse
condition, some of the same discourse-sensitive patterns emerge. As shown in Figure
2.5, two of the three verbs that showed discourse-driven argument realization patterns
(stuff and fill) still show more Locatum-object responses with given locatums (cf.
Figure 2.2). Ultimately, the significant effect of locatum-givenness is not entirely
attributable to the fact that given locatums were more likely to be realized as
pronouns. It seems likely that locatum-givenness and locatum-pronominality both
have some independent influence on construction choice, but there is too much overlap
between them in these data to fully verify this claim.
2.2.3 Discussion
This study provides strong evidence that discourse-givenness influences early word
order choices. But givenness did not influence word order across the board. Some
verbs were influenced by givenness, while others were not. Givenness of the locatum
affected construction choice, but givenness of the location did not. Also, choice of
referring expression seems to have influenced the effects of givenness. This section
addresses each of these issues in turn.
62 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
condition
proportion
0.0
0.2
0.4
0.6
0.8
1.0
control
locatum
location
drop (locatum-obj)
control
locatum
location
pour (locatum-obj)
control
locatum
location
cover (location-obj)
stuff (alternating) squirt (alternating)
0.0
0.2
0.4
0.6
0.8
1.0fill (location-obj)
Figure 2.5: Study 1A: Proportion of locatum-object constructions in responses withlexical postverbal arguments.
2.2. STUDY 1A: CHILD LOCATIVES 63
Differences among verbs. The two alternating verbs (stuff, squirt) and the location-
object verb fill showed discourse-driven alternation patterns, but the location-
object verb cover and the two locatum-object verbs (pour, drop) did not. Verb type
alone cannot account for this finding, because the two location-object verbs showed
such different behavior. The semantic properties of the events also fail to account
for these results. For alternating locative verbs, the locatum-object construction is
generally associated with a “partitive” interpretation, and the location-object con-
struction with a “holistic” interpretation (see §1.2.1). But in this study, the manner
and end-state of the events were held constant by verb (see §2.2.1). The locations
were fully affected in the vignettes for stuff (and for fill and cover), and partially af-
fected in the vignettes for squirt (and for pour and drop). Thus, degree of affectedness
does not account for the different verb-based patterns.
Another possibility is that the differences are due to children’s prior experience
with the verbs. To estimate how much experience children have with these verbs,
I conducted a small corpus study of uses of these verbs in speech to seven children
in the CHILDES Database (MacWhinney 2000).7 The results are presented in Table
2.4. Importantly, the three verbs that showed no alternation in the experimental child
data (cover, drop, pour) are more than twice as frequent in CDS as the verbs that
showed discourse-sensitive alternation patterns. Also, children rarely (if ever) hear
the infelicitous variant of these non-alternating verbs. The location-object verbs7These seven corpora were the same ones used in de Marneffe et al. (Submitted): Brown corpus
(Adam), Brown (Sarah), Clark (Shem), Demetras1 (Trevor), Kuczaj (Abe), Sachs (Naomi), Suppes
(Nina). I use these to provide a basis of comparison with the corpus results in Study 2, which are
partially taken from the data extracted by de Marneffe and her colleagues. I extracted the locative
verbs using the CLAN tool, searching only for tokens labeled as verbs (in the morphology tier).
Mislabeled nouns were removed from the data, as was one figurative use of cover (viz. Don’t ya think
that about covers it? Mother to Sarah).
64 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
Locatum-object Location-object
Verb Verb type One-arg. Two-args. One-arg. Two-args. Total
stuff alternating 0 0 1 1 2squirt alternating 0 1 16 2 19fill location-obj. 0 0 10 4 14cover location-obj. 0 0 43 5 48drop locatum-obj. 86 15 0 0 101pour locatum-obj. 33 24 1 0 58
Table 2.4: CHILDES: Distribution of verbs in parental speech to 7 children.
only occurred with location-objects. There was only one instance of a locatum-
object verb in what could be construed as a ‘V + location’ construction (viz.,
Pour one more glass. Father to Abe), but here, glass is almost certainly a measure
phrase denoting the locatum (as in a glass of milk) rather than an actual location
argument. Additional examples from the corpora are given in (42) and (43).
(42) Parental Locatum-object constructions in CHILDES
a. Pour your own milk. (Mother to Naomi)
b. He dropped some beans. (Mother to Adam)
c. Okay, could you squirt some lemon in my tea please? (Mother to Nina)
(43) Parental Location-object constructions in CHILDES
a. I’m protecting myself against your maniac mother; she wants to squirt
me. (Father to Abe)
b. Lie down, Nomi, and I’ll cover you with the blanket. (Father to Naomi)
c. Because I stuffed it with meat. (Mother to Abe)
Given the results from naturalistic CDS, we can reasonably assume that the chil-
dren in my study already had enough exposure to the three non-alternating verbs
(cover, drop, and pour) to learn that they resist alternation. In other words, the
2.2. STUDY 1A: CHILD LOCATIVES 65
argument structure of these verbs was entrenched (see §1.3), so children performed at
ceiling and avoided discourse-motivated errors. But with verbs that children have very
little exposure to (and which are, consequently, non-entrenched), discourse-pressure
was able to motivate alternations, whether the verb allows alternation in adult speech
(squirt, stuff ) or not (fill).
The frequency distributions of these verbs further suggest that the givenness effects
were not due to item-based statistical learning. Discourse-sensitive ordering emerged
with precisely those verbs for which there is very little evidence in the input. Of
course, this does not eliminate the possibility that statistical learning is driving the
effects. Perhaps children learn that given information is positioned earlier in general
(e.g., by observing patterns with subject choice), and they apply this principle in
ordering postverbal arguments—unless they have had enough experience with a verb
to know that it prohibits alternation. In other words, children apply global ordering
preferences in production until they learn item-specific patterns that conflict with the
global ones. I address other potential accounts of givenness effects below.
Differences in locatum- and location-givenness. Another pressing question
is why construction choice was influenced by locatum-givenness, but not location-
givenness. One explanation may lie in the selection of verbs in this study. The
most extreme effect of locatum-givenness was for the verb fill (see Figure 2.2).
This is a low-frequency location-object verb, and previous work has shown that
children tend to (erroneously) use this verb in locatum-object constructions (see
Bowerman 1982; Gropen 1989; Kim et al. 1999). Perhaps the children in my study
had some rudimentary understanding that fill prefers/requires location-objects, but
they still had a relatively unstable representation of this verb’s argument structure.
When discourse pushed for a locatum-object response, children were, therefore,
less able to inhibit such a response than in other discourse conditions. This would
66 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
account for the fact that children made errors with this verb around 90% of the time
when the locatum was given, and about 30% of the time otherwise. Perhaps word
order is most susceptible to discourse pressure when the discourse favors an argument
structure that children have not yet fully learned to suppress. The non-alternating
locatum-object verbs in this study are both high-frequency in CDS, and the children
in my study were at ceiling with these verbs. Thus, to give location-givenness a fair
chance, it would have been best to include a low-frequency (and potentially otherwise
error-prone) locatum-object verb.
Another possible explanation is related to a bias to attend more to locatums
than locations. There was a systematic asymmetry between the locatums and
locations in this study. The locatums were always in motion, while the loca-
tions remained stationary. This may have boosted the overall prominence of the
locatums. As Fisher (2000:16) noted, children show “a basic predisposition to at-
tend to first movers in causal events” (cf. Dowty 1991). Studies on the use of novel
verbs of transfer have also suggested that children are more likely to attend to mov-
ing objects rather than stationary ones (Gentner 1978; Gropen 1989; 1991a, 1991b).
Of course, for these transfer events, it would generally be impractical to move the
location during the event. Thus, this asymmetry in the stimuli was not simply an
artifact of the experimental design.
Although this locatum advantage was equally present in all of the experimental
stimuli (regardless of discourse condition), the boosted non-linguistic prominence of
the locatum may have interacted in important ways with the discourse. If children
were predisposed to attend to locatums, then the locatum-given condition pro-
vided a cue consistent with this predisposition (yes, the locatum is more prominent),
while the location-given condition provided a conflicting cue (no, the location
is more prominent). Thus, when the cues were aligned, discourse succeeded in influ-
encing word order, but when the cues conflicted, the potential effect of discourse was
2.2. STUDY 1A: CHILD LOCATIVES 67
cancelled out, resulting in syntactic choices that were not significantly different from
the baseline (see the discussion of “cue clash” in de Marneffe et al. Submitted). This
account yields the prediction that boosting the visual salience of the location over
the locatum would yield a more robust discourse-effect for location givenness (see
Gropen 1989; Gropen et al. 1991b for suggestive evidence in this direction).
The idea that locatums are generally more prominent than locations also
receives some support from the pronoun data. Children were generally more likely to
pronominalize the locatum than the location and to to use pronominal locatums
for discourse-new information (see Figure 2.3). To the extent that pronouns are
used for contextually prominent referents, and not just discourse prominent ones
(see §1.5), the (over)use of pronominal locatums compared to locations suggests
that children did, indeed, consider the locatums to be more prominent than the
locations. While further research is necessary to fully understand the difference
between locatum and location-givenness, a number of factors (both linguistic and
non-linguistic) likely conspire together to effect this asymmetry.
The role of referring expressions. There were strong associations between given-
ness and the choice of referring expressions. Pronouns were mostly used for given
information, and indefinite NPs mostly for new information. This is consistent with
previous findings that young children use referring expressions to mark discourse in
a target-like way, though their choices are not entirely error-free (see §1.4.1). For
the present purposes, the choice of referring expression is especially important as it
relates to word order. De Marneffe et al. (Submitted) found that the pronominality
of the postverbal arguments of the dative verbs give and show was highly predic-
tive of child construction choice (see also Bresnan et al. 2007; Bresnan & Nikitina
2009; Bresnan & Ford 2010; Anttila et al. In press). Also, Hickmann et al. (1996:610)
argued that, at least from age 7 on, givenness is either not marked at all, or marked
68 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
via both word order (‘global markings’) and referring expressions (‘local markings’).
In particular, Hickmann and her colleagues observed that discourse-new NPs tend to
occur in postverbal position (e.g., in presentative constructions like, There’s a cat.),
but only if the NP is also marked for newness (e.g., with an indefinite article).
In the present study, there was no compelling evidence that NP definiteness played
an independent role in construction choice. Definite and indefinite NPs were approxi-
mately equally likely to occur in the direct-object position, but definite NPs referring
to given information were more likely to occur in the direct-object position than those
referring to new information.
Pronominality may have influenced the results, though. Nearly all locatum-
pronouns immediately followed the verb, whether given or new. On the other hand,
pronouns were far more likely to be used for given referents than for new ones, and
the data involving only lexical postverbal arguments exhibited the same discourse-
sensitive alternation patterns, though locatum-givenness did not reach significance
in these data.8 While givenness may primarily influence word order by first influencing
the choice of referring expressions (by increasing the rate of pronominalization), this
study provides preliminary evidence that both givenness and pronominality have an
independent effect on construction choice, consistent with the findings of Arnold
et al. (2000) for adult construction choice (see also Wasow 2002; de Marneffe et al.
Submitted).
Summary. Study 1A indicates that three- and five-year-olds are sensitive to dis-
course status and that discourse influences the order of postverbal arguments either8The patterns in the data with lexical postverbal arguments also suggest that the difference
between locatum- and location-givenness is not attributable to preponderance of locatum-
pronouns over location-pronouns. As shown in Figure 2.5, the location-given condition still
appears to have no effect construction choice.
2.2. STUDY 1A: CHILD LOCATIVES 69
directly (given-first), or indirectly through the choice of referring expressions (short-
first). This effect is verb-dependent, and is apparently blocked when the verb’s ar-
gument structure is entrenched. Also, the effect of givenness on word order is likely
influenced by the extra-linguistic prominence of the referents. I have argued that
the location-given condition failed to influence word order because the discourse
prominence of the location was in conflict with the contextual/conceptual promi-
nence of locatum. In the locatum-given condition, though, these two sources of
prominence aligned and discourse had a robust effect on word order.
But what mechanisms underlie this effect? Is the effect driven by addressee needs,
speaker needs, and/or statistical learning? Discourse-sensitive ordering is probably
not learned on a verb-by-verb basis from the input. If it were, discourse effects should
only emerge with verbs that show robust evidence of given-first ordering in the input,
but the effects found in this study were with very infrequent verbs, one of which does
not generally alternate in adult speech. At present, though, we cannot rule out the
possibility that discourse-sensitive word order choices were learned more globally from
the input (e.g., by observing that subjects are generally more given than objects).
Abstracting away from this possibility, do the current data decide between an
addressee- or a hearer-based account? This question is particularly difficult to ad-
dress, because both the speaker and the addressee were jointly attending to the stim-
uli. In one sense, then, it should be impossible to decide because the speaker and the
addressee shared the exact same background knowledge. To fairly assess whether chil-
dren accommodate to the addressee’s knowledge state or their own, the two knowledge
states would need to be readily distinguishable.
This important caveat aside, one aspect of the current study argues in favor of a
speaker-based account. As discussed in Chapter 1, addressee-based accounts claim
that speakers use given-before-new ordering to facilitate comprehension. In this study,
though, responses with one postverbal argument showed the same discourse effect as
70 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
those with two. The Given-New Contract does not straightforwardly apply to cases
where the given referent is realized, but the new one is not. In these cases, children
did not wait to mention the new information until the addressee had a chance to
retrieve the given information (see §1.5.1). They simply never mentioned the new
information.
In fact, it is not intuitively clear whether producing ‘verb + given information’
or ‘verb + new information’ would better facilitate comprehension. Considering the
following:
(44) E: . . .What did the girl do with the ketchup?
a. She squirted it/the ketchup. (verb + given information)
b. She squirted a/the hotdog. (verb + new information)
Although (44b) may be more informative than (44a) (especially under conditions
where knowledge of the event is not shared), both responses are easily comprehensible
and relevant to the discourse. In essence, the addressee-based approach accounts for
the data with two postverbal arguments, but does not fare as well with the one-
argument data.
A speaker-based approach, however, could readily account for discourse sensitive
ordering in both subsets of the data. The claim from this account is that given entities
are more accessible in memory, and, therefore, processed faster and produced earlier.
If the child is only going to produce one postverbal argument, then choosing the one
that is most accessible should facilitate production in the same way as mentioning
that information first and then the new information. In sum, the nature of the
mechanism driving this effect is still open for debate, though there is strong evidence
against item-based statistical learning, and some evidence in favor of a speaker-based
account over an addressee-based one.
2.3. STUDY 1B: ADULT LOCATIVES 71
2.3 Study 1B: Adult locatives
2.3.1 Methods
The participants for this study were 14 adults (8 male/6 female), who were students
at or affiliates of Stanford University. The methods for the study were identical to
those in Study 1A, except that the auditory stimuli were not presented with a child-
directed speech style. Participants were tested individually in a quiet room on or near
the Stanford campus and received monetary compensation for their participation.
2.3.2 Results
Of a possible 252 adult responses (18 per participant), 238 (94%) met the inclusion
criteria of being initial responses with the target verb and at least one postverbal
argument. In a few cases (n=13), adults used a different verb in their initial responses.
As with in Study 1A, these responses were excluded from the analysis. One additional
response (viz. She’s pouring a glass of orange juice. S41, control condition) was
excluded because it involved a measure phrase as the locatum argument, which
contained some of the same lexical material (glass) as the location.
Word order. Unlike the child data, the adult data showed very different patterns
for responses with one postverbal argument compared to those with two. There were
only 13 adult responses with one postverbal argument (see 45).
(45) a. Squirting the towel. (S48, control condition)
b. Filling the jar. (S47, locatum-given condition)
c. The boy covered the flower. (S41, location-given condition)
These were all location-object constructions (mostly with location-object verbs).
They were fairly evenly distributed across the discourse conditions: five in the control
72 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
condition (with cover, fill, and squirt), five in the locatum-given condition (with
cover and fill), and three in the location-given condition (with cover and squirt).
Figure 2.6 shows the distribution of locatum- and location-object responses
across verbs and conditions for the data with two postverbal arguments. As ex-
pected, adult choices with the non-alternating verbs showed essentially no alter-
nation. Locatum-object verbs were always realized with locatum-objects. And
except for one use of fill with a locatum-object (viz. Filling it into the cup.
S38, locatum-given condition), location-object verbs were always realized with
location-objects.
condition
proportion
0.0
0.2
0.4
0.6
0.8
1.0
control
locatum
location
drop (locatum-obj)
control
locatum
location
pour (locatum-obj)
control
locatum
location
cover (location-obj)
stuff (alternating) squirt (alternating)
0.0
0.2
0.4
0.6
0.8
1.0fill (location-obj)
Figure 2.6: Study 1B: Proportion of locatum-object constructions in responses withtwo postverbal arguments.
2.3. STUDY 1B: ADULT LOCATIVES 73
Counter to expectations, however, there was very little evidence of alternation
with the two alternating verbs. There was only one use of stuff in a two-argument
location-object construction and two uses of squirt:
(46) a. So she just stuffed the purse with the shirt.
(S49, location-given condition)
b. He’s squirting the towel with water. (S42, control condition)
c. Squirting the towel with the water. (S50, locatum-given condition)
Otherwise, the alternating verbs pattern identically to the locatum-object verbs.
Clearly, discourse condition did not have a strong influence on adult’s syntactic
choices. Their choices were essentially fully determined by the verb.
Word order and referring expressions. Given that adults were tested with
the same procedure as children, one might wonder whether they found the task too
unnatural and, therefore, did not adapt their responses to the discourse. In other
words, perhaps they treated the task as if it did not involve participating in a coherent
conversation. If so, adult referring expressions, like word order, should show no
evidence of discourse sensitivity. But this was not the case. As in the child data,
givenness (given, new) and choice of referring expression (indefinite NP, definite NP,
pronoun) were highly correlated for locatums (Spearman’s r=0.55, p<0.0001) and
for locations (Spearman’s r=0.66, p<0.0001).
Like children, adults often used pronouns to mark discourse status. In fact, adults
used pronouns exclusively for given information (see Figure 2.7). Adults were also
more likely to use indefinite NPs for new information than for given information.
Overall, 93% of discourse-given referents were marked as given (either with definite
NPs or pronouns), compared to 68% of discourse-new referents (definite NPs only).
Of course, these definite NPs for discourse-new information were appropriate given
the context of joint attention.
74 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
The only arguably infelicitous referring expressions in the adult data were the
occasional indefinite NPs for given information (n=11). All of these were bare nouns
referring to mass noun locatums (see 47).
(47) a. E: What’s she doing with the milk? (locatum-given condition)
S38: Pouring milk into the bowl.
b. E: What’s she doing with the dirt? (locatum-given condition)
S43: She’s filling a jar with dirt.
Given the fact that only part of the mass of these locatums was transferred to the
location, these indefinite NPs for given information were not particularly unnatural
(see §2.1 and §2.2.2).
Thus, while adults did not use word order to mark discourse status, their use of
referring expressions reflected sensitivity to the discourse. In fact, the distributions of
referring expressions in the child and adult data were strikingly similar (cf. Figures
2.3 and 2.7). For both groups, given arguments (whether locatums or locations)
were mostly pronominal, new locatums were about as likely to be definite NPs as
indefinite ones, and new locations were mostly realized as definite NPs.
2.3.3 Discussion
The pressing question raised by the adult data is this: Why did the two verbs that
purportedly alternate in adult speech (stuff and squirt) show so little evidence of
alternation? The reason cannot be due to the partitive/holistic properties of the
event. In the vignettes for stuff, the location was holistically affected. In those
for squirt, it was only partially affected. If construction choices for stuff and squirt
were fully determined by these semantic properties, stuff should have patterned like
location-object verbs and squirt like locatum-object verbs (cf. Jeffries & Willis
1984). Instead, both patterned like locatum-object verbs.
2.3. STUDY 1B: ADULT LOCATIVES 75
given locatums new locatums
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
given locations new locations
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
Figure 2.7: Study 1B: Distribution of referring expression types across given vs. newlocatums (above) and locations (below).
76 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
This lack of alternation might be related to the extra-linguistic prominence of the
locatum (see §2.2.3). Perhaps the contextual salience of the moving locatums
clashed with the discourse prominence of the location in the location-given con-
dition, yielding construction choices no different from the baseline. For the adults,
though, the baseline construction choice was essentially at ceiling with a nearly cat-
egorical preference for locatum-object constructions, so there was no room for the
locatum-given condition to have an effect. In other words, the two discourse con-
ditions both failed, but for different reasons—one because of a cue clash and one
because of a ceiling effect.
But even if this is so, why did the adults have such a strong baseline preference
for locatum-object constructions? There are too few examples of stuff and squirt
in the CDS data presented above to provide insight into how adults use these verbs
in spontaneous speech (see Table 2.4). To check whether adult-adult speech reflects
the patterns found in Study 1B, I analyzed uses of stuff and squirt in the Associated
Press Worldstream segment of the English Gigaword Third Edition corpus (Graff et al.
2007). The corpus was parsed using the Stanford Parser (Klein & Manning 2003a,
2003b), and the parses were not hand-corrected. Verb uses were then extracted using
tgrep2 (Rohde 2001).9 Table 2.5 gives the frequencies of these two verbs with both
the locatum and location as postverbal arguments.10
Like the experimental data, these corpus data show a strong preference for locatum-
object constructions in sentences with both postverbal arguments. But this preference
is much more extreme in the experimental data (99% locatum-object constructions)
than in the corpus data (77% locatum-object constructions). Closer examination of
9I am very grateful to Uriel Cohen Priva for providing these data.10I excluded 16 examples (about golf) with the form stuff NP into X feet, because they prohibit
alternation (cf. Stuff a wedge into 4 feet. vs. *Stuff 4 feet with a wedge.). I thank Shawn Hughes for
this observation.
2.3. STUDY 1B: ADULT LOCATIVES 77
Verb Locatum-object Location-object Total
stuff 485 (76%) 150 (24%) 635squirt 56 (85%) 11 (16%) 67
Total 541 (77%) 161 (23%) 702
Table 2.5: APW: Distribution of verbs in the English Gigaword Third Edition corpus.
the corpus data suggests that the types of participants in the experimental materials
may be largely responsible for the adults’ syntactic choices. In the corpus results for
squirt, all the location-object responses, with one possible exception (viz. squirt-
ing them with ketchup and mustard) involved animate/body part locations. A few
examples are given below:
(48) a. squirt each other with yogurt
b. squirted the man’s face with mace-like spray
c. squirted a detainee with water in February
By contrast, all the locatums in the experimental stimuli were inanimate. Perhaps
adult speakers basically only use squirt in location-object constructions when the
location is human or perhaps otherwise particularly salient.
In the corpus data for stuff, location-object constructions also had a fairly lim-
ited set of locations. About 40% of them can be easily classified as animates/body
parts. An additional 23% of these constructions involved pronominal locations.
And animacy and pronominality aside, 37% of these sentences fit into one of four
categories:
(49) a. Stuff oneself with something (n=13):
stuffing themselves with unripe mangoes
b. Stuff someone’s mouth/face with something (n=12):
stuffing your face with the candy
78 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
c. Stuff someone’s pockets with something (n=14):
stuffing his pockets with wads of cash
d. Stuff a ballot box with something (n=16):
stuffing ballot boxes with votes for Kabila
Of course, some of this uniformity (especially the numerous examples with ballot
boxes) is likely due to the news genre of the corpus. But adults may generally prefer
locatum-object constructions with stuff, except with a limited set of event types or
fixed expressions.
Although the corpus results cannot fully explain why adults avoided location-
object constructions with stuff and squirt, they offer suggestive evidence that this
preference is more general, and not a by-product of the experimental setting. The
baseline construction choices of the adults may have been driven predominately by
global preferences for ordering the types of locatums and locations used in my
study (see Table 2.1).
One potential problem for this interpretation, though, is that Gropen found al-
ternations in adult uses of stuff with seemingly comparable stimuli (Gropen 1989).
Gropen showed adults two-panel picture sequences of a woman stuffing clothes into a
hamper or a boy stuffing toys into a box (see Gropen 1989:31). When locatives were
elicited with questions like those in (50), adults produced locatum-object responses
91% of the time and location-object sentences otherwise. But when they were
elicited with questions like those in (51), they produced locatum-object responses
only 56% of the time.11
(50) a. Locatum-given: What is the woman doing to the clothes?
b. Locatum-given: What is the boy doing to the toys?11Gropen’s picture sequences varied the degree of affectedness of the location (e.g. “half filled”
vs. “filled” vs. “bulging”), but the analyses ignored this distinction.
2.3. STUDY 1B: ADULT LOCATIVES 79
(51) a. Location-given: What is the woman doing to the hamper?
b. Location-given: What is the boy doing to the box?
Nevertheless, Gropen’s study differed from mine in a number of ways, and any one
of these might be responsible for the divergent results.
First, in my study, the PP in the question prompt was a with-phrase, not a to-
phrase (e.g., What was she doing with the ketchup? ). Highlighting the affectedness
of the given element by using a to-phrase could have driven the effect in Gropen’s
study. Second, in my study, a single item was stuffed into a comparably sized con-
tainer. The container’s larger size in Gropen’s study might have made it more salient
than (or at least as salient as) the locatums, encouraging more location-object
constructions. Third, the stimuli in my study were films, not picture sequences. So
participants saw the locatums in motion, which might have boosted their overall
salience, yielding a greater preference for locatum-object constructions (see §2.2.3).
And finally, Gropen’s participants were asked to produce the referent for the given
information before describing the event (e.g., Point to the box! . . . Say box! . . . Say
stuffing! . . .What is the boy doing to the box?). Perhaps saying the word immedi-
ately prior to producing the description made given-first responses more likely (e.g.,
by boosting their activation in memory).
At this stage, there are several plausible explanations for why the adult data in
my study showed no evidence for discourse-sensitive word order preferences, in spite
of discourse-sensitive choices of referring expressions. In all likelihood, the reason
has to do with the particular properties of the events or event participants, rather
than an unwillingness to use word order to mark discourse status generally. Whether
discourse can influence adults to produce one locative construction over the other
remains an open question.
80 CHAPTER 2. STUDY 1: LOCATIVE ALTERNATION
2.4 Summary
Study 1 confirms that preschool children and adults take account of the discourse in
formulating their utterances. In the child data, discourse sensitivity was manifested
both in the choice of referring expressions and syntactic constructions. In the adult
data, only referring expressions showed sensitivity to the discourse, but I have argued
that the null result for word order may be due to various properties of the target
events and not to a general predisposition against discourse sensitive ordering.
Except for the effects of givenness on word order, adult and child response pat-
terns were very similar. Both groups used more pronouns and definite NPs for given
information than for new information. For both groups, sentences with one postver-
bal argument were more likely to be location-object constructions, and those with
two were more likely to be locatum-object constructions (except for responses with
cover and fill). Also, for the three high-frequency verbs, children’s syntactic choices
looked just like adults’: Drop and pour were produced in locatum-object construc-
tions and cover in location-object constructions. For the three lower frequency
verbs, children, but not adults, were influenced by locatum-givenness: They were
more likely to realize given locatums as the direct object than new locatums.
These data clearly show that early errors with verbs like fill can be attributed to
discourse pressure (cf. Bowerman 1982).
Taken together, the results of this study show that preschool children are well on
their way to the target system with respect to locative alternation. And the points
of divergence in the child and adult responses highlight the importance of taking the
discourse context (as well as the extralinguistic context) into account when analyzing
child production data.
Chapter 3
Study 2: Dative alternation
3.1 Introduction
Study 1A revealed that children use discourse-sensitive ordering for postverbal ar-
guments. The extent to which these results generalize across different verbs, how-
ever, is unclear because Study 1 only considered locative verbs. Also, the verbs in
Study 1 were not well balanced for frequency, nor were the frequency differences
evenly distributed among semantic verb classes. There were two low-frequency al-
ternating verbs, one low-frequency non-alternating verb, and three high-frequency
non-alternating verbs. Study 2 addresses the following questions: (i) Do effects of
discourse on word order generalize to other verbs, and in particular, to alternating
dative verbs as in (52)? and (ii) How does givenness influence alternating verbs that
are high frequency in child-directed speech (CDS)?
(52) Dative alternation
a. She gave the hat to the man. (theme-first: prepositional dative variant)
b. She gave the man the hat. (recipient-first: double-object dative variant)
81
82 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
If discourse influences early word order only when children are unsure about the verb’s
argument structure, then discourse is unlikely to affect construction choices with high
frequency verbs. On the other hand, discourse might influence word order both when
children are less familiar with the verb (Study 1A) and when they know that the verb
permits two variants.
I predict that syntactic choices with highly frequent dative verbs will show given-
before-ordering. Like the locative verbs in Study 1A, though, some dative verbs may
show more susceptibility to discourse pressure than others. If so, I expect patterns
of argument realization in the input or semantic properties of the verb to underlie
this difference in sensitivity. Unlike in Study 1A, patterns of alternation should not
be attributable to child uncertainty regarding syntactic structure, because children
should have ample prior experience with these verbs. I also expect discourse to
influence both choice of referring expression and choice of syntactic construction.
Study 2 differs from Study 1 in that it involves an extra discourse condition, one
in which a non-event participant (viz. the location of the transfer event) was given
information. This condition was added to test whether the mention of additional
referents in the immediate discourse would influence argument structure choices, even
if children did not mention that element in their response (cf. Arnold & Griffin
2007). In the control condition in Study 1, only the agent was mentioned, but in
the two experimental conditions, one of the two postverbal arguments was mentioned
along with the agent. For Study 2, the location-given condition provides an
additional control condition, in which both theme and recipient are discourse-new,
but two elements (the location and the agent) are mentioned in the discourse.
This condition also serves to balance the number of animate entities (agents and
recipients) and inanimate entities (themes and locations) highlighted in the
discourse. Given that this condition does not directly affect the discourse status of
3.2. STUDY 2A: CHILD DATIVES 83
the postverbal arguments, I expect that responses in this condition will pattern like
those in the control condition (agent-given only).
For Study 1A, I tested two groups of children, three-year-olds and five-year-olds,
and both groups showed the same discourse patterns. For this study, I focused on four-
year-old children, in order to test a group of children with less linguistic experience
than five-year-olds, but who would be more likely than three-year-olds to produce
constructions with multiple postverbal arguments. I also varied the methods for the
child and adult versions of this study in hopes that the adult control study would
provide a more natural discourse situation for the participants. Ultimately, I argue
that children (and adults) indeed prefer given-before-new ordering with the postverbal
arguments of high-frequency dative verbs, although other factors again moderate the
role of discourse in construction choice.
3.2 Study 2A: Child datives
3.2.1 Methods
Participants. I collected data from 66 four-year-olds (3,10-5;4, M=4;8; 31 male/33
female).1 The data from two of the children were excluded, because they failed to
produce any responses with both post-verbal arguments.
Materials. This study tested four alternating dative verbs: give, read, show, and
throw. Compared to the verbs in Study 1, these verbs are very frequent in CDS. The
estimated token frequency of these verbs in the same seven CHILDES corpora used
1Five children were under 4;0, and 11 children were over 5;0.
84 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
in Study 1A (see §2.2.3) is as follows: give (n=988), read (n=348), show (n=321),
and throw (n=321).2
The stimuli for this study were five filmed vignettes: one warm-up vignette and
one vignette per dative verb. Each critical-item vignette involved one male actor,
one female actor, and one inanimate object. As in Study 1, the vignettes were silent,
no actor or item appeared in more than one vignette, and the referents for the two
postverbal arguments were matched for number of syllables; male actors were referred
to as boys and female actors as girls (see 53). Also, a photograph was made of each
actor and item individually.
(53) Description of vignettes in Study 2
a. Warm up: a male actor runs around a tree
b. Give: a female actor gives a hat to a male actor
c. Read: a female actor reads a book to a male actor
d. Show: a male actor shows a shirt to a female actor
e. Throw: a male actor throws a ball to a female actor
I used eight nursery rhymes as fillers. These were selected based on how familiar the
children were with them during pilot testing and the types of syntactic constructions
they contain. Most of the verbs in the rhymes were intransitive; none were dative.3
Design. This study used a 4 x 4 design with discourse condition (control, theme-
given, recipient-given, location-given), and verb (give, show, read, throw) as
independent factors.2I extracted these estimates with the CLAN command “freq,” which counted the number of these
forms marked as verbs (in the morphology tier).3The rhymes were these: The Itsy Bitsy Spider; Jack and Jill; It’s Raining, It’s Pouring; Mary
Had a Little Lamb; Humpty Dumpty; I’m a Little Tea Pot; Hey Diddle, Diddle; Twinkle, Twinkle
Little Star.
3.2. STUDY 2A: CHILD DATIVES 85
Procedure. Discourse context was manipulated via the four discourse conditions
mentioned above. Each participant saw all five vignettes. The order of the four critical
item vignettes and the pairing of the vignette with one of the four discourse conditions
were counterbalanced in a Latin squares design. The four orders of vignettes and
four pairings of vignette and discourse condition were then fully crossed, yielding
16 experimental lists. Each list included all four dative verbs, one per discourse
condition. These lists were circulated through the participant pool, so that each list
was completed by four participants.
There were two phases of the experiment, an initial labeling phase and a test
phase. For the labeling phase, I (the experimenter) introduced the task:
(54) This game has fun movies and pictures! First, I want to show you my pictures.
See if you know what these things are!
I then asked children to label a series of pictures: a picture of the tree from the
warm-up vignette, the four given elements from the dative vignettes (agent, theme,
recipient, location), the other three themes, the other three recipients, and
six filler pictures depicting scenes from nursery rhymes.4 This labeling phase took
2-3 minutes.
After the labeling phase, I announced that we were ready for the movies. Before
showing each vignette, I introduced the verb and asked the child to label the picture
of the given information, which was pasted inside a colorful piece of folded paper.
The child labeled the picture, and I ratified the label and began the prompt for the
vignette as shown in (55).4The pictures of the non-given themes and recipients were presented first, followed by the tree
and the four given elements, and finally the filler pictures. Within these three subgroups, the order of
the pictures was drawn randomly for each participant. This ordering minimized the chances that the
non-given themes and recipients would be considered given information during the subsequent
test phase.
86 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
(55) Example prompts for Study 2A
Vignette: a female (agent) gives a hat (theme) to a male (recipient)—the
scene takes place near a bush (location)
E(xperimenter): This one’s with giving! Say giving! [[Child: giving]]Great!
And . . .
a. Control condition (agent-given):
E: this one goes with the red picture! [[E and child view picture of agent]]
What’s that? [[Child: a girl!]] Yeah! Now let’s watch the movie about
her! [[E shows vignette]] Can you find the girl in that movie? [[Child
points]] Yeah, there she is! Now I’ll start the story about her! Once upon
a time there was a very nice girl. One day, the girl was outside, and
[[E shows vignette again]] what did she do?
b. Location-given condition:
E: this one goes with the green picture! [[E and child view picture of
location]] What’s that? [[Child: a bush! ]] Yeah! Now let’s watch the
movie about it! [[E shows vignette]] Can you find the bush in that
movie? [[Child points]] Yeah, there it is! Now I’ll start the story about
it! Once upon a time there was a green leafy bush. One day, a girl was
outside by the bush, and [[E shows vignette again]] what did she do?
c. Theme-given condition:
E: this one goes with the blue picture! [[E and child view picture of
theme]] What’s that? [[Child: a hat! ]] Yeah! Now let’s watch the movie
about it! [[E shows vignette]] Can you find the hat in that movie?
[[Child points]] Yeah, there it is! Now I’ll start the story about it! Once
upon a time there was a nice brown hat. One day, a girl was outside with
the hat, and [[E shows vignette again]] what did she do?
3.2. STUDY 2A: CHILD DATIVES 87
d. Recipient-given condition:
E: this one goes with the yellow picture! [[E and child view picture of
recipient]] What’s that? [[Child: a boy! ]] Yeah! Now let’s watch the
movie about him! [[E shows vignette]] Can you find the boy in that
movie? [[Child points]] Yeah, there he is! Now I’ll start the story about
him! Once upon a time there was a very nice boy. One day, a girl was
outside with the boy, and [[E shows vignette again]] what did she do?
Children in this study labeled the discourse-given information once and heard it
mentioned five times. The question children answered (i.e., What did he/she do?)
was the same across discourse conditions. Finally, between each vignette, children
randomly chose two nursery rhyme filler pictures (different from the filler pictures
shown during the labeling phase). Then we recited or sang the corresponding rhymes.
Coding and analysis. Audio and video recordings were made of all experimental
sessions and transcribed for analysis. The statistical analyses for this study were the
same as those used for Study 1 (see §2.2.1). And as in Study 1, I only included initial
responses to the elicitation prompt that contained the target verb and at least one
postverbal argument. The primary coding was based on the referent of the direct
object and the presence or absence of a recipient-PP. Overall, children gave four
types of responses: (i) ‘verb + theme’ (56a), (ii) ‘verb + recipient-PP’ (56b), (iii)
‘verb + theme + recipient-PP’ (56c), and (iv) ‘verb + recipient + theme’ (57).
Responses of the first three types were coded as prepositional datives, and those of
the last type were coded as double-object datives.
88 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
(56) Responses coded as prepositional datives
a. Read a book. (control condition; S11, 4;1)
b. Read to a little boy. (location-given condition; S43, 4;9)
c. She readed a book to the boy. (recipient-given condition; S13, 4;5)
(57) Responses coded as double-object datives
a. Give the boy a hat. (control condition; S03, 5;2)
b. Read him a book. (recipient-given condition; S50, 4;1)
c. He showed the lady a t-shirt. (recipient-given condition; S44, 5;0)
Only themes and recipients were counted as postverbal arguments. In one case,
the child apparently assigned the recipient role to the location when the loca-
tion was given:
(58) E: One day, a girl was outside by the bush and what did she do?
Child (S20, 3;10): Give it the hat.
This was the only response of this type, and I excluded it from the analysis.
As in Study 1, I also coded the form of the two postverbal arguments. I coded
NPs with definite determiners (e.g., the hat, her hat, that hat) as definite, and those
with an indefinite determiner (e.g., a hat) or no determiner (e.g., books) as indefinite.
Recipients of the form somebody or someone were also coded as indefinite NPs.
3.2.2 Results
Of 256 possible responses (4 per child), 172 (67%) met the inclusion criteria (viz. initial
responses that used the target verb and involved at least one postverbal argument).
All of these responses were included in the analysis except for the response in (58)
and one other response that failed to pattern like the others (see 59). This was the
only case in which the theme was realized in a prepositional phrase.
3.2. STUDY 2A: CHILD DATIVES 89
(59) E: One day, a lady sat down with the book and what did she do?
Child (S55, 5;4): Read the boy to the book.
Of the responses that failed to satisfy the inclusion criteria (n=82), 54 (64%) in-
volved a non-target verb, and 20 (24%) involved the target verb but no postverbal
arguments.5 In the 10 remaining cases, the child either declined to respond at all
(n=5), said, I don’t know (n=3), or only produced an NP (n=2 ; e.g., a shirt).
Importantly, a few factors led to categorical syntactic behavior. While these
results are important, they cannot be analyzed in a probabilistic model. Consequently,
I review these results here and exclude them from the statistical analyses below.
First, there were 55 responses that only contained one postverbal argument. These
responses were always prepositional datives, most involving the theme as in (60),
and a few involving the recipient as in (61).
(60) theme-only responses: ‘verb + theme’
a. Showing a shirt. (S12, 4;9, control condition)
b. She read the book. (S63, 4;11, location-given)
c. He throwed the ball. (S42, 4;6, recipient-given)
(61) recipient-only responses: ‘verb + to + recipient’
a. Read to a kid. (S02, 4;7, control condition)
b. She readed to him. (S04; 3;11, recipient-given condition)
One-argument responses were given for all of the verbs. For give, show and throw,
all one-argument responses were of the form in (60). For read, half the responses5Non-target verbs used in the initial responses were the following: (a) for give: hand, pass, put;
(b) for read: get; (c) for show: ask, bring, buy, get, give, hang, hide, look at, pull, put, shake, take,
talk, touch, wiggle; (d) for throw: catch, give, pass, play, stand, and toss. The action in the vignette
for show was clearly the most difficult one for children to interpret.
90 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Discourse condition
Verb Construction Control Location Theme Recip. Total
give V + theme 0 2 1 0 3
read V + theme 4 6 6 2 18V + to + recip. 3 1 0 5 9
show V + theme 2 1 0 2 5
throw V + theme 6 4 4 6 20
Total 15 14 11 15 55
Table 3.1: Study 2A: Distribution of responses with one postverbal argument.
involved a theme-object (see 60), and half a recipient-PP (see 61). Children never
produced constructions with a recipient-object and an elided theme (e.g., *She
read her.), which is generally illicit in adult English (but cf. He taught her.).6 The
distribution of child responses with one postverbal argument is given in Table 3.1.
Second, there were 28 two-argument uses of the verb throw, all of which were
realized in prepositional datives (e.g., He threw a ball to her.). While throw allows
alternation in adult English (cf. He threw her a ball.), children in this study never
used this verb with double-object datives (recipient-first). Finally, there were 36
two-argument responses in the theme-given condition, and all of these were realized
as prepositional datives (theme-first). I discuss these categorical effects further in
Section 3.2.3.
Figure 3.1 gives the proportion of prepositional datives by verb and discourse
condition for the responses with two postverbal arguments. For simplicity, I exclude
the data from the location-given condition from this figure. As discussed below,6Not all dative verbs are equally felicitous with just one postverbal argument. Children generally
avoided the same infelicitous one-argument responses that the adult control group avoided (see Table
3.3). Neither children nor adults used constructions like #show to her or #throw to her.
3.2. STUDY 2A: CHILD DATIVES 91
the location-given condition did not differ significantly from the control condition
or the theme-given condition. The theme-given and recipient-given conditions
consistently yielded the expected patterns, except with throw, which only occurred in
prepositional datives. Across the three verbs that showed alternation, children used
prepositional datives (theme-first) most often when the theme was given (100%),
second most when the location was given (94%), third most in the control condition
(73%), and least when the recipient was given (42%).
condition
proportion
0.20.40.60.81.0
control
theme
recipient
give
control
theme
recipient
read
show
0.20.40.60.81.0
throw
Figure 3.1: Study 2A: Proportion of prepositional datives.
92 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Word order. To test the effects of givenness on construction choice, I fitted a re-
gression model predicting the occurrence of prepositional datives (theme-first). Con-
struction type (prepositional datives vs. double-object datives) was the binary out-
come variable, and discourse condition (control vs. location-given vs. recipient-
given) was the primary predictor variable. I also included age and order (trial num-
ber: 1-4) as control variables, and participant and verb as random effects. Again,
this model excluded the three (partially overlapping) subsets of data that showed no
variation with respect to construction choice: (i) responses with only one postverbal
argument, (ii) responses with throw, and (iii) responses in the theme-given condi-
tion. Age, order, and the random effect for verb did not contribute significantly to
the model and were, therefore, removed. For age, I tried a model with age as a contin-
uous variable (measured in months) and one with age as a binary variable (younger
vs. older children). Neither measure of age was significant.
As expected, the regression showed no difference between the location-given
condition and the control condition (B=6.694, SE=7.707, p=0.385). At this point,
little can be said about the results in the location-given condition. Statistically,
this condition differed from neither the control condition nor the theme-given condi-
tion. The difference between the control condition and the recipient-given condition
just missed significance (B=-10.654, SE=5.669, eB=2.36e-05, p=0.060). When the
model was built to only include a comparison between the control condition and the
recipient-given condition, the difference was significant (B=-1.6337, SE=0.7067,
eB=0.1952, p=0.021). So theme-givenness had a categorical effect on construc-
tion choice (always yielding prepositional datives—theme-first), and recipient-
givenness had relatively small effect (encouraging more double-object datives).
Another way to think about the effects of givenness in these data is to consider the
distribution of different patterns of information flow across the two construction types.
Are prepositional datives and double-object datives equally likely to have a flow of
3.2. STUDY 2A: CHILD DATIVES 93
information that coheres with the discourse? Figure 3.2 shows the distribution of
the three information structure patterns evidenced in the data: (i) given-before-new,
(ii) new-before-given, and (iii) new-new (i.e., responses in the control condition and
the location-given condition). Though prepositional datives were more prevalent
in general, all of the responses with new-before-given ordering were prepositional
datives, and almost all the new-new responses were. The majority of double-object
datives had given-before-new ordering (67%) while just under 40% of the prepositional
datives did. Double-object datives (recipient-first) may, then, be better indicators
of information status than prepositional datives (theme-first), even if given themes
are more likely to occur in first position than given recipients.
prepositional datives double object datives
num
ber o
f res
pons
es
010
2030
4050
new-newgiven-before-newnew-before-given
Figure 3.2: Study 2A: Distribution of information structure patterns.
94 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Word order and referring expressions. As expected, givenness (given, new)
and choice of referring expression (indefinite NP, definite NP, pronoun) were highly
correlated for themes and recipients (for each: Spearman’s r=0.55, p<0.0001).
Figure 3.3 gives the distribution of referring expression types across given versus new
arguments in all of the analyzable data with two postverbal arguments. When an
argument was given (whether theme or recipient), that argument was realized as
a pronoun over 80% of the time. And with one exception (see 62), all other given
arguments were realized as definite NPs.
(62) E: A lady sat down with the book, and what did she do?
Child (S17, 5;2): Read a book to the boy.
Overall, about half of the new arguments were realized as definite NPs and the other
half as indefinite ones (ca. 45% each). Only 10% of all discourse new arguments were
pronominal.
Can the effects of givenness in these data be reduced to a pronoun-first (short-
first) and/or a definite-first preference? Figure 3.4 provides the distribution of the
different types of themes and recipients for the three verbs that showed alterna-
tion.7 Pronominal themes always occurred in prepositional datives (theme-first),
regardless of whether the theme was given (n=23) or new (n=6). There were only
five responses in which both postverbal arguments were pronominal (all prepositional
datives). The categorical effect of theme-pronominality reflects the fact that pronom-
inal themes are strongly dispreferred in double-object datives in American English
(cf. ??She gave him it. vs. She gave it to him.). The children in my sample may have
never heard pronominal themes in double-object datives.7The numbers in the remainder of this subsection are taken from this data set. Recall that all
responses with throw were prepositional datives (theme-first), regardless of the form of the referring
expressions.
3.2. STUDY 2A: CHILD DATIVES 95
given themes new themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
given recipients new recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
Figure 3.3: Study 2A: Distribution of referring expression types across given vs. newthemes (above) and recipients (below).
96 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
In general, pronominal recipients were only slightly more likely to occur in
double-object datives (recipient-first) than in prepositional ones. But here, we
should disregard responses with co-occurring pronominal themes, since theme-
pronominality had an overriding effect on construction choice. Among the responses
with lexical themes, there were 15 pronominal recipients for given information
and five for new information. Those for given information occurred in double-object
datives 80% of the time. Only one (20%) of the five pronominal recipients for
new information was in a double-object dative. Thus, when the theme was lexical
and the recipient was given, almost all (i.e., all but three) pronominal recipients
occurred in double-object datives.
As for lexical nouns, definite NP themes (n=25) occurred in prepositional datives
(theme-first) 76% of the time, while indefinite ones (n=34) occurred in prepositional
datives 56% of the time. Recall that most lexical themes referred to discourse-new
information. There were only three lexical themes for given information (2 definite,
1 indefinite). Like all given themes, these occurred in prepositional datives. Of the
definite lexical themes for new information (n=23), 74% occurred in prepositional
datives, compared to 54% of indefinite themes for new information (n=33). But if
we restrict the data set to responses in which the recipient was also lexical, then
86% of the definite themes for new information (n=14) occurred in prepositional
datives, compared to 74% of the indefinite ones (n=23). Perhaps theme-definiteness
had a small independent effect on choice of referring expression, but more data are
needed to substantiate this claim.
Definite and indefinite NP recipients were equally likely to occur in double-
object constructions (13% of the time). Like lexical themes, there were very few
lexical recipients for given information (n=6, all definite). For these six responses,
all of the co-occurring themes were also lexical. Two (33%) of these were double-
object datives. If we ignore the responses with co-occurring pronominal themes,
3.2. STUDY 2A: CHILD DATIVES 97
indefinite NP themes definite NP themes pronominal themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
indefinite NP recipients definite NP recipients pronominal recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
Figure 3.4: Study 2A: Proportion of prepositional (theme-first) vs. double-objectdatives (recipient-first) for each type of theme (above) and recipient (below).
98 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
there were 15 definite lexical recipients for new information and 18 indefinite ones—
20% of the definite ones occurred in double-object datives, as did 17% of the indefinite
ones. It seems unlikely that recipient-definiteness played any independent role in
construction choice.
Because nearly all given arguments were pronominal (and very few new ones
were), we cannot rule out the possibility that givenness influenced construction choice
indirectly via pronominality (short-first). The effects of givenness are not readily
attributable to NP definiteness, though, especially since there were so few cases in the
theme- and recipient-given conditions in which both postverbal arguments were
lexical. In fact, there were only two such cases in the theme-given condition (both
prepositional datives), and six in the recipient-given condition (67% prepositional
datives). Also, with all else held equal, there was little evidence that definite NPs
for new information were more likely to occur in the first postverbal position than
indefinite ones.
3.2.3 Discussion
This study corroborates the finding of Study 1A that discourse-givenness influences
the word order choices of preschool children. Crucially, the data in this study ex-
hibited a clear given-before-new ordering preference. When the theme was given,
children categorically produced prepositional datives (theme-first). When the re-
cipient was given, children were significantly more likely to produce double-object
datives (recipient-first) than in the control condition. Like Study 1, though, not
all verbs were equally affected by discourse pressure, and givenness of one of the
postverbal arguments was more influential than givenness of the other. Also, refer-
ring expressions may have played an important role in construction choice.
3.2. STUDY 2A: CHILD DATIVES 99
Differences among verbs. Unlike the other three verbs, throw only occurred in
prepositional datives (theme-first). This may seem surprising, because throw is
highly frequent in CDS and permits both constructions (cf. Throw her the ball. vs.
Throw the ball to her.). Verb semantics might explain why children avoided double-
object datives with throw. As mentioned in Chapter 1, the double-object construction
(recipient-first) is associated with the meaning cause to possess, while the preposi-
tional dative is often associated with the meaning cause to go. The throwing event in
this study involved a more long-distance and ballistic type of transfer than the other
events (see Pinker 1989; Rappaport Hovav & Levin 2008). Perhaps for this verb, the
cause to go meaning was so salient that the semantic preference outranked pressure
from the discourse to use a double-object construction (cf. Conwell & Demuth 2007).
In fact, according to Rappaport Hovav & Levin’s (2008) account of the dative
alternation, the verbs that alternated in my data (give, read, show) only have the
caused possession meaning and are predicted to alternate based on givenness (and
heaviness). But throw can have either the caused possession or caused motionmeaning
and is predicted to alternate primarily based on the meaning. If children interpreted
the throwing event as caused motion, the pattern of results is precisely what we would
expect given Rappaport Hovav & Levin’s predictions. The cause to go aspect of the
throwing event may have also been particularly salient for pragmatic reasons. In this
event, one actor threw a ball to another. By age four children have generally had
ample experience throwing and catching balls. The focus of such events is more on
the motion of the ball, and less on who happens to have possession of the ball in the
end. In essence, children may interpret these events as being more about the motion
of the theme and less about possession.
Another possibility is that children avoid double-object constructions with throw
because of patterns in the input. An investigation of CDS in the same childes
corpora used in Study 1 yielded over 300 uses of throw. But only 14 instances of
100 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
throw in the corpus data involved two postverbal arguments, all of which occurred in
prepositional datives (theme-first).8 A few examples of these are given below.
(63) a. No, throw it to me. (Mother to Adam)
b. You threw it to Mommy? (Mother to Nina)
c. Can you throw the balloon to me? (Mother to Nina)
d. Do you want me to throw the ball to you? (Father to Abe)
The other uses of throw mostly involved the sequence ‘throw + theme (+ directional
phrase)’ as in (64) (see Rappaport Hovav & Levin 2008:136). Many of them directed
the child not to throw things.
(64) a. Hey, you’re not supposed to throw books. (Mother to Abe)
b. Go ahead and throw it in the sink. (Mother to Abe)
c. We’ll throw snowballs at each other. (Mother to Naomi)
d. They were throwing hay on their back. (Mother to Nina)
This is the same syntactic construction as found with prepositional datives, and my
corpus search provided no evidence that children are exposed to double-object datives
with throw. Perhaps for children of this age, the ‘throw + theme’ pattern of argument
realization is entrenched. Throw was the only verb that failed to alternate in the input
(see Table 3.2).9 The present data do not decide between a semantic and an input-
based entrenchment account for the throw results. Also, semantic factors may well
be responsible for the distributional patterns in the input.8There were also 15 uses of throw with an at-PP (e.g., I’ll throw beer bottles at you; Father to
Trevor, 3;10). To be conservative, I omitted these prepositional uses. Including them would not
change the result that all uses of throw had the theme in the first postverbal position.9For the verbs give and show, the frequencies with two postverbal arguments and the proportion
of prepositional datives were calculated from de Marneffe et al.’s (Submitted) data.
3.2. STUDY 2A: CHILD DATIVES 101
Verb Prepositional datives Double-object datives Totalthrow 14 (100%) 0 (0%) 14read 63 (47%) 71 (53%) 134give 171 (26%) 477 (74%) 648show 46 (26%) 129 (74%) 175
Table 3.2: CHILDES: Distribution of verbs with both postverbal arguments inparental speech to 7 children.
Givenness and the role of referring expressions. Both theme-givenness and
recipient-givenness had a significant effect on construction choice, but the effect
of theme-givenness was categorical, while the effect of recipient-givenness was
gradient. In Chapter 2, I attributed the difference between locatum-and location-
givenness to differences in the non-linguistic prominence of those arguments. I argued
that the locatum, as the entity in motion, was more contextually salient than the
location. Thus, when the location was given, the contextual prominence of
the locatum conflicted with the linguistic prominence of the location. But this
explanation does not adequately account for the differences between theme-givenness
and recipient-givenness. Unlike the locatums, not all the themes in this study
were in motion. The theme changed physical locations in the vignettes for give
and throw, but not in those for show and read. Also, the recipients, by virtue of
their humanness, should have been at least as contextually prominent as any of the
themes, which were all inanimate.
The difference in theme- and recipient-givennes, however, may be due to an
asymmetry in how pronominal themes and recipients are realized. Pronomi-
nal themes are strongly dispreferred in final position in American English, while
pronominal recipients are perfectly acceptable there (see Green 1971; Erteschik-
Shir 1979; Collins 1995; Bresnan & Nikitina 2009; Anttila et al. In press, among
others):
102 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
(65) She read the boy a book.
a. ?? She read the boy it.
b. ?? She read him it.
(66) She read a book to the boy.
a. She read a book to him.
b. She read it to him.
In my data with two postverbal arguments (even including those for throw), there
were only six lexical themes for given information and seven lexical recipients for
given information. If children had already acquired the target-like restriction against
pronominal themes in double-object constructions, this would essentially ensure that
all the given themes appeared in prepositional object constructions (theme-first).
The fact that pronominal themes used for discourse new information also categor-
ically appeared in prepositional datives supports the idea that children, like adults,
avoid pronominal themes in double-object constructions, independent of the discourse
context. Because this restriction does not apply to pronominal recipients, recip-
ients referring to given information (even though mostly pronominal) were not as
likely a priori to be ordered first.
To what extent, though, can we attribute the effects of word order to pronominal-
ity? All pronominal themes were realized in prepositional datives, but so were all
given themes, whether pronominal or lexical. For the three verbs that showed alter-
nation, there were only three lexical themes for given information and 11 pronom-
inal themes for new information. Given the sparsity of the data in which theme-
givenness and theme-pronominality were dissociated, we cannot determine whether
theme-givenness has an independent effect on word order apart from influencing the
form of the referring expression.
3.2. STUDY 2A: CHILD DATIVES 103
Recipient-pronominality may have also been a driving force in the effect of the
recipient-given condition. Most given recipients were pronominal, and pronomi-
nal recipients were more likely to appear in double-object datives (recipient-first)
than in prepositional ones (theme-first). But in the data with lexical themes, this
was only true for pronominal recipients that referred to given information. Four
of the five pronominal recipients that referred to new information appeared in
prepositional datives. The data with two lexical postverbal arguments also hint at
an independent influence of recipient-givenness, but again there were too few cases
to say for sure. As with theme-givenness, then, we cannot fully assess whether
recipient-givenness influences word order directly or indirectly by influencing the
type of referring expression.
Summary. Study 2A reveals that four-year-olds prefer given-before-new order with
dative verbs, though some dative verbs (at least throw) apparently resist discourse-
sensitive ordering, perhaps because of semantic or pragmatic properties of the verb,
or because of how these verbs are used in the input, or both. Givenness of both
postverbal arguments influenced dative syntax, though theme-givenness had more
of an effect than recipient-givenness, likely because of asymmetric restrictions on
the placement of theme and recipient pronouns.
Givenness had a massive impact on the choice of referring expression. The ex-
tent to which givenness influences word order independent of pronominality remains
an open question. Interestingly, though, when children did use double-object con-
structions (recipient-first), these responses generally exhibited the most discourse-
coherent flow of information (i.e., given-before-new) (cf. Clifton & Frazier 2004;
Brown et al. Submitted). The same cannot be said for prepositional datives, though
there were more given-before-new than new-before-given orderings among these re-
sponses as well.
104 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Two important caveats are in order, though, with respect to the interpretation of
these data. First, children saw a picture of the discourse-given information prior to
viewing the vignette. In Study 1A, children also saw a picture prior to the vignette,
but the picture was of the entire scene, not just the discourse-given information.
This procedural change was intended to provide a natural context for labeling the
discourse-given information (and not the other items) prior to viewing the vignette,
thus increasing the child’s awareness that that element was, indeed, discourse-given.
But this manipulation also meant that the visual stimuli were not held constant across
the discourse conditions. Although children described the same event (i.e., vignette)
regardless of the discourse condition, they saw a different picture prior to viewing
the vignette, depending on the discourse condition. Consequently, we cannot rule
out the possibility that seeing the picture of the given information immediately prior
to viewing the vignette was responsible for the discourse effects. In other words,
the visual context (physical copresence) and not just discourse context (linguistic
copresence) might have affected the responses (see §1.5).
Second, children in this study were told that the vignette and the corresponding
story were about the discourse-given information (e.g., Now let’s watch a movie about
it! . . .Now I’ll start the story about it! ; see 55). Aboutness or topicality is strongly
associated with discourse-givenness, but the two need not overlap perfectly, as dis-
cussed in Section 1.5. Perhaps the effects of discourse were due more to the fact that
I highlighted the topicality of the given elements than simply that the elements were
mentioned in the discourse. These two design issues are remedied in Study 3.
3.3. STUDY 2B: ADULT DATIVES 105
3.3 Study 2B: Adult datives
3.3.1 Methods
Participants. I collected data from 28 adults (10 male/18 female). One participant
was replaced, because, as she reported afterward, she answered the questions as if I
had asked What happened? rather than the actual questions.
Materials. Adult participants saw the same critical item vignettes as children. There
were, however, four additional critical item vignettes for the adult study. These
involved the same four dative verbs as the other events. None of the actors or items
appeared multiple times in any of the vignettes, and each vignette had one male actor
and one female actor. A description of the additional vignettes is given in (67).
(67) Description of the additional critical-item vignettes in Study 2B
a. Give: a female actor gives a hat to a male actor
b. Read: a male actor reads a book to a female actor
c. Show: a male actor shows a coat to a female actor
d. Throw: a male actor throws a ball to a female actor
Eight filler vignettes were also included in this study. These were the intransitive
fillers from Study 1 and two additional vignettes (both with skipping). Like children,
adults also saw a picture of the given information before viewing the vignette. Adults
were also given an unrelated pencil and paper distractor task between each vignette.10
Procedural changes for adults. I adapted the methods of Study 2A so that
the adult task would seem more natural. At the beginning of the session, adult10For this task, participants saw the two words that make up a compound noun (e.g., DISH
WASHER) and were asked to choose which of three spelling options they preferred (e.g., dishwasher,
dish-washer, dish washer).
106 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
participants were told that they would be describing simple films and working on a
paper task between each film. Instead of introducing the verb and having the adult
repeat it, I gave the adults a list of the verbs (dative verbs and fillers) and asked them
to read the list out loud at the beginning of the session. I then told them that each
of their answers should involve one of the verbs on the list.
I also emphasized that the order of the films needed to be random. I asked
participants to randomly choose from a stack of cards, open the card and show me
the picture inside so that I could find the corresponding film. I specified which stack
they should draw from so that (i) the order of films would always be a critical item
followed by a filler, and (ii) the four vignettes tested in the child study would be tested
before the additional vignettes. The four vignettes from Study 2A were tested first
to ensure the validity of comparisons between the child and adult responses for the
first set of vignettes, in case the vignettes tested later in the study showed evidence
of priming or practice effects.
When adults drew a card and showed me the picture, I said (as though to myself),
Okay, the one with the X, where X was always the given information. I then pre-
sented the vignette. This method not only served the stated purpose of randomizing
the films, but it also ensured that adults, like children, saw a picture of the given in-
formation and heard a label for it prior to viewing the vignette. The prompt format
for adults is given below.
(68) Example prompts for Study 2B
Vignette: a female (agent) gives a hat (theme) to a male (recipient)—the
scene takes place near a bush (location)
[[E shows vignette]] . . .
a. Control condition (agent-given):
E: Here, the woman was outside, and what did she do?
3.3. STUDY 2B: ADULT DATIVES 107
b. Location-given condition:
E: Here, a woman was outside by the bush, and what did she do?
c. Theme-given condition:
E: Here, a woman was outside with the hat, and what did she do?
d. Recipient-given condition:
E: Here, a woman was outside with the man, and what did she do?
As in Study 1A, the agent was the only given referent in the control condition.
Because the experimenter labeled the picture of the agent prior to the prompt, the
agent was referred to with a definite NP (68a). In the other conditions, the first
mention of the agent was in the prompt itself, and hence indefinite (68b)-(68d).
After responding to each film, adults were asked to spend one minute working on
paper task. The four dative films seen by the children were always shown first, sepa-
rated by filler vignettes and presented in the random order chosen by the participant.
The other four dative films were presented in a second block with the same method.
3.3.2 Results
Of 224 possible responses (8 per participant), 214 met the inclusion criteria (viz. initial
responses that used the target verb and involved at least one postverbal argument).
All other responses were excluded from the analysis.
Some adult responses (n=58) only contained one postverbal argument. These
were very similar to those in the child data (cf. Tables 3.1 and 3.3). All one-argument
responses were prepositional datives (theme-first). With show and throw, the omit-
ted argument was always the recipient (e.g., He showed the jacket.). With read,
the omitted argument was sometimes the recipient (e.g., She read the book.) and
sometimes the theme (e.g., She read to the boy.). Also, adults never used one-
argument responses with give, and children rarely did. The only clear difference
108 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Discourse condition
Verb Construction Control Location Theme Recip. Total
read V + theme 0 2 2 0 4V + to + recip. 10 7 3 7 27
show V + theme 1 1 2 1 5
throw V + theme 6 5 5 6 22
Total 17 15 12 14 58
Table 3.3: Study 2B: Distribution of responses with one postverbal argument.
between the child and adult one-argument responses was that children preferred the
‘read + theme’ pattern over the ‘read + to + recipient’ pattern (67% vs. 33%),
while the opposite preference emerged in the adult data (13% vs. 87%).
Unlike the child data, though, there was variation in construction choice among the
two-argument adult responses in the theme-given condition and with the verb throw.
Consequently, these data were included in the statistical analyses below. Figure 3.5
gives the proportion of prepositional datives by verb and discourse condition in the
adult data (excluding the location-given condition, which did not differ from the
control condition).
Word order. To test the effects of givenness on construction choice, I fitted a regres-
sion model with construction type (prepositional dative vs. double-object dative) as
the binary outcome variable and discourse condition (control vs. location-given vs.
theme-given vs. recipient-given) as the primary predictor variable. I also included
order (trial number: 1-8) and video set (1 vs. 2) as control variables and random ef-
fects for participant and verb. Order, video set, and the random effect for participant
did not contribute significantly to the model and were, therefore, removed.
Table 3.4 gives the results of this model. As expected, the location-given con-
dition did not differ from the control condition. The negative coefficient for the
3.3. STUDY 2B: ADULT DATIVES 109
condition
prop
ortio
n N
P P
P
0.2
0.4
0.6
0.8
1.0
control
theme
recipient
give
control
theme
recipient
read
show
0.2
0.4
0.6
0.8
1.0throw
Figure 3.5: Study 2B: Proportion of prepositional datives in responses with twopostverbal arguments.
110 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Factors Coefficient SE Odds P-value
theme-given condition 2.6464 0.5714 7.4120 0.000456 ***location-given condition -0.1015 0.5725 0.9035 0.859271 n.s.recipient-given condition -0.6146 0.5944 0.5409 0.301131 n.s.
Table 3.4: Study 2B: Results of model predicting prepositional datives in responseswith two postverbal arguments.
recipient-given condition reveals that prepositional datives (theme-first) were less
likely when the recipient was given than in the control condition, but this difference
was not significant. But there was a significant effect of the theme-given condition.
When the theme was given, participants were over seven times more likely to use a
prepositional dative (theme-first) than in the control condition. The fit of the model
was good (C-statistic=0.83, Dxy=0.67, R2=0.97).
Curiously, the distribution of different patterns of information flow across the two
construction types was quite different from that in the child data (cf. Figures 3.2
and 3.6). New-new (i.e., control condition) responses were more likely to be double-
object datives (recipient-first) than prepositional ones (theme-first), while given-
before-new and new-before-given patterns were basically equally common in the two
construction types.11 In these data, only 31% of double-object datives involved given-
before-new ordering (compared to 67% in the child data); 46% of prepositional datives
had given-before-new ordering (similar to the 39% in the child data). Unlike children,
adults apparently did not reserve double-object datives for discourse-consistent word
orders. This difference is likely due to asymmetries in how adults and children referred
to the recipient, which I discuss below.11The preponderance of new-new responses among double-object datives is mostly due to the data
for give and show. For read and throw, double-object datives were rare, regardless of the pattern of
information flow.
3.3. STUDY 2B: ADULT DATIVES 111
prepositional datives double object datives
num
ber o
f res
pons
es
010
2030
4050
new-newgiven-before-newnew-before-given
Figure 3.6: Study 2B: Distribution of information structure patterns.
112 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Word order and referring expressions. As in the child data, givenness (given,
new) and choice of referring expression (indefinite NP, definite NP, pronoun) were
significantly correlated for both themes (Spearman’s r=0.35, p<0.0001) and recip-
ients (Spearman’s r=0.53, p<0.0001). Figure 3.7 shows the distribution of referring
expression types across given versus new arguments.12 Half of the given themes were
realized as pronouns. With one exception (see 69), the rest were realized as definite
NPs.
(69) E: Here, a woman was outside with the hat, and what did she do?
S26: Gave him a hat.
When the recipient was given, all recipients were pronominal. Curiously, among
the recipients for new information, 38% were pronominal and all but two of the
lexical ones were definite. Thus, adult speakers almost categorically avoided indefinite
recipients.
In the regression model presented in Table 3.4, theme-givenness had a significant
effect on construction choice (more theme-first responses). There were also more
recipient-first responses when the recipient was given, but this effect did not
reach significance. Are these findings attributable to choices of referring expressions?
Figure 3.8 gives distributions of different types of themes and recipients across
the two constructions.
There were only two responses with indefinite NP recipients, both in the control
condition—one in a prepositional dative (70a) and one in a double-object dative (70b).
(70) a. Read a book to a little kid. (S12)
b. Give a man a hat. (S21)12For one response (viz. She gave the man [xxx] hat. S27, theme-given condition), the definiteness
of the theme was not determinable because of noise in the recording. This response is excluded for
the graphs and analyses that distinguish between definite and indefinite themes.
3.3. STUDY 2B: ADULT DATIVES 113
given themes new themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
given recipients new recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
Figure 3.7: Study 2B: Distribution of referring expression types across given vs. newthemes (above) and recipients (below).
114 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
indefinite NP themes definite NP themes pronominal themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
indefinite NP recipients definite NP recipients pronominal recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
Figure 3.8: Study 2B: Proportion of prepositional (theme-first) vs. double-objectdatives (recipient-first) for each type of theme (above) and recipient (below).
3.3. STUDY 2B: ADULT DATIVES 115
Adults occasionally produced indefinite NP themes (n=32), but only one was for
given information (see 69). Definite and indefinite NP themes referring to new
information were essentially equally likely to occur in prepositional datives (31% vs.
28%, respectively). Definite themes for given information were more likely to occur
in prepositional datives (53%) than those for new information (31%). Thus, these
data provide no convincing evidence that NP definiteness played an independent role
in construction choice. If anything, it seems that theme-givenness played a role over
and above the choice of referring expressions.
But are the effects of givenness attributable (at least in part) to pronominality
(short-first)? As in the child data, pronominal themes always occurred in preposi-
tional datives (theme-first). In the adult data, all pronominal themes referred to
given information, and there were only three responses in which both the theme and
the recipient were pronominal (see 71).
(71) a. He threw it to her. (S08)
b. He was showing it to her. (S08)
c. I guess he showed it to her. (S12)
Although recipient-pronominality did not have a categorical effect on construc-
tion choice, recipient-pronouns were much more likely to occur in double-object
datives (77%) than in prepositional ones (23%). But only 46% of adults’ pronom-
inal recipients referred to given information. If recipient-givenness played a
role over and above the choice of referring expression, then given pronominal re-
cipients should have been more likely than new ones to occur in double-object
datives (recipient-first). This apparently was not the case. In the data with lexical
themes, 74% of pronominal recipients for given information were in double-object
datives (recipient-first), compared to 86% of those for new information.
116 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Factors Coefficient SE Odds P-value
Location-given & recip. NP -0.4486 0.7656 0.6385 0.55793 n.s.Theme-given & recip. NP 3.1410 1.2349 23.127 0.01098 *Control & recip. pro -4.0564 1.3810 0.0173 0.00331 **Location-given & recip. pro -2.3074 1.0024 0.0995 0.02134 *Theme-given & recip. pro -3.6292 1.3964 0.0265 0.00935 **Recipient-given & recip. pro -2.0050 0.8103 0.1347 0.01335 *
Table 3.5: Study 2B: Results of model predicting prepositional datives in responseswith two postverbal arguments and lexical themes.
Given the distribution of referring expressions in the adult data, we can ask two
questions: (i) Were lexical themes for given information significantly more likely to
appear in prepositional datives than those for new information? and (ii) In the data
with lexical themes, did discourse condition play a significant role in the position
of pronominal recipients? To address these questions, I created a new factor by
crossing the levels of discourse condition (n=4) with the levels of recipient pronom-
inality (n=2), omitting the empty cell for lexical recipients in the recipient-given
condition. This yielded a seven-level factor, and I fitted a regression model compar-
ing the level with the control condition and lexical recipients to each of the other
levels, which are listed in Table 3.5. I ran this regression only on the subset of the
data containing lexical themes. I also included video set and trial order as control
variables and participant and verb as random effects. As before, video set, trial order,
and the random effect were not significant in the model and were, therefore, removed.
The results of final model are presented in Table 3.5, and the model fit was good
(C-statistic=0.90, Dxy=0.81, R2=0.86).
As expected, there was no significant difference between responses in the con-
trol condition with lexical recipients (the comparison level) and responses in the
location-given condition with lexical recipients. There was, however, a significant
3.3. STUDY 2B: ADULT DATIVES 117
difference between responses in the comparison level and those in the theme-given
condition with lexical recipients. When both postverbal arguments were lexical,
prepositional datives (theme-first) were over 20 times more likely when the theme
was given than in the control condition. In other words, theme-givenness signifi-
cantly affected construction choice even when both postverbal arguments were lexi-
cal. So the significant effect of theme-givenness is not fully attributable to theme-
pronominality.
All the levels involving recipient-pronouns were significantly different from the
comparison level. Regardless of the discourse condition, prepositional datives (theme-
first) were significantly less likely when the recipient was pronominal than when
it was lexical and in the control condition (hence the negative coefficients for these
levels). In other words, if the recipient was pronominal, it was significantly more
likely to be realized in double-object datives (recipient-first). There was no signifi-
cant difference between the four level in which the recipient was pronominal. This
means that pronominal recipients referring to discourse-new information were as
likely to appear in double-object datives as those referring to discourse-given informa-
tion. Thus, recipient-pronominality had an independent influence on construction
choice, but there is no evidence that recipient-givenness influenced construction
choice beyond encouraging pronominal recipients.
3.3.3 Discussion
Like the children in Study 2A, the adults in this study showed a reliable preference
for given-before-new ordering. And as in the child data, the effects of givenness were
not the same across the board. Some verbs seemed to be more affected by discourse
pressure than others. The effects of theme- and recipient-givenness were not
equally robust, and theme- and recipient-pronominality showed different patterns
across conditions and construction types.
118 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
Discourse condition
Verb Control Location Theme Recipient Total
give 14 14 13 12 53show 12 14 12 13 51read 4 4 8 5 21throw 8 7 8 8 31
Total 38 39 41 38 156
Table 3.6: Study 2B: Distribution of responses with two postverbal arguments.
Differences among verbs. All four verbs in the adult data showed some evidence of
discourse-sensitive alternation. But Figure 3.5 shows more extreme patterns of such
alternation for give and show than for read and throw. The difference among verbs
in the adult data may be due in part to an imbalance in the number of responses
with two postverbal arguments for each of these verbs. Responses with two postver-
bal arguments were more common for give and show than for read and throw. Table
3.6 provides the distribution of responses with two postverbal arguments across dis-
course conditions. Given the sparsity of the data for read and throw in each discourse
condition, the apparent differences in discourse effects may not be meaningful.13
Givenness and the role of referring expressions. Theme-givenness had a sig-
nificant effect on adult construction choice, even in the subset of data where both
postverbal arguments were lexical. Thus, the fact that pronominal themes al-
ways occurred in prepositional datives (theme-first) does not fully account for the
effect of theme-givenness. The influence of recipient-givenness is less straight-
forward. Recipients were generally more likely to occur in double-object datives13But if they are, the difference for throw might well be due to the same semantic/pragmatic
and or usage factors that likely influenced the child construction choices for throw (see §3.2.3). The
reason read should be less sensitive to the discourse than give and throw is unclear to me.
3.3. STUDY 2B: ADULT DATIVES 119
(recipient-first) in the recipient-given condition than in the control condition,
but not significantly so. The regression including recipient-pronominality provided
no evidence that recipient-givenness influenced construction choice over and above
the role it played in increasing (actually, maximizing) the amount of recipient-
pronominalization. Pronominal recipients were significantly more likely to appear
in double-object datives, but given pronominal recipients did not show a greater
tendency to appear in double-object constructions than new ones. And because all
given recipients were pronominal, we cannot assess whether given lexical recipi-
ents (like given lexical themes) would have been more likely to appear in double-
object datives than new ones.
The fact that theme-givenness had a more robust effect on construction choice
may well be related to the differential distribution of referring expressions for themes
and recipients. But the complexity of this distribution makes a fair comparison
of the effects of theme-givenness and recipient-givenness especially difficult. We
cannot compare the realization of given lexical themes and recipients, because
there were no given lexical recipients. There were given pronominal themes
and recipients, but pronominal themes are only felicitous in prepositional da-
tives (theme-first), while pronominal recipients can occur in either prepositional
datives or double-object datives (see §3.2.3). Thus, pronominal themes that refer to
given information are a priori more likely to appear in first position than pronominal
recipients that refer to given information. Finally, we cannot compare pronominal
themes and recipients that refer to new information, because adults only used
pronominal themes for given information (and because of the restriction against
theme pronouns in double-object datives).
Setting aside the comparison with theme-givenness, then, the question remains
as to why discourse-new recipient-pronouns were as likely to appear in double-
object datives as discourse-given ones. The reason may be simply that adults prefer
120 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
short-before-long ordering with datives, regardless of whether the short element is
given (cf. Arnold et al. 2000; Wasow 2002; Bresnan et al. 2007; Bresnan & Nikitina
2009; Bresnan & Ford 2010). If so, we might think of construction choice in terms of
the decision tree outlined below:
(72) Hypothetical decision tree for dative construction choice:
Step 1: Will I use a pronominal theme?
• If yes, strongly increase bias toward prepositional datives.
• If no, continue to Step 2. . .
Step 2: Will I use a pronominal recipient?
• If yes, increase bias toward double-object datives.
• If no, continue to Step 3. . .
Step 3: Is the theme (recipient) given?
• If yes, increase bias toward order that yields given-before-new ordering.
• If no, maintain baseline bias.
The current findings can be captured by this simple model, though the model would
likely need to be modified to account for more complex adult utterances produced in a
less controlled communicative setting (e.g., those with especially long or syntactically
complex arguments). But this type of model has the disadvantage that it essentially
ignores the role that givenness plays in determining the choice of referring expression.
Pronominal recipients for new information and those for given information pat-
terned the same, but why did adults use so many pronominal recipients for new
information in the first place? Recall that all recipients in my stimuli were hu-
man (and that the gender of the recipient was always different from the gender of
the agent). Perhaps recipient-animacy or humanness encouraged higher propor-
tions of recipient-pronouns. Themes (and locatums and locations in Study
3.3. STUDY 2B: ADULT DATIVES 121
1B) were all inanimate and were only pronominalized when they referred to given
information.
The absence of a significant difference between realization of given pronominal
recipients and new ones might be explained as follows: When adults treat an argu-
ment as though it were especially prominent (whether or not it was actually prominent
in the discourse), this influences both choice of referring expression and word order.
In other words, if adults decide to pronominalize an argument as though it were
given, they also treat it as though it were given in how they realize it syntactically.
Hence, any potential advantage of pronominal recipients in the recipient-given
condition is neutralized. Whether animacy or humanness or some other factor was
responsible for the initial choice to treat the argument as especially prominent is an
open question. The claim here is simply that this decision could result in choices to
mark the status of this referent both on the referring expression and in the syntax.
So there are two possible conclusions: (i) Recipient-pronominality affects con-
struction choice, but recipient-givenness does not independently influence construc-
tion choice, or (ii) When adults decide (for whatever reason) to treat a recipient
as though it were given, they mark this decision both in their choice of referring
expressions and in their choice of syntactic construction. Further research is neces-
sary to evaluate these explanations, and it may be impossible to fully rule out either
alternative.
Summary. Adults showed a robust preference for given-before-new ordering when
the theme was given, and for short-before-long ordering for both postverbal argu-
ments. Both types of constructions involved more given-before-new ordering than
new-before-given ordering, though double-object datives (recipient-first) had more
responses with new-new ordering than did prepositional datives (theme-first). This
differences is most likely due to the abundance of pronominal recipients when both
122 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
arguments were new. With themes, the short-before-long preference aligned per-
fectly with the given-before-new preference, because only given-themes were pronom-
inalized. With recipients, the short-before-long preference was evidenced regard-
less of the discourse condition. While discourse givenness alone does not provide a
straightforward account of recipient-ordering, we cannot rule out the possibility
that whatever drives adults to use pronominal recipients for new information also
drives them to order those arguments first.
3.4 Summary
Givenness influences preschooler’s choice of dative constructions, though it may do so
directly or indirectly. Responses in the theme-given condition were always preposi-
tional datives, and responses in the recipient-given condition were more likely to be
double-object datives. Whether these preferences reflect more of a given-before-new
preference or a short-before-long preference is difficult to determine because of the
nearly perfect alignment of givenness and pronominality. For adults, there is com-
pelling evidence that theme-givenness influences construction choice regardless of
the form of the theme. The case for recipient-givenness is less clear. If anything,
recipient-pronominality seemed to have a stronger influence on construction choice
than givenness per se, but we cannot rule out an alternative cause that simultaneously
influences both the referring expressions and the syntax.
Beyond showing sensitivity to the discourse, there were many other similarities
between the child and adult responses. For both groups, responses with one postver-
bal argument were always prepositional datives (theme-first) and the decision to
mention one or two postverbal arguments seemed to depend more on the verb than
on the discourse. Both groups also categorically avoided realizing theme pronouns
in double-object datives (recipient-first). And both groups were more likely to use
3.4. SUMMARY 123
throw in prepositional datives, though this preference was categorical for children and
gradient for adults.
Perhaps the most striking difference between the child and adult data was the
asymmetry in patterns of pronominalization. Children pronominalized themes and
recipients under the same conditions (i.e., givenness) and at essentially the same
rates (see Figure 3.3). Adults were less likely than children to pronominalize given
themes and more likely to pronominalize new recipients (cf. Figure 3.7). Ap-
parently children chose to pronominalize arguments based almost solely on discourse
status, whereas animacy or some other measure of non-linguistic prominence must
have influenced adult choices.
This difference might be due, at least in part, to task differences. Children heard
the discourse given entity mentioned six times and produced it once prior to describing
the vignette, but the adults never produced it and only heard it mentioned twice (once
when the picture was viewed and once in the question prompt). These procedural
differences were intended to make the adult task more age-appropriate, but they may
have resulted in the discourse context being less striking. If the animate/human
entities in the vignettes were already much more salient, and the discourse did not
sufficiently emphasize the prominence of a different entity in the adult task, this could
explain some of the asymmetry in adult patterns of pronominalization. Regardless,
the primary purpose of of the adult control group was to confirm that adults’ ordering
of postverbal arguments is influenced by givenness in a task similar to the one given
to children. This was indeed the case.
This study confirms that early discourse-driven ordering preferences for postverbal
arguments are not limited to locative verbs, but emerge with dative verbs as well. And
discourse sensitivity was found with verbs that are extremely frequent in CDS. Thus,
children show discourse sensitivity not only when they have not yet fully mastered
a verb’s argument structure (as I argued in Chapter 2), but also when they have
124 CHAPTER 3. STUDY 2: DATIVE ALTERNATION
had ample opportunity to learn that both variants are acceptable. Given that adult
choices of dative constructions also reflect discourse-sensitive ordering, it could be
that children use this ordering because they learned it from observing patterns in the
input, whether verb specific or not. The data are also consistent with both addressee-
and speaker-based explanations of this ordering preference. Children may have used
given-before-new (or short-first) ordering because they thought the addressee would
have an easier time understanding it, or because this order was easier for them to
produce (see §1.5). I return to the issue of why children use given-before-new ordering
in the next chapter. The next and final study also shifts the focus away from adult
comparisons and toward the relative importance of the factors responsible for child
construction choices.
Chapter 4
Study 3: Dative alternation revisited
4.1 Introduction
Study 3 was designed as a follow-up to Study 2A, primarily to address these questions:
(i) What role does semantic verb class (viz. caused motion verbs vs. caused possession)
play in influencing child construction choice? (ii) What role does verb frequency play?
and (iii) How do these factors interact with the influence of discourse givenness?
Another goal of this study is to ensure that the effects of givenness in Study 2A were
not due to either the differences in the visual scene across the discourse conditions or
the explicit references to aboutness (see §3.4). In Study 3, I did not show the children
a picture of the given information prior to the vignette, and I never said what the
vignette was about. Assuming that young children are sensitive to the discourse, I
expect to find given-before-new ordering despite the procedural changes.
There may, however, be complex interactions between discourse condition, se-
mantic verb class, and/or verb frequency. Thus far, I have argued that children use
discourse-sensitive ordering when unsure of a verb’s argument structure (Study 1A)
and when they know that the verb allows two alternative orderings (Study 2A). But
in Study 2A, the high-frequency verb throw only occurred in prepositional datives
125
126 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
(theme-first)—perhaps because children interpreted the throwing event as a caused
motion event, or perhaps because they had experienced ample evidence for ‘throw +
theme’ in the input, but not for ‘throw + recipient’. If children resist double-object
datives with caused motion events generally, they should avoid double-object datives
for such events even if they lack experience with the verbs themselves. So assuming
that children will interpret the events depicting caused motion verbs as caused mo-
tion events, then caused motion verbs should resist given-before-new ordering more
than the caused possession verbs, regardless of their frequency. But if the categorical
syntactic choices with throw in Study 2A were (instead) a result of prior experience
with this verb, children might use given-before-new ordering more with low-frequency
caused motion verbs than high-frequency ones. In other words, if verb-specific input
patterns moderate the availability of given-before-new ordering, then low-frequency
caused motion verbs should alternate (as in Study 1A), while high-frequency ones
resist alternation (like throw in Study 2A). For the caused possession verbs, though,
we might find given-before-new ordering with both high- and low-frequency verbs,
given that prior linguistic experience, semantic verb class, and discourse all seem to
support this choice.
With this study, I also return to the messy question of why children prefer given-
before-new ordering. As discussed in Chapter 1, this tendency could be driven by
addressee-based production strategies, speaker-based production strategies, or prefer-
ences learned from the input. Like Studies 1A and 2A, this study cannot definitively
answer this question. Here, I simply take up a small piece of the problem and con-
sider whether child production data corroborates the assumptions of speaker-based
accounts. These accounts argue that speakers prefer given-before-new ordering be-
cause this ordering is easier to produce: Given information is more accessible and can
be processed faster (by the speaker) and produced faster, buying time to plan the
4.1. INTRODUCTION 127
production of the new information (see §1.5.2). Does child language data yield any
evidence that using given-before-new ordering actually facilitates production?
Research on adult language has provided ample evidence that disfluencies like long
pauses, fillers, and restarts signal planning and production difficulties (see Goldman-
Eisler 1972; Levelt 1989; Smith & Clark 1993; Clark 1994; Clark & Wasow 1998;
Arnold et al. 2000; Fox Tree 2001; Clark & Fox Tree 2002; Gallo et al. 2008; Tily
et al. 2009, among others). For example, H. Clark and his colleagues have found
that adults use the filler uh to indicate a short but unavoidable delay in production
and um to indicate an imminent long delay (Smith & Clark 1993; Clark 1994; Clark
& Fox Tree 2002). Also, Tily et al. (2009) found that dative constructions in adult
speech corpora were significantly more likely to be disfluent if the speaker chose the
less favored variant in that context. For Tily et al.’s study, probability of a speaker’s
choice was determined by the model of the dative alternation in Bresnan et al. 2007,
which includes the givenness, length, and definiteness of the arguments, among other
factors (see also Arnold et al. 2000).
There is relatively little work on disfluencies in acquisition, but disfluencies indi-
cate planning and production difficulty in child language as well (see H. Levin et al.
1967; MacWhinney & Osser 1977; Rispoli 2003; Bannard & Matthews 2008; McDaniel
et al. 2010, and references therein). H. Levin et al. (1967) showed 24 school children
(aged 5-12) several physical demonstrations and elicited descriptions of varying con-
ceptual difficulty. For example, the experimenter mixed two colorless liquids together
to create a colorful one and asked the children these questions: (i) Tell me what you
see here. (ii) Watch and see what happens. What happened? and (iii) Why do you
think that happened? Children used more pauses, longer pauses, more hesitations,
128 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
and a slower speech rate when trying to explain the demonstration than when simply
describing the scene or event.1
In a larger study, McDaniel et al. (2010) tested two groups of children (aged 3;5-
5;9 and 6;1-8;10) and an adult control group in an elicited production task targeting
relative clauses of differing complexity (short and long subject relative clauses, and
short and long object relative clauses). Like adults, children were more disfluent
when producing more complex relative clauses. Also, children were generally more
disfluent than adults, and younger children were more disfluent than older children.
Recent work by Kidd and her colleagues (2009) also indicates that children younger
than three (2;4-2;8) have already learned that disfluencies (specifically, the filler uh)
are associated with production difficulty. When shown an image of a novel object
next to a familiar object, toddlers were faster to shift their gaze to the novel object
when the novel word was preceded by uh (e.g., Look at the uh mog! ) than when it
was produced in a fluent sentence (Look at the mog! ).
Together, these results suggest that preschool children should both produce more
disfluencies when experiencing production difficulty and that they understand that
fillers like uh signal some such difficulty. In light of these findings, I assess whether
disfluencies in four-year-olds’ dative constructions reveal greater production and plan-
ning difficulties with one pattern of argument ordering over another. If given-before-
new ordering is driven (at least in part) by speaker-based processes, I predict fewer
disfluencies in responses with given-before-new ordering than in other responses.1Variation in the data was not attributable to age, though there were too few children in each
age group (n=6) to draw firm conclusions about potential age effects. Also, H. Levin et al. (1967)
did not distinguish between different types of “hesitations,” which included filled pauses like um and
uh (see p. 561).
4.2. METHODS 129
4.2 Methods
Participants. I collected data from 26 four-year-olds (4;0-4;11, M=4;5; 13 male/12
female). Data from one child was excluded, because he only responded to questions
with I don’t know.
Materials. I tested the children on 12 alternating dative verbs. These verbs were
counterbalanced for semantic type (caused possession only vs. caused motion or
caused possession) and frequency (high vs. low) (see Table 4.1). The semantic verb
classes were chosen according to the classification in Levin (1993) (see also Rappa-
port Hovav & Levin 2008; Gropen et al. 1989). For convenience, I refer to the verbs
that only mean caused possession as “caused possession verbs” and those can mean
either caused motion or caused possession as “caused motion verbs.”2 Verb frequency
was determined by how many times the verb was used in child directed speech in the
seven CHILDES corpora used for Studies 1 and 2. Verbs with under 35 uses were
chosen as low-frequency verbs (range: 0-32), and verbs with over 300 uses were chosen
as high-frequency verbs (range: 321-1482).
The stimuli for this study were 13 filmed vignettes: one warm-up vignette and 12
critical-item vignettes, one per verb. Each dative vignette involved one male actor,
one female actor, and one inanimate object. As with Studies 1 and 2, the vignettes
were silent and no actor or item appeared in more than one vignette. The referents
for the themes were all monosyllabic, familiar words, and the position (right/left
side of the screen) and gender (male/female) of the agent and recipient were
counterbalanced both within and across verb classes. The vignette for read and the
warm-up vignette were the same as in Study 2; all other vignettes were new. Table 4.1
provides the list of dative verbs and brief descriptions of the corresponding vignettes.2These labels are further abbreviated as “possession” or “poss.” and “motion” or “mot.” in the
tables.
130 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
Semantic type Verb freq. Verb Agent Theme Recipient
possession
highgive male dress femaleread female book maleshow male star female
lowhand female cup maleloan female brush malepass male plate female
motion
highbring male plant femaletake female broom malethrow male box female
lowhurl female boot malekick male ball femaletoss female hat male
Table 4.1: Study 3: List of verbs and events.
Design. This study used a 3 x 2 x 2 x 12 design with discourse condition (con-
trol, theme-given, and recipient-given), semantic verb class (caused possession vs.
caused motion), verb frequency in CDS (high vs. low), and verb (bring, give, hand,
hurl, kick, loan, pass, read, show, take, throw, toss) as independent factors.
Procedure. Discourse context was manipulated via the three discourse conditions
mentioned above. Each participant saw all 13 vignettes. The order of the 12 critical
item vignettes was semi-random. After the warm-up vignette, children randomly
chose a card from one of two piles: one for the caused possession verbs and one for
the caused motion verbs. Half of the children started with a caused possession verb
and half with a caused motion verb. Children subsequently chose a card from each
pile, alternating between caused possession and caused motion verbs until (i) all 12
were presented, (ii) the session ran over time (15-20 minutes), or (iii) the child refused
to continue.
4.2. METHODS 131
The pairing of the vignette with one of the three discourse conditions was coun-
terbalanced across verbs, semantic verb class, and verb frequency. To achieve this,
I constructed three experimental lists. In each list, there were four verbs per dis-
course condition and two verbs from each semantic class in each condition, one high-
frequency and one low-frequency. These three lists were rotated through the partici-
pant pool, such that each child was presented with one list and each list was seen an
approximately equal number of times.
Unlike Study 2A, there was no initial labeling phase in this experiment. Pilot-
ing revealed that children already knew labels for all the themes and recipients.
Eliminating the labeling phase ensured that none of the discourse-new items were
labeled prior to the child’s description of the target event. Also, children did not see
a picture of the given information prior to viewing the vignette (as they did in Study
2), nor were they asked to point to it. These changes were intended to eliminate
the possibility that the discourse effects in Study 2 were due to the visual stimuli
(especially the pre-vignette picture), rather than the discourse context.
I (the experimenter) introduced the study by telling children that I had a game
with movies and cherries (the filler items). After the warm-up vignette and the first
cherry game, the child choose a card from one of the two piles mentioned above.
Sample prompts for this study are given below.
(73) Example prompts for Study 3
Vignette: a female (agent) tosses a hat (theme) to a male (recipient)
E(xperimenter): This one’s with tossing! Say tossing! [[Child: tossing]]
Great! And . . .
132 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
a. Control condition (agent-given):
E: this movie has a girl in it! Say girl! [[Child: Girl]] Yeah, now
let’s watch! [[E shows vignette]] I see the girl! Do you see her? [[Child
affirms]] Great! Now I’ll start the story! Once upon a time there was a
very nice girl. One day, the girl was outside, and [[E shows vignette
again]] what did she do?
b. Theme-given condition:
E: this movie has a hat in it! Say hat! [[Child: Hat]] Yeah, now let’s
watch! [[E shows vignette]] I see the hat! Do you see it? [[Child affirms]]
Great! Now I’ll start the story! Once upon a time there was a very nice
hat. One day, a girl was outside with the hat, and [[E shows vignette
again]] what did she do?
c. Recipient-given condition:
E: this movie has a boy in it! Say boy! [[Child: Boy]] Yeah, now let’s
watch! [[E shows vignette]] I see the boy! Do you see him? [[Child
affirms]] Great! Now I’ll start the story! Once upon a time there was a
very nice boy. One day, a girl was outside with the boy, and [[E shows
vignette again]] what did she do?
As in Study 2A, the question children answered was the same across discourse
conditions (viz. What did he/she do?). Aside from the differences mentioned above,
the only other change to the procedure in Study 2A was that children were not
explicitly told that the movie or story was “about” the given information (see 55).
Thus, the children in this study were not explicitly told that the given information
had a central role in the event.
Between each critical item, children played the cherry game. I hid three plastic
cherries under one of three identical, opaque dishes, had the children count, shuffled
4.2. METHODS 133
the dishes, and asked the children to find the cherries. After the initial hiding game,
the children chose whether to play the role of the finder or the hider.
Transcription and coding. The audio sessions were transcribed by a research
assistant who was blind to the experimental hypotheses. Because one of the goals
of this study was to analyze disfluency data, the research assistant was instructed
to transcribe the utterances exactly as the child said them, including fillers, restarts,
prolonged determiners, and extra long pauses. For prolonged determiners, she was
told to write two colons (::) after determiners that sounded especially long. For
pauses, she was asked to put a hash mark (#) in the transcript wherever she noticed
a pause within the utterance that sounded longer than usual.3 I then checked each
transcription and disagreements were resolved in consultation.
This method for coding pauses and lengthened determiners was subjective, but it
provided an efficient way to estimate production difficulties that have no correspond-
ing added lexical material (like um, uh, and restarts) (cf. Clark & Fox Tree 2002:81).
To provide a more precise description of the pause data, I subsequently measured (in
milliseconds) all the elements annotated as pauses using the freeware program Tran-
scriber 1.5.1 (Ingham & Reid 2001). For utterance initial filled pauses, I measured
from the onset of the filler to the onset of the following word. For utterance medial
unfilled and filled pauses, I measured the pause from the offset of the preceding word
to the onset of the following word (i.e., the pause duration including the filler).
Once the transcriptions were completed, I coded the utterances for verb produced,
number of postverbal arguments (1 vs. 2), construction type (prepositional dative
vs. double-object dative), type of referring expression for the postverbal arguments
(pronoun, definite NP, indefinite NP), and the categories of disfluencies mentioned3Long pauses prior to the onset of the utterance were not included. Many of these may have
been due to inattention, rather than to difficulty processing. Also, discriminating extra long pauses
before the utterance onset proved especially subjective.
134 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
above. As with Study 2, when the child spontaneously reformulated (self-corrected) a
response, the coding was based on the reformulated version, not the original response.
The coding criteria for construction type and referential expression were also the same
as in Study 2 (see §3.2.1). All responses coded were initial responses to the elicitation
prompt. Responses were included in the analysis as long as they involved (a) one of
the targeted dative verbs, even if it was not the verb primed for that particular item,
and (b) at least one postverbal argument.
4.3 Results
Of 300 possible responses (12 per child), 254 (85%) were produced. Children failed to
respond to all 12 elicitation questions either because we ran out time in the session
or (more commonly) because the child refused to finish the session. Of the 254
elicited responses, 225 (89%) met the criteria for inclusion in the analysis. Responses
that contained a non-dative verb or no postverbal arguments were excluded (n=29).4
Although children always repeated the target verb when prompted to do so, they
often replaced the target verb with another dative verb when describing the event.
For example, when prompted with the low-frequency verb loan (as in: This one’s
with loaning! Say loaning! ), most children used the verb give to describe the vignette.
Overall, give was the most common substitute verb. The verbs hand, pass, throw, and
toss were also used as substitutes for the target verb. Responses with substitute dative
verbs were included in the analysis. Table 4.2 shows the distribution of responses for
each verb.4The majority of these were excluded because they contained no postverbal arguments (n=19).
One was excluded because the child failed to respond at all, and 9 were excluded because they
contained a non-target verb (viz. put, share, look).
4.3. RESULTS 135
Verb producedTyp
e
Freq.
Target
Args.
Target give hand pass throw toss Total
caused
possession high
give 1 arg. 2 — 22 args. 14 — 3 17
read 1 arg. 7 72 args. 10 10
show 1 arg. 2 1 32 args. 12 1 1 14
low
hand 1 arg. 1 — 1 22 args. 2 13 — 2 1 18
loan 1 arg. 1 1 22 args. 12 3 15
pass 1 arg. 3 — 32 args. 2 12 — 1 15
Subtotal 51 44 1 10 0 2 108
caused
motion
high
bring 1 arg. 2 2 42 args. 15 1 16
take 1 arg. 1 1 22 args. 14 3 17
throw 1 arg. 6 — 62 args. 16 — 16
low
hurl 1 arg. 6 1 72 args. 13 13
kick 1 arg. 3 32 args. 13 1 14
toss 1 arg. 1 4 — 52 args. 2 12 — 14
Subtotal 43 32 0 5 36 1 117Total 94 76 1 15 36 3 225
Table 4.2: Study 3: Patterns of verb use and substitutions.
136 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
As in Study 2, all responses with one postverbal argument (n=46) were preposi-
tional datives. All but four of these responses involved the theme as in (74). The
remaining four (all with the verb read) involved the recipient as in (75).
(74) theme-only responses: ‘verb + theme’
a. He gave a plate. (control condition; S19 , 4;11)
b. Hand it. (recipient-given condition; S12, 4;8)
(75) recipient-only responses: ‘verb + to + recipient’
a. Readed to a boy. (control condition; S04, 4;8)
b. She read to the little boy. (theme-given condition; S19, 4;11)
One-argument responses were given for all the target verbs except take, which was
never produced with postverbal arguments. The distribution of responses with one
postverbal argument is given in Table 4.3 along with the proportion of one-argument
responses per verb. One-argument responses were distributed fairly evenly across
discourse conditions. Apparently, the primary factor contributing to a verb’s realiza-
tion with one versus two postverbal arguments was the verb itself. Read and bring
were among the most likely to occur with one-argument and give and pass among the
least likely (see Table 4.4). Because these one-argument responses show no variation
in construction choice (always prepositional datives), they were excluded from the
analyses that follow.
Of the 12 verbs elicited, only eight were produced in constructions with two
postverbal arguments. Two of them, hand (n=2) and pass (n=15), only appeared in
prepositional datives (theme-first). The distribution of responses with two postver-
bal arguments across conditions and constructions is given in Table 4.5. Because of
the imbalance in the frequency of use of each verb, I present these data in a table
with frequencies, rather than in the type of figures used in the previous studies. The
two verbs that showed no alternation were also excluded from the ensuing analyses.
4.3. RESULTS 137
Discourse condition
Verb Construction Control Theme Recipient
bring V + theme 2
give V + theme 3 5 3
hand V + theme 1
kick V + theme 2 1
pass V + theme 2
read V + theme 2 1V + to + recipient 2 1 1
show V + theme 1 1
throw V + theme 6 5 5
toss V + theme 1 1
Total 14 18 14
Table 4.3: Study 3: Distribution of responses with one postverbal argument.
Verb One-argument Two-arguments Total
bring 2 (100%) 0 (0%) 2give 11 (12%) 81 (88%) 92hand 1 (33%) 2 (67%) 3kick 3 (19%) 13 (81%) 16pass 2 (12%) 15 (88%) 17read 7 (41%) 10 (59%) 17show 2 (14%) 12 (86%) 14throw 16 (28%) 42 (72%) 58toss 2 (33%) 4 (67% ) 6
Total 46 (20%) 179 (80%) 225
Table 4.4: Study 3: Proportion of responses with one vs. two postverbal arguments.
138 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
Discourse condition
Type Verb Control Theme Recipient Total
poss.
give 20/26 (77%) 24/24 (100%) 16/31 (52%) 60/81(74%)read 2/2 (100%) 3/4 (75%) 3/4 (75%) 8/10 (80%)show 1/6 (17%) 3/3 (100%) 1/3(33%) 5/12 (42%)hand 0/0 2/2 (100%) 0/0 2/2 (100%)pass 5/5 (100%) 4/4 (100%) 6/6 (100%) 15/15 (100%)
Subtotal 28/39 (72%) 36/37 (97%) 26/44 (59%) 102/120 (85%)
mot.throw 14/14 (100%) 13/14 (93%) 14/14(100%) 41/42 (98%)kick 1/2 (50%) 5/5 (100%) 6/6 (100%) 12/13 (92%)toss 2/2 (100%) 1/1 (100%) 0/1 (0%) 3/4 (75%)
Subtotal 17/18 (94%) 19/20 (95%) 20/21 (95%) 56/59 (95%)
Total 45/57 (79%) 55/57 (96%) 46/65 (71%) 146/179 (82%)
Table 4.5: Study 3: Proportion of prepositional datives in responses with two postver-bal arguments.
Word order. Excluding the categorical data mentioned above (data with one
postverbal argument, and data with the verbs hand and pass), there were 162 re-
sponses across six verbs. To test the effects of givenness on construction choice
in these data, I fitted a logistic regression predicting the presence of prepositional
datives (theme-first). Syntactic construction (prepositional dative vs. double-object
dative) was the binary outcome variable, and discourse condition (control vs. theme-
given vs. recipient-given), verb frequency (high vs. low), and verb type (caused
possession vs. caused motion) were the main predictor variables. I also included age
and order (i.e., trial number: 1-12) as control variables, and participant and verb as
random effects. Verb frequency, age, and order did not contribute significantly to the
model and were, therefore, removed.5
5For age, I tried a model with age as a continuous variable (measured in months), and as a binary
variable (younger vs. older children). Neither was significant.
4.3. RESULTS 139
Factors Coefficient SE Odds P-value
theme-given condition 3.2741 1.2564 26.42 0.00916 **recipient-given condition -0.6687 0.7380 0.5124 0.36486 n.s.verb type=possession -4.5729 1.8072 0.0103 0.01139 *
Table 4.6: Study 3: Results of model predicting prepositional datives in responseswith two postverbal arguments.
The results of this model are given in Table 4.6, and the model fit was good
(C-statistic=0.98, Dxy=0.96, R2=0.93). Positive coefficients indicate a preference
for prepositional datives (theme-first) and negative coefficients for double-object
datives (recipient-first). As expected, there was a significant effect of theme-given
condition. When the theme was given, children were significantly more likely to use
a prepositional dative (theme-first) than in the control condition. In the recipient-
given condition, prepositional datives were less likely than in the control condition
(hence the negative coefficient for the recipient-given condition), but this difference
was not significant.
There was a significant effect of semantic verb type. Consistent with the analysis
of the dative alternation in Rappaport Hovav & Levin (2008) (see §3.2.3), preposi-
tional datives were significantly less likely with caused possession verbs than with
caused motion verbs. As shown in Figure 4.1, caused motion verbs were almost
exclusively realized in prepositional datives (about 95% of the time, regardless of
discourse condition). The proportion of prepositional datives with caused possession
verbs was more varied, ranging from 53% in the recipient-given condition to 98%
in the theme-given condition.
The different patterns shown in Figure 4.1 suggest an interaction between seman-
tic verb type and discourse condition. Discourse condition influenced construction
choice more for caused possession verbs than for caused motion verbs. Although
140 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
this interaction was not statistically significant (and was therefore removed from the
model), the significant difference between the control condition and the theme-given
condition was clearly driven by the data involving caused possession verbs.
condition
prop
ortio
n N
P P
P
0.0
0.2
0.4
0.6
0.8
1.0
control
theme
recipient
motion
control
theme
recipient
possession
Figure 4.1: Study 3: Proportion of prepositional datives.
Although this analysis does not show the expected effect of the recipient-given
condition on construction type (viz. significantly more double-object datives with
given recipients), the distribution of information structure patterns across con-
struction type is informative. Figure 4.2 gives the distribution of all the analyzable
responses with two postverbal arguments across construction types for three informa-
tion structure patterns: (i) given-before-new, (ii) new-before-given, and (iii) new-new.
The last category labels responses in the control condition, where neither postverbal
argument was discourse given.
4.3. RESULTS 141
prepositional datives double object datives
num
ber o
f res
pons
es
010
2030
4050
new-newgiven-before-newnew-before-given
Figure 4.2: Study 3: Distribution of information structure patterns.
142 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
For both prepositional datives (theme-first) and double-object datives (recipient-
first), the most common pattern was given-before-new. For prepositional datives, the
new-new (control condition) and new-before-given (recipient-given condition) re-
sponses were equally frequent. But for the double-object datives, while there were
some new-new responses, there were hardly any new-before-given (theme-given con-
dition) responses. This is consistent with the pattern found in Study 2A (cf. Figure
3.2). Again, the double-object construction seemed to be particularly resistant to
new-before-given ordering.
Word order and referring expression. As in the previous studies, givenness
(given, new) and choice of referring expression (indefinite NP, definite NP, pronoun)
were highly correlated for themes (Spearman’s r=0.55, p<0.0001) and recipients
(Spearman’s r=0.62, p<0.0001). Figure 4.3 gives the distribution of referring expres-
sion types across given versus new arguments in all of the analyzable two argument
data. Except for three indefinite themes for given information, all given arguments
were realized either as pronouns (57%) or definite NPs (41%). New arguments were
mostly realized as definite NPs (57%) or indefinite (38%) NPs. Only 6% of new
arguments were pronominal.
Do choices of referring expressions underlie children’s syntactic choices in these
data? Figure 4.4 gives the distributions of different types of arguments across the
two construction types for the verbs that alternated. These data closely replicate
the patterns found in Study 2A: All pronominal themes, whether given (n=34) or
new (n=3), occurred in prepositional datives (theme-first). Pronominal recipi-
ents were more likely to occur in double-object datives (recipient-first) than in
prepositional datives, but they were used in both. There were no responses in which
both postverbal arguments were pronominal. Most pronominal recipients referred
4.3. RESULTS 143
given themes new themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
given recipients new recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0
indefinite NPsdefinite NPspronouns
Figure 4.3: Study 3: Distribution of referring expression types across given vs. newthemes (above) and recipients (below).
144 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
to given information (n=29), though there were a few that referred to new infor-
mation as well (n=5). Pronominal recipients for given information occurred in
double-object datives 55% of the time, and all but one (80%) of the five pronominal
recipients for new information were in double-object datives.6
As for lexical nouns, definite NP themes (n=65) occurred in prepositional datives
89% of the time, and indefinite ones (n=60) 56% of the time. Of the definite NP
themes, 14 referred to given information and 51 to new information. All of those
referring to given information occurred in prepositional datives, compared to 86% of
those referring to new information. If the data are restricted to responses with lexical
recipients, 91% of definite NP themes referring to new information occurred in
prepositional datives. The only uses of indefinite NPs for given information were the
three responses in (76)-(78), two double object datives and one prepositional dative.
(76) E: . . .One day, a man was holding the box, and what did she do?
Child (S02, 4;3): He throwed the girl a box.
(77) E: . . .One day, a lady sat down with the book, and what did she do?
Child (S25, 4;9): Read someone a book.
(78) E: . . .One day, a girl was outside with the hat, and what did she do?
Child (S23, 4;3): Throw a hat to the man.
Aside from these three instances, all responses in the theme-given condition involved
prepositional datives (theme-first) with definite themes (lexical or pronominal).
Of the indefinite themes for new information, only 58% occurred in prepositional
datives (compared to 86% of definite NP themes for new information). If the data
are restricted to responses with lexical recipients, 76% of indefinite NP themes
6This is true in the data with lexical themes as well, since there were no responses with two
pronominal postverbal arguments.
4.3. RESULTS 145
indefinite NP themes definite NP themes pronominal themes
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
indefinite NP recipients definite NP recipients pronominal recipients
proportion
0.0
0.2
0.4
0.6
0.8
1.0 theme-first
recipient-first
Figure 4.4: Study 3: Proportion of prepositional (theme-first) vs. double-object da-tives (recipient-first) for each type of theme (above) and recipient (below).
146 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
referring to new information occurred in prepositional datives (compared to 91% of
definite NP themes for new information). So the data for lexical themes provide
some preliminary evidence of an independent influence of both givenness (viz. more
given definite NP themes in prepositional datives than new ones) and NP definiteness
(viz. more definite new themes in prepositional datives than indefinite ones), but
more data are needed to determine whether these differences are significant. As for
lexical recipients, even when we exclude data with pronominal themes, definite
NP recipients referring to given information (n=29) and those referring to new
information (n=43) were essentially equally likely to appear in double-object datives
(recipient-first) (10% vs. 14%, respectively).
Nevertheless, because there was more variation in choice of referring expression
for these data than for those in Study 2A, we can provide an initial assessment of
whether recipient-givenness and/or pronominality independently influenced con-
struction choice. As in Study 2B, I created a new factor by crossing the levels
of condition (n=3) by the levels of recipient pronominality (n=2), omitting the
empty cell for pronominal recipients in the theme-given condition. This yielded
a five-level factor, and I fitted a regression model comparing the level with the con-
trol condition and lexical recipients to each of the other levels, which are listed in
Table 4.7. In this analysis, I only included data with lexical themes. I included age
and trial order as control variables, and participant and verb as random effects. Age
and trial order were not significant and were, therefore, removed. This results in the
model presented in Table 4.7. Again, the fit of the model was good (C-statistic=0.99,
Dxy=0.97, R2=0.79).
This model shows that only the case in which the recipient was pronominal and
given was significantly different from the comparison level (responses in the control
condition with lexical recipients). In other words, while pronominality and given-
ness of the recipient influence construction choice in conjunction with each other,
4.3. RESULTS 147
Factors Coefficient SE Odds P-value
Recipient-given & recip. pro -2.2509 1.06043 0.1053 0.0338 *Control & recip. pro -4.1560 2.6668 0.0157 0.1191 n.s.Recipient-given & recip. NP 0.3012 1.0544 1.3513 0.7751 n.s.Theme-given & recip. NP 0.2747 1.2607 1.3161 0.8275 n.s.
Table 4.7: Study 3: Results of model predicting prepositional datives in responseswith two postverbal arguments and lexical themes.
there is no strong evidence in these data for an independent effect of either recipient-
givenness or recipient-pronominality. This model also fails to show a significant
difference between responses in the control condition with two lexical postverbal ar-
guments (76% prepositional datives) and those in the theme-given condition (88%
prepositional datives), though the trend is in the expected direction. Thus, the data
excluding pronouns altogether (i.e., responses with two lexical postverbal arguments)
fail to offer convincing evidence for an effect of givenness independent of the type of
referring expression. In fact, in all three discourse conditions, most of these responses
were prepositional datives (theme-first). But the expected pattern did emerge (at
least to some extent) for a few of the verbs. These patterns with each of the eight
verbs produced are given in Table 4.8 (cf. Table 4.5).
In sum, then, the current data replicate previous findings that theme-pronominality
perfectly determines construction choice. This is true whether the pronouns refer
to discourse-given themes or discourse-new ones. But pronouns for discourse new
themes are very rare. These data also provide a few clues that theme-givenness and
theme-definiteness may both independently influence construction choice, though
more data are needed to substantiate this claim. Finally, there is very little evidence
here that recipient-givenness or recipient-pronominality alone have a robust influ-
ence on construction choice, though recipients that are both given and pronominal
148 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
Discourse condition
Type Verb Control Theme Recipient Total
poss.
give 20/24 (83%) 6/6 (100%) 12/14 (86%) 38/44 (86%)read 2/2 (100%) 1/2 (50%) 1/1 (100%) 4/5 (80%)show 1/5 (20%) 2/2 (100%) 0/1(0%) 3/8 (38%)hand 0/0 0/0 0/0 0/0pass 3/3 (100%) 1/1 (100%) 2/2 (100%) 6/6 (100%)
Subtotal 26/34 (76%) 10/11 (91%) 15/18 (83%) 51/63 (81%)
mot.throw 13/13 (100%) 4/5 (80%) 9/9 (100%) 26/27 (96%)kick 0/0 2/2 (100%) 4/4 (100%) 6/6 (100%)toss 1/1 (100%) 0/0 0/0) 1/1 (100%)
Subtotal 14/14 (100%) 6/7 (86%) 13/13 (100%) 33/34 (97%)
Total 40/48 (83%) 16/18 (89%) 28/31 (90%) 84/97 (87%)
Table 4.8: Study 3: Proportion of prepositional datives in responses with two lexicalpostverbal arguments.
are significantly more likely to appear in double-object datives (recipient-first) than
those that are new and lexical.
Disfluencies. Of the 179 responses with two postverbal arguments, 58 (32%) were
coded as involving at least one disfluency. A few disfluencies are given in bold face in
(79), where a hash mark (#) represents a long pause and double colons (::) an extra
long vowel (also see 80-82).
(79) a. She give #[932ms] him a a a a a #[369ms] a broom.
(S02, 4;3, recipient-given)
b. She threw a boot to a:: man (S03, 4;8, control condition)
4.3. RESULTS 149
Overall, there were 42 responses with unfilled pauses, 15 responses with filled pauses,
25 responses with restarts, and three with prolonged determiners.7
I measured all the filled and unfilled pauses in these data. The frequency and
length of these pauses are given in Table 4.9.8 Pause length varied substantially both
within and across pause types. Recall that initial filled pauses were measured from
the filler onset to the onset of the following word, while utterance medial pauses
were measured from the offset of the previous word to the onset of the following
word. Because of this difference, I report the initial pauses separately in Table 4.9.
Regardless of position, filled pauses with uh were shorter (and less frequent) than
those with um.
Unfortunately, the literature does not provide a gold standard for counting child
pauses as speech disruptions. H. Levin et al. (1967) counted child pauses 80ms and
longer as disfluencies, whereas MacWhinney & Osser (1977) set the cut-off at 250ms,
McDaniel et al. (2010) at 500ms, and Rispoli (2003) at 1-3s (measured with a stop-
watch). All the unfilled pauses in my data were coded as such because the research
assistant who transcribed the data perceived them as “extra long.” None of these
pauses were as short as 80ms, which suggests that this perceptual measure is more
conservative than H. Levin et al.’s (1967) cut-off point. Seven of the unfilled pauses7In the majority of responses with restarts (76%), the restarts began prior to the onset of the
second postverbal argument. Six of the restarts were in double-object datives: three beginning at the
recipient, two at the theme, and one before either postverbal argument. The remaining responses
with restarts (n=19) were prepositional datives. In these, there were 21 restarts altogether: four
beginning at the recipient, 10 at the theme, and seven before either postverbal argument. Overall,
double-object datives and prepositional datives contained approximately the same proportion of
restarts (18% in double-object datives, 13% in prepositional datives).8There was one utterance with two adjacent fillers (viz., She gave the broom to her s- uh um son
S20; 4;0, control condition). Because the utterance medial fillers were measured from the offset of
the previous word to the onset of the following word, I excluded this complex filled pause from the
lengths measures reported in the table.
150 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
Pause type N Length range (ms) Average length (ms)
uh (initial) 1 91–91 91uh (medial) 3 230–270 247
um (initial) 2 733–979 856um (medial) 9 651–3287 1422
Unfilled pauses (medial) 53 118–2429 740
Total 68 91–3287 802
Table 4.9: Study 3: Frequency and length of pauses in responses with two postverbalarguments.
(13%) fell below the 250ms cut-off, 16 (30%) were shorter than 500ms, and 44 (83%)
were shorter than 1000ms. In the analysis below, I only included (as disfluencies)
unfilled pauses that were 500ms or longer. As McDaniel et al. (2010) argued (and
the distribution of the perceptible pauses in my data corroborates), this is likely a
reasonably conservative measure of pauses as speech disruptions in child data.
Restricting disfluent pauses to those over 500ms resulted in recoding 6 of the 58 re-
sponses with disfluencies as fluent. The remaining disfluencies were then categorized
according to their position in the utterance. When disfluencies occurred immedi-
ately before a postverbal argument or within the argument itself, that argument was
counted as disfluent (see 80-81). When disfluencies only occurred on or before the
verb (see 82), those utterances were categorized as “pre-argument only.”
(80) Disfluent themes:
a. Give a #[628ms] his #[1611ms] give #[1922ms] a plant to somebody.
(S22, 4;0, control condition)
b. He throwed the uh[270ms] package to a girl. (S04, 4;8, control condition)
c. She gave #[2429ms] the comb to the boy. (S06, 4;6, recipient-given)
4.3. RESULTS 151
Discourse condition
Disfluency type Control Theme-given Recipient-given Total
Pre-argument only 4 (36%) 2 (18%) 5 (45%) 11Theme 6 (43%) 1 (7%) 7 (50%) 14Recipient 11 (50%) 5 (23%) 6 (27%) 22Both arguments 3 (60%) 1 (20%) 1 (20%) 5
Total 24 (43%) 9 (17%) 19 (37%) 52
Table 4.10: Study 3: Distribution of disfluencies in responses with two postverbalarguments.
(81) Disfluent recipients:
a. Give a broom to uh[230ms] the man. (S23, 4;3, control condition)
b. She read it to um[1285ms] a boy. (S11, 4;0, theme-given)
c. He gave the plant to the (#[235ms]) to #[859ms] the #[1087ms] lady.
(S06, 4;6, theme-given)
(82) Pre-argument only:
a. She #[1689ms] throw it to the man. (S10, 4;1, theme-given)
b. He #[629ms] he showed the woman a star. (S21, 4;6 , recipient-given)
c. Um[733ms] she gave it to the man to brush his hair.
(S17, 4;1, theme-given)
When a response had both a pre-argument disfluency and a disfluency on one of the
arguments (n=7), it was grouped together with the disfluencies for that particular
argument. The distribution of the remaining 52 responses is provided in Table 4.10.
The distribution of these disfluencies supports the claim that disfluencies indicate
production difficulty. When the theme was given, the recipient was more likely to
be disfluent than the theme: 23% of disfluent recipients occurred in theme-given
condition, but only 7% of disfluent themes did. Similarly, when the recipient
152 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
was given, the theme was more likely to be disfluent than the recipient: 50% of
disfluent themes occurred in recipient-given condition, but only 27% of disfluent
recipients did. Also, the highest proportion of disfluencies overall occurred in
the control condition, where the children did not have the benefit of hearing either
postverbal argument prior to giving their response.
The disfluency data were collected primarily to test the hypothesis that utter-
ances with given-before-new indicate fewer processing difficulties than other orders,
assuming that placing given information first gives the speaker a processing advantage
by allowing them to mention the most accessible information first and save the less
accessible information for later. But this hypothesis is partially tied to the position
of the disfluencies in the sentence. If speakers need extra time to formulate the new
information, and therefore use given-before-new ordering, they might still produce the
new information disfluently after they have produced the given information. They
bought themselves some time, but that was insufficient to eliminate the difficulty.
The disfluent utterances in (79b) and (81b-81c) are potential examples of this. Of
course, the extra time may have been sufficient to avert the disfluency, so we need
not see disfluencies in sentence-final new information, but finding them there would
not invalidate the claim that given-before-new facilitates production.
In light of this, my primary aim with the disfluency data is to test the follow-
ing hypothesis: Except for utterances in which disfluencies only arise after the first
postverbal argument, there should be significantly fewer disfluencies in utterances
with given-before-new ordering than in those with new-new ordering or new-before-
given ordering. According to this measure, 38 (21%) of the 179 utterances with
two postverbal arguments were disfluent. As expected, the smallest proportion of
disfluent responses (15%) was for responses with the given-before-new ordering. Re-
sponses with new-before-given ordering had the second highest proportion of disflu-
encies (21%), and new-new responses (i.e., the responses in the control condition)
4.3. RESULTS 153
Factors Coefficient SE Odds P-value
given-before-new -1.0724 0.4845 0.3422 0.02687 *new-before-given -0.5338 0.4979 0.5864 0.28363 n.s.
Table 4.11: Study 3: Results of model predicting the presence of disfluencies.
were the most likely to be disfluent (30% disfluent). The fact that new-before-given
order shows fewer disfluencies than new-new order is also not surprising, because chil-
dren producing new-before-given order, still have the advantage that one argument
as been mentioned, even if they do not capitalize on that advantage by ordering that
argument first.
To test whether this difference is significant, I fitted a logistic regression model
with a disfluency factor (disfluent vs. fluent) as the binary outcome variable and
information structure (with the three levels mentioned above) as the primary predictor
variable. I also included construction choice (prepositional dative vs. double-object
dative), order, and age as control variables; but none of these were significant, so I
removed them from the model. I also included participant and verb as random effects.
The random effect for verb was not significant and was, therefore, also removed. The
results of the final model are given in Table 4.11. The model fit was acceptable
(C-statistic=0.82, Dxy=0.64, R2 = 0.96).
This model reveals that responses with given-before-new ordering were signifi-
cantly less likely to be disfluent than those with new-new ordering (the comparison
condition; i.e., responses in the control condition), while responses with new-before-
given order were not significantly different from those with new-new order. The
difference between given-before-new order and new-before-given order (15% vs. 21%
disfluent) did not reach significance. Thus, these data show some evidence that given-
before-new facilitates production (given-before-new utterances were significantly less
disfluent than new-new utterances), but they do not offer statistical confirmation of
154 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
the finding that given-before-new utterances were less disfluent than new-before-given
ones. This difference might prove significant if more data were included in the analy-
sis. At this point, we can say that given-before-new provides a processing advantage,
but the extent of this advantage remains unclear.
4.4 Discussion
The findings from this study confirm the primary results of Study 2A: Discourse
givenness influences how preschoolers order the two postverbal arguments of dative
verbs. The results of this study also allay concerns that the discourse effects evidenced
in Study 2A were due to differences in the visual stimuli or emphasis on the topicality
of the given information (see §3.2.3). In this study, the visual stimuli were held
constant across the discourse conditions, and children were never explicitly told that
the event was about the given information. This study also replicates the findings in
Study 2A that discourse did not influence all the verbs equally, and theme-givenness
had a more robust influence on syntactic choices than recipient-givenness. These
data also further address the role of verb semantics, frequency, and argument length
in dative construction choice and offer some preliminary support for a speaker-based
analysis of why children use given-before-new ordering.
Differences among verbs. This study systematically manipulated the semantic
class of the verbs and the frequency of the verbs in CDS (see Table 4.1). The frequency
manipulation was largely unsuccessful, though, because children often substituted a
different dative verb for the target verb (see Table 4.2). For the six verbs of caused
possession as well as for bring and take, children generally opted to describe the
event with the verb give. With the vignettes for throw (high-frequency), hurl (low-
frequency) and toss (low-frequency), children overwhelmingly described the events
with the verb throw (93% of the time). Aside from the vignette for throw, the one for
4.4. DISCUSSION 155
kick was the only one in the caused motion class that successfully elicited the target
verb. The low-frequency verbs loan and hurl were never produced, and the low-
frequency verb hand was only produced four times with both postverbal arguments
(see Table 4.5). The only low-frequency verbs produced fairly often were pass and
kick. Except for one use of kick, all of these responses were prepositional datives
(theme-first). These data do not offer any support for the hypothesis that children
are more likely to be influenced by discourse pressure with verbs when they are
relatively inexperienced with the verb’s argument structure (see the discussion of
entrenchment in §2.2.3 and §3.2.3).
But pass and kick may not be good representatives of low-frequency verbs. The
children seemed very comfortable with the verb kick, as evidenced at least in part
by the fact that they almost never substituted another verb for it. Perhaps the
children in my sample had already had ample experience hearing kick used in the
input. If so, there is a reasonably high likelihood that these uses of kick in the input
involved theme-objects and recipient-PPs (or other directional PPs), rather than
double-object datives like, Kick me the ball. Pass is also problematic given the shared
linguistic experiences of the children in this study. All the children attended the same
nursery school, where the daily snack time routine involves passing snacks around the
table. During this time, children are repeatedly asked (by the teacher) to pass the
apples, pass the Saltines, and so on. The recipient of this passing activity is rarely
mentioned—the general idea is that the snack should be passed around the table to
every child. When children want to have something passed to them, the generally
say something like I want an apple, not Please pass me an apple. In essence, this is a
high-frequency verb at least for the children in this sample, and they probably rarely
(if ever) hear it in a double-object construction.9 In the end, then, these data do not9Another possible concern with pass is that it can be used to describe throw-type events (e.g.,
Pass me the ball.). This was apparently not an issue, though. While children used pass for several of
156 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
allow us to assess whether children are more likely to use given-before-new ordering
with low-frequency verbs nor how frequency interacts with the semantic properties of
the event (i.e., caused possession vs. caused motion).
While the frequency manipulation failed, children did obligingly produce verbs in
both semantic verb classes (see Table 4.5), and verbs in the caused motion class were
significantly more likely to be used in prepositional datives (theme-first) than those
in the caused possession class. This is consistent with the view that the children
overwhelmingly interpreted these events as caused motion events, rather than change
of possession events, and, therefore, realized these verbs as prepositional datives (see
Rappaport Hovav & Levin 2008). In essence, the theme is construed as an object
moving to a goal (viz. the recipient), hence the appropriateness of the directional
to-phrase. Unlike the throwing event in Study 2A, though, these caused motion events
did not generally involve the type of routine ball play discussed in Section 3.2.3. The
event targeting kick involved a ball, but the events targeting throw, toss, and hurl
involved a box, a boot, and a hat, respectively. Consequently, these events are less
likely to have been experienced by the children. Nevertheless, they described these
events the same way they described those involving a ball. These results suggest that
the children’s caused motion verbs are less susceptible to discourse pressure and more
likely to be used in prepositional datives, regardless of whether the event represents
a typical throwing routine.
Responses for the vignettes targeting bring and take also deserve attention. De-
spite the fact that these verbs are very common in CDS (and in spontaneous child
speech), the children in this study essentially never produced them. The bring and
take vignettes were almost always described with the verb give (see Table 4.2). In
these vignettes, the agent walked a few feet with the theme in hand and then
the same events they also used give with, they never used it for the events involving ballistic caused
motion (throw, toss, hurl, and kick) (see Table 4.2).
4.4. DISCUSSION 157
gave the theme to the recipient. When describing these two events, children used
give with prepositional datives 66% of the time, while the other uses of give were
actually somewhat more likely to be prepositional datives (79% of the time). But
the added motion for the bring and take vignettes, may not have been particularly
salient to the children, since the agents only moved a few feet with the theme. This
might (in part) explain both why children did not more strongly prefer prepositional
datives with these events and why they substituted give for bring and take. More
likely, though, children avoided these verbs, because they require a particular deictic
context. Bring indicates movement toward and take movement away from the deictic
center—a central character, generally the speaker or addressee (Fillmore 1997:98-99).
The children’s avoidance of these verbs is reasonable given that the stimuli failed to
provide the necessary deictic relationships.
Givenness and the role of referring expressions. In Study 2A, the effect
of theme-givenness was categorical, whereas here it was gradient. Similarly, in
Study 2A, recipient-givenness had a weak but (marginally) significant influence
on construction choice, whereas in this study there were more double-object datives
(recipient-first) when the recipient was given, but not significantly so. This dif-
ference between the two studies, if meaningful, might be attributed to procedural
differences or even to differences in the vignettes themselves. Also, the addition of
more vignettes involving throw-type events may have reduced the likelihood of a ro-
bust effect for recipient-givenness. The children may have been more attuned to
attending to the motion of the theme, given the saliency of the theme movement in
the vignettes for kick, hurl, throw, and toss. Or perhaps producing more prepositional
datives in response to these events heightened the overall use of prepositional datives
and made children more likely to persist in that choice. Another possible reason is
that there were proportionally fewer responses with pronominal postverbal arguments
158 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
in these data than in those for Study 2A. At this point, we cannot rule out any of
the above explanations.
This difference aside, Study 3 replicates the finding that theme-givenness more
strongly affects construction choice than recipient-givenness. The reason for this
difference is likely related to the difference in how theme and recipient pronouns
are realized. In the responses with two postverbal arguments (excluding the verbs that
failed to alternate), 67% of given themes were pronominal, and 49% of given recip-
ients were pronominal. Recall that pronominal themes are only felicitous in prepo-
sitional datives (theme-first), while pronominal recipients are felicitous in both
constructions. This gives theme-givenness an a priori advantage over recipient-
givenness (see §3.2.3).
In Study 2A, givenness and pronominality were almost completely overlapping.
There were very few lexical NPs referring to given information and very few pronouns
referring to new information. In this study, there was still a robust correlation between
discourse condition and the choice of referring expression and very few pronouns
referring to new information, but there were more lexical NPs for given information
than in Study 2A. For the verbs that showed alternation, there were 17 lexical themes
in the theme-given condition. All but two of these—the responses with indefinite
given themes (see 77 and 76)—were realized in prepositional datives (theme-first).
Thus, when the theme was given, it was almost always realized in a prepositional
dative, even when it was lexical. Although the regression including pronominality
(see Table 4.7) did not show a significant difference between responses in the control
condition with two lexical arguments (the comparison condition) and those in the
theme-given condition with two lexical arguments, there was a higher proportion of
prepositional datives in the latter than in the former (88% vs. 76%).
As for recipients, when the recipient was given and pronominal, double-object
datives were significantly more likely than when the recipient was lexical and in
4.5. SUMMARY 159
the control condition. Pronominal recipients in the control condition were not
significantly more likely to appear in double-object datives than lexical recipients
in the same condition (though these were rather rare). Recipients may need to be
both given and pronominal to have a robust effect on word order. Perhaps with more
data, we would see a clear independent effect of both givenness and pronominality
in preschooler’s dative constructions, but the current data show (at the least) that
these two factors conspire together to encourage given-before-new ordering.
4.5 Summary
This study replicates the findings of the two previous studies that preschooler’s word
order choices are influenced by givenness, and it offers additional insight into the
question of whether this influence is direct or indirect (through the choice of referring
expressions). Apparently discourse has the strongest influence on word order when
the argument is both given and pronominal. Neither pronominality nor givenness
alone fully account for the results. This study also provides the best support so
far that semantic verb class can override discourse pressure to use given-before-new
ordering. Nevertheless, because the frequency manipulation was not fully successful,
we cannot rule out the possibility that the caused motion verbs were more resistant
to alternation because of verb-specific input patterns, rather than a general semantic
rule. Regardless, givenness clearly does not have a blanket effect on construction
choice, but interacts with a battery of factors that drive argument expression.
While the current data are consistent with both a speaker- and addressee-based
account for given-before-new ordering, the disfluency results provide some support
for a speaker-based account. The speaker-based account hinges on the notion that
160 CHAPTER 4. STUDY 3: DATIVE ALTERNATION REVISITED
using given-before-new ordering eases the speaker’s processing load.10 And the result
of the model predicting the presence of disfluencies confirms that given-before-new
orders were less disfluent than new-new orders. They were also less disfluent than
new-before-given orders, but not significantly so. Additional data are needed to show
conclusively that given-before-new ordering provides a substantial boost to processing
beyond the advantage gained by simply having heard one of the arguments mentioned.
The disfluency data also provide evidence that children process language similarly
to adults and that they signal processing difficulty in the same way. Though children
produced only a few filled pauses, these uses closely reflected the arguably subtle
difference between um and uh in adult language (see §4.1). If additional data corrob-
orate this difference in child language, this would provide new evidence that children
learn a wide range of subtle correspondences in the input, even when those are not
directly tied to referential meaning. In the next chapter, I bring together the results
of Studies 1-3 and further consider what the findings reveal about the extent to which
early syntactic choices are governed by learned correspondences, semantic properties
of the events and participants, communicative pressure, and processing factors.
10Of course, given-before-new might also lighten the addressee’s processing load, but my data do
not provide any measure of ease of comprehension (but see Clifton & Frazier 2004; Brown et al.
Submitted).
Chapter 5
Conclusion
5.1 Overview
The acquisition of word order is a complex problem, in part because children (like
adults) integrate information from multiple sources during production. This disser-
tation opens a window onto that complexity. Based on the results of three elicited
production studies, I have argued that givenness (specifically, explicit mention in the
immediately prior discourse) influences early word order choices. Previous work has
suggested that this is true for the choice of syntactic subject (Braine et al. 1990;
Braine & Brooks 1995), but my studies show that it is true for the choice of object as
well. Together, these studies provide converging evidence that young children attend
to givenness and use that information in formulating their utterances.
These findings are important in light of previous claims that children begin to
mark information status via referring expression quite early, but do not manage to
do so with word order until well after preschool (see §1.4). Based on data from
English locative and dative constructions, I have argued that preschool children use
word order in addition to referring expressions to mark givenness. In sum, I have
addressed the following questions:
161
162 CHAPTER 5. CONCLUSION
(83) a. Do children use given-first or new-first ordering?
b. Are all verbs equally affected by givenness?
c. Are all arguments equally affected by givenness?
d. Are givenness effects attributable to the type of referring expressions?
e. Why do children use given-before-new ordering?
This chapter provides a summary of my arguments as they relate to each of these
questions.
5.2 Do children use given-first or new-first ordering?
In all three of my studies with preschoolers, givenness was a significant predictor
of construction choice, and the preferred order was consistently given-before-new.
My data provide no evidence for a (non-adult-like) new-first preference. Some other
developmental work, however, indicates that young children use new-first ordering in
contexts where adults prefer given-first ordering (see §1.4.2). Perhaps the clearest
case for this is the work on conjunct ordering by Narasimhan & Dimroth (2008).
These researchers found that German-speaking three- and five-year-olds preferred
new-before-given ordering. In their tasks, the experimenter, who could not see the
visual stimuli, asked participants to label an object in a transparent cylinder (e.g., a
spoon). Afterward, another item (e.g., a ball) was added to the initial item in the
same cylinder, and again the experimenter asked the participant, Was ist da drin?
(‘What’s in there?’). In responding to this question, children were significantly more
likely to mention the newer item first and the older item second (e.g., a ball and a
spoon), while caregivers preferred the opposite ordering.
This study differs from mine in many ways. One important difference might be
the linguistic phenomena tested (conjunct ordering vs. argument ordering; see Bock
5.2. DO CHILDREN USE GIVEN-FIRST OR NEW-FIRST ORDERING? 163
& Warren 1985). For example, if children use given-before-new ordering to ease the
processing load, they may need this facilitation more with argument ordering than
conjunct ordering. When describing the filmed vignettes in my study, children had
to juggle multiple event participants and sort out the roles of those participants. In
the conjunct ordering task, though, children needed only to label the objects, all of
them familiar items. Perhaps this task failed to tax the processing system in a way
that would motivate given-before-new ordering, leaving room for a competing bias to
influence syntactic choices (e.g., a bias to attend to new items).1
Similarly, if children use given-before-new ordering to facilitate comprehension,
perhaps they recognized that their utterances would be easy to comprehend with
either ordering, so there was no need to observe the Given-New Contract (see §1.5.1).
Another potentially important difference is that the referents for the given and new
information were generally in the visual scene the same amount of time in my studies,
whereas the discourse-new item was new to the visual scene in the conjunct study.
This difference could have substantially increased the salience of the new argument,
neutralizing or overriding prominence brought about by the initial labeling.
Importantly, though, the labeling task, unlike my tasks was not embedded in
a coherent discourse. Participants labeled one item and subsequently labeled two
items, each time answering the same question (What’s in there? ). The two labeling
instances were not obviously part of the same conversation. In a follow-up study
in which this task was embedded in a coherent discourse, Narasimhan and Dimroth
found that four- and five-year-old children no longer showed a preference for new-
before-given (nor one for given-before-new) ordering (Narasimhan & Dimroth 2009;1One caveat here is that adults in this study were more likely to use given-before-new ordering—
obviously not because they had more difficulty retrieving the referent of the new item. But adult
performance could be based on other processes as well. Even if adults do not need to use given-
before-new ordering to facilitate production, they may do so because they have learned a preference
for this ordering through ample experience with more complex linguistic phenomena.
164 CHAPTER 5. CONCLUSION
Dimroth & Narasimhan 2010). Alongside my results, the differences in these two
conjunct ordering studies suggests that children need both a conversational context
and a situation that taxes production and/or comprehension (at least to some extent)
to capitalize on the given-before-new ordering strategy.
5.3 Are all verbs equally affected by givenness?
Different verbs showed different levels of susceptibility to discourse-pressure. Some
verbs showed given-first patterns of alternation (e.g., squirt, fill, give), while others
essentially never alternated (e.g., pour, throw). Discourse-driven alternation patterns
were found with verbs that are low-frequency in child-directed speech (CDS), whether
they allow alternation in adult speech (e.g., squirt) or not (e.g., fill). Given-before-
new ordering was also found with high-frequency verbs, but only those that show
alternation in CDS (e.g., give), and not with those that do not (e.g., pour, throw).
In light of these findings, I have argued that children do not learn given-before-new
ordering on a verb-by-verb basis, but they may well learn to resist given-before-new
ordering on a verb-by-verb basis. In other words, children choose constructions that
yield given-before-new ordering unless a stronger pressure (like evidence from usage-
based patterns) demands a different choice.
In addition to patterns of usage, semantic properties of the verbs may also override
any pressure to use given-before-new ordering. In the dative studies, children strongly
preferred prepositional dative constructions (theme-first) with verbs of caused mo-
tion, and in particular with events involving ballistic motion. This result is consistent
with the idea that when children interpret an event as a caused motion event, they
(presumably like adults) describe these events with prepositional dative constructions
and reserve double-object dative constructions for events of caused possession. Al-
ternatively, children may have learned from the input that these verbs rarely occur
5.4. ARE ALL ARGUMENTS EQUALLY AFFECTED BY GIVENNESS? 165
in double-object dative constructions. Regardless, properties of the individual verbs
(and perhaps of semantic classes of verbs) must be considered in any investigation of
how discourse influences argument structure. Verb-specific properties (be they seman-
tic, distributional, or both) can mask or nullify potential discourse effects. Givenness
is just one piece of the argument structure puzzle.
5.4 Are all arguments equally affected by givenness?
Givenness did not have a blanket effect on postverbal arguments either. In my
locative data, locatum-givenness was a significant predictor of construction choice,
but location-givenness was not. I have argued that this asymmetry was due to
discourse-external sources of prominence. In particular, the locatum, as the mov-
ing item in the visual scene, was more prominent (visually and/or conceptually) than
the location, which was stationary throughout (see §2.2.3). Thus, children (and
adults) were faced with a cue clash: When the location was given, the discourse
context indicated that the location was more prominent, but discourse-external
factors indicated that the locatum was more prominent. So the two sources of
prominence clashed, and the potential effect of discourse was cancelled out, yielding
syntactic choices that were no different from the baseline. But in the locatum-given
condition, discourse-prominence aligned with discourse-external prominence, and the
locatum was significantly more likely to be ordered first.
In the dative data, theme-givenness had a more robust influence on word order
than recipient-givenness. In Study 2A, both theme-givenness and recipient-
givenness were reliable predictors of construction choice, but theme-givenness had
a categorical effect, and recipient-givenness a gradient one. In Study 3, theme-
givenness was a robust (though not quite categorical) predictor of construction choice.
The effect of recipient-givenness did not reach significance, though there were more
166 CHAPTER 5. CONCLUSION
double-object dative constructions (recipient-first) when the recipient was given
than in the control condition. The difference between theme- and recipient-
givenness is not easily attributable to a clash between discourse-external and -internal
sources of prominence. If anything, recipients (by virtue of their humanness) should
have been more visually/conceptually prominent than themes (all inanimate), yield-
ing a cue clash in the theme-given condition, not in the recipient-given condition.
I have argued that the difference between theme- and recipient-givenness is due
to an asymmetry in pronoun realization patterns. In American English, recipient-
pronouns can felicitously occur in either the first or second postverbal argument
position, but theme-pronouns are only felicitous in the first position (see §3.2.3).
Because most given arguments were pronominalized, this difference in pronoun real-
ization can account for the advantage of theme-givenness over recipient-givenness
in influencing word order choices.
5.5 Are the effects due to referring expressions?
All of my studies reveal robust correlations between givenness and choice of refer-
ring expression in the speech of preschool children, and discourse-sensitive syntactic
choices often coincided with discourse-sensitive choices of referring expressions. These
findings were expected in light of the fact that word order and referring expressions
provide two ways of marking the same function—givenness. Given arguments tend to
be realized in the first postverbal argument position (given-first) and given arguments
tend to be pronominal (or definite NPs). The end result is harmonic alignment (Aissen
1999; Bresnan et al. 2007; Bresnan & Nikitina 2009; Bresnan & Ford 2010; de Marn-
effe et al. Submitted): The first postverbal argument tends to be given, short, and
definite, the second is more likely to be new, longer, and indefinite.
5.5. ARE THE EFFECTS DUE TO REFERRING EXPRESSIONS? 167
This alignment suggests that preschool children are well on their way to adult-like
production choices, but it also raises a difficult question: Does givenness influence
early syntax directly (given-first) or indirectly via some property of the referring
expressions (short-first and/or definite-first)? Although most work on givenness and
child language ignores this question (see §1.4), de Marneffe et al.’s (Submitted) corpus
study of give and show revealed that pronominality (length) significantly affected child
choices of dative constructions, while givenness did not. Specifically, prepositional
dative constructions (theme-first) were more likely when the theme was pronominal,
and double-object dative constructions (recipient-first) were more likely when the
recipient was pronominal. There was, however, no significant effect of theme-
givenness or recipient-givenness. De Marneffe and her colleagues argued that the
use of pronouns for new information in child speech might have masked any givenness
effects. For example, if children produce a pronominal theme when the recipient
is given, they are faced with two conflicting cues—the pronominality of the theme
encourages prepositional dative constructions (short-first), but the discourse status
of the recipient encourages double-object dative constructions (given-first). If the
pressure to use short-first ordering wins out in this cue clash, the use of pronouns for
new information would dilute any effects of givenness.
One difficulty in working with corpus data, though, is that discourse status is of-
ten difficult to assess. Naturalistic conversation takes place in settings where there is
substantial variation in the number of competing discourse referents, the contextual
salience of those referents, and their degree of givenness (e.g., number of mentions,
referring expressions used in prior mentions, etc.). These difficulties are exacerbated
with child data, because there is no established rubric for deciding under what circum-
stances children are likely to notice givenness in the surrounding discourse. While
elicited production studies have their own disadvantages, they have the advantage
that the discourse context and the situational context can be carefully controlled.
168 CHAPTER 5. CONCLUSION
To my knowledge, my studies are the first to consider the influence of givenness and
type of referring expressions on the ordering of postverbal arguments in a controlled,
experimental setting.
What do the current data tell us about the effects of givenness and referring ex-
pression type on early word order? Across my studies, there was little compelling
evidence that definiteness of lexical arguments was driving the effect of discourse
givenness. Definite NPs were produced for given and for new information. If defi-
niteness (rather than givenness per se) were responsible for construction choices, then
definite NPs referring to new information should have been just as likely to occur
in first position as those referring to given information. But in every case in which
an argument’s givenness was a significant predictor of construction choice, definite
NPs for that argument were more likely to occur in first position if they referred to
given information than if they referred to new information. All three studies provide
some evidence that givenness has an independent influence on construction choice
and is not simply a by-product of a preference for definite NPs as the first postverbal
argument. In fact, once givenness is taken into account, there is rather little evidence
for any preference for ordering definite NPs first.
While the children’s syntactic choices are not reducible to NP definiteness, they
may be largely attributable to pronominality. Pronominal arguments (whether given
or new) were more likely than lexical ones to occur in the first postverbal position.
In fact, all pronominal themes occurred in the first postverbal position, as did all
but two of the children’s pronominal locatums. In Study 1A, locatum-givenness
was a significant predictor of construction choice, and 76% of given locatums were
pronominal. In the dative studies, theme-givenness was a significant predictor of
construction choice. In Study 2A, 83% of given themes were pronominal, as were 68%
of given themes in Study 3. Recipient-givenness was also a significant predictor of
construction choice in Study 2A, where 81% of given recipients were pronominal.
5.5. ARE THE EFFECTS DUE TO REFERRING EXPRESSIONS? 169
Because the majority of these given arguments were pronominal and very few new
arguments were, independent effects of givenness or pronominality are difficult to
assess. Across my studies, though, there was a (non-significant) tendency for given
arguments to be ordered first in the data excluding pronouns. In other words, even
when both postverbal arguments were lexical, given arguments were more likely to
be ordered first.
Beyond these findings, there were a few additional clues that the givenness effects
are not reducible to pronominality (short-first). In Study 2A, 67% of pronominal
recipients that referred to given information occurred in the first postverbal position
(n=18), compared to only 14% of those that referred to new information (n=7). Also
in Study 3, where given recipients showed a non-significant tendency to occur in
the first position, a regression crossing discourse condition with referring expression
type revealed that recipients were significantly more likely to be positioned first
only if they were both given and pronominal.
Because of the sparsity of lexical arguments referring to given information, I can-
not tell whether givenness would have robustly influenced syntax had it not been for
the increase in pronominality. Since pronominality and word order both mark given-
ness, when givenness is not marked via pronominality, it should not be too surprising
to find that it is not marked via word order either. Similarly, some caution is advisable
in assuming that the type of referring expression is chiefly responsible for the syntac-
tic choices in cases where givenness and choice of referring expression fail to overlap
(e.g., the occasional uses of pronouns for new information; but cf. Anttila et al. In
press). Something is presumably responsible for the choice of referring expression in
these cases as well. When pronominal arguments for discourse-new information were
ordered first in my data, some factor other than discourse givenness may have led to
the choice of referring expression and the choice of word order. Further research is
170 CHAPTER 5. CONCLUSION
needed to determine whether this is the case and what that factor might be. Never-
theless, while it may be tempting to “avoid pesky questions” raised by pronominality
(Snyder 2003:52), researchers who choose to ignore pronominal data cannot ignore
the issue of pronominality altogether. My data highlight the strength of the tendency
to realize discourse-given arguments as pronouns. Whatever else causes speakers to
mark (or not mark) givenness via pronominality may well have the same effect on
their tendency to mark givenness via word order.
5.6 Why do children use given-before-new ordering?
In Chapter 1, I outlined three possible explanations for why children might use given-
before-new ordering: (i) to ease comprehension (an addressee-based approach), (ii)
to ease production (a speaker-based approach), or (iii) because they learned this
convention from the input and want to replicate it in their own speech (a statistical-
learning approach). Here again, though, a form and function problem arises. Each of
these three functions (viz. facilitation of comprehension, facilitation of production,
or reliance on conventions learned from the input) encourages production of the same
form (viz. given-before-new ordering). Perhaps all three explanations together con-
tribute to early given-before-new ordering. Definitively eliminating any of them may
be impossible.
Nevertheless, my data yielded three clues that restrict the range of possible expla-
nations. First, children preferred to realize given information as the first postverbal
argument of locative verbs even when no other postverbal argument was mentioned
(see 84).
(84) Vignette: male actor squirts water on a towel
E: . . .What’s the boy doing with the towel?
Child (S12, 3;3): Squirting the towel.
5.6. WHY DO CHILDREN USE GIVEN-BEFORE-NEW ORDERING? 171
I have argued that an addressee-based approach cannot neatly account for this ten-
dency (see §2.2.3). In these cases, children do not wait to mention the new informa-
tion until the addressee has had a chance to retrieve the information already given.
Instead, they simply omit the new information altogether.
Second, in Study 1A, children showed the most robust tendency to order given
information first with the verb fill. Children produced errors (i.e., locatum-object
constructions) with fill 88% of the time when the locatum was given, but only 25%
of the time in the control condition. This finding indicates that children do not use
given-before-new ordering because of verb-specific patterns they learn from the input.
Child uses of given-before-new ordering may be an attempt to replicate more global
given-before-new patterns in the input, but if children learn this pattern on a verb-
by-verb basis, we should find it only with high frequency verbs that provide clear
evidence of discourse-driven alternations. Give and show might be good candidates
for such verbs (see de Marneffe et al. Submitted), but fill is not (see §2.2.3).
The disfluency data in Chapter 4 yield a third clue about the source of early given-
before-new ordering. If children use this order to facilitate production, we should find
evidence of facilitation when this order is used. Indeed, four-year-olds’ dative con-
structions were significantly more fluent when they involved given-before-new ordering
than in the control condition (new-new ordering). Given-before-new responses were
also more fluent than new-before-given responses, though not significantly so. If more
data support this finding, this would be compelling evidence that given-before-new
ordering substantially eases the child’s processing load. Even if given-before-new or-
dering provides a significant processing advantage, this does not prove that children
actually use this ordering to facilitate processing. But these data do lay some of the
groundwork necessary to make a case for a speaker-based approach.2
2Of course, evidence that given-before-new ordering facilitates processing for the speaker might
also be taken as support for the claim that it facilitates processing for the addressee as well.
172 CHAPTER 5. CONCLUSION
While determining the precise motivation underlying given-before-new ordering is
difficult, this issue has important implications for our understanding of how children
move toward more adult-like production choices. If given-before-new ordering is a by-
product of processing limitations, then we can expect to find this ordering preference
in even younger children as long as other factors (like syntactic complexity) do not in-
tervene. The studies of transitivity alternations by Braine and his colleagues provide
some support along these lines (Braine et al. 1990; Braine & Brooks 1995). Alterna-
tively, if children use given-before-new ordering primarily to facilitate the addressee’s
comprehension, this skill should improve with age as children gain a better under-
standing of the needs and perspective of their addressee. Similarly, if children use
given-before-new ordering to replicate global conventions in the input, this preference
should increase with increased linguistic experience.
My studies do not show a clear developmental trajectory. Study 1A tested three-
and five-year olds and Studies 2A and 3 tested four-year-olds. Regardless of how
age was coded, my data yielded no interaction between age and discourse condition.
In other words, younger children and older children showed the same preference for
given-before-new ordering. This may be taken as support for a processing account,
or it may simply indicate that children in this age range are all at about the same
stage of development in ability to mark givenness via word order. Studies of children
belonging to a larger age range may be needed to determine the developmental course
of this discourse sensitivity. Regardless, my studies show that the syntactic choices
of children as young as three reflect an adult-like sensitivity to givenness, even the
givenness of postverbal arguments. This suggests that previous studies have under-
estimated (i) the ability of young children to mark givenness via word order and (ii)
the potential for early argument structure errors to be discourse-driven (see §1.4).
In the end, my results underline the fact that preschool children can and do
integrate information from multiple sources as they process language for production.
5.7. ADDING NEW INFORMATION TO THE DISCOURSE 173
Like adults, children take information from the grammar, patterns of language use,
the properties of the event, and properties of the situational and discourse contexts
into account when they formulate their utterances. Regardless of the final word
on what drives given-before-new ordering, these studies add to a growing body of
evidence pointing toward continuity in how children and adults process language for
production (e.g., Bannard & Matthews 2008; McDaniel et al. 2010; de Marneffe et al.
Submitted).
5.7 Adding new information to the discourse
This dissertation provides insight into the development of both information structure
and argument structure, but much remains to be done. While I have considered
two linguistic manifestations of discourse-sensitivity (viz. choice of word order and
choice of referring expression), these do not exhaust the linguistic tools for marking
givenness. Intonation patterns are also highly indicative of information structure,
at least in adult speech (e.g., Firbas 1966; Halliday 1970; Prince 1981). To date,
there are very few studies that consider whether young children use intonation in
an adult-like way to mark discourse givenness (but see Bates 1976; Grünloh et al.
In prep.). An analysis of the intonation contours children use for given versus new
information could provide additional insight into the range of devices preschoolers
exploit in marking information status. Once all of these tools are taken into account,
researchers will have an excellent basis for assessing not only whether children mark
givenness in an adult-like way, but also under what circumstances they choose one
marker over another, or choose to exploit multiple markers simultaneously.
Because my studies only manipulated discourse givenness, they do not reveal what
kinds of givenness have the strongest impact on early word order (see §1.4). Discourse-
givenness (linguistic copresence) may be a more important determinant of word order
174 CHAPTER 5. CONCLUSION
than physical copresence (cf. Campbell et al. 2000), but my data do not address this
issue. Givenness might also play a stronger role in some communicative contexts
than others (e.g., story telling vs. general conversation). Because the conversational
context was held constant within my studies, this question remains unanswered. My
studies also leave open the question of whether the degree of givenness plays a role in
how children use argument structure to mark information status. Mentioning an item
more or fewer times might also influence child ordering preferences. The field is ripe
for further investigation into the factors that influence early word order choices. In
the end, I suspect that our best efforts will support the conclusion of Bates (1976:200):
“[T]he acquisition of word order probably reflects an interaction of internal pragmatic
and semantic constraints, and high frequency environmental models, all competing
for limited channel use during communication.”
Bibliography
Abbot-Smith, K., & H. Behrens. 2006. How known constructions influence the
acquisition of other constructions: The German passive and future constructions.
Cognitive Science 30.995–1026.
Aissen, J. 1999. Markedness and subject choice in Optimality Theory. Natural
Language and Linguistic Theory 17.673–711.
Allen, S.E.M. 2000. A discourse-pragmatic explanation for argument representation
in child Inuktitut. Linguistics 38.483–521.
Ambridge, B., J.M. Pine, C.F. Rowland, & C.R. Young. 2008. The effect of
verb semantic class and verb frequency (entrenchment) on children’s and adults’
graded judgements of argument-structure overgeneralization errors. Cognition
106.87–129.
Anderson, J.M. 1977. On case grammar: Prolegomena to a theory of grammatical
relations . Atlantic Highlands, NJ: Humanities Press.
Anderson, M. 2006. Affectedness. In The Blackwell companion to syntax , ed. by
M. Everaert & H. van Riemsdijk, volume 1. Oxford: Blackwell.
Anderson, S.R. 1971. On the role of deep structure in semantic interpretation.
Foundations of Language 7.387–396.
175
176 BIBLIOGRAPHY
Anttila, A., M. Adams, & M. Speriosu, In press. The role of prosody in the
english dative alternation. Language and Cognitive Processes.
Arad, M. 2006. The spray-load alternation. In The Blackwell companion to syntax ,
ed. by M. Everaert & H. van Riemsdijk, volume 4. Oxford: Blackwell.
Ariel, M. 2001. Accessibility theory: An overview. In Text representation:
Linguistic and psycholinguistic aspects , ed. by T. Sanders, J. Schilperoord, &
W. Sporren, volume 8, 29–87. Amsterdam: John Benjamins.
Arnold, J., 1998. Reference form and discourse patterns . Stanford University
dissertation.
——, S. Brown-Schmidt, & J. Trueswell. 2007. Children’s use of gender and
order-of-mention during pronoun comprehension. Language and Cognitive Pro-
cesses 22.527–565.
——, & Z.E. Griffin. 2007. The effect of additional characters on choice of referring
expression: Everyone counts. Journal of Memory and Language 56.521–536.
——, A. Losongco, T. Wasow, & R. Ginstrom. 2000. Heaviness vs. newness:
The effects of structural complexity and discourse status on constituent ordering.
Language 76.28–55.
Arnon, I. 2010. Re-thinking child difficulty: The effect of NP type on child pro-
cessing of relative clauses in Hebrew. Journal of Child Language 37.27–57.
Baayen, H. 2008. Analyzing linguistic data: A practical introduction to statistics
using R. Cambridge, UK: Cambridge University Press.
——, D.J. Davidson, & D.M. Bates. 2008. Mixed-effects modeling with crossed
random effects for subjects and items. Journal of Memory and Language 59.390–
412.
BIBLIOGRAPHY 177
Bannard, C., & D. Matthews. 2008. Stored word sequences in language learning.
Psychological Science 19.241–248.
Bates, E. 1976. Language and context: The acquisition of pragmatics . New York:
Academic Press.
——, & B. MacWhinney. 1979. A functionalist approach to the acquisition of
grammar. In Developmental pragmatics , ed. by E. Ochs & B.B. Schieffelin, 167–
211. New York: Academic Press.
Behrens, H. 2007. The acquisition of argument structure. In Valency: Theoretical,
descriptive and cognitive issues , ed. by T. Herbst & K. Götz-Votteler, 193–214.
New York: Mouton de Gruyter.
Birner, B. 2006. Inferential relations and noncanonical word order. In Drawing
the boundaries of meaning: Neo-Gricean studies in pragmatics and semantics in
honor of Laurence R. Horn, ed. by B. Birner & G. Ward, 31–51. John Benjamins
Publishing Co.
——, & G. Ward. 1998. Information status and noncanonical word order in English.
Amsterdam: John Benjamins.
——, & ——. 2009. Information structure and syntactic structure. Language and
Linguistics Compass 3.1167–1187.
Bock, J.K., & D.E. Irwin. 1980. Syntactic effects of information availability in
sentence production. Journal of Verbal Learning and Verbal Behavior 19.467–
484.
——, & R.K. Warren. 1985. Conceptual accessibility and syntactic structure in
sentence formulation. Cognition 21.47–67.
178 BIBLIOGRAPHY
Bowerman, M. 1982. Reorganizational processes in lexical and syntactic devel-
opment. In Language acquisition: The state of the art , ed. by E. Wanner &
L. Gleitman, 319–346. Cambridge, UK: Cambridge University Press.
—— 1985. What shapes children’s grammars? In The crosslinguistic study of
language acquisition, ed. by D.I. Slobin, volume 2, 1257–1319. Hillsdale, NJ:
Lawrence Erlbaum Associates.
—— 1989. Learning a semantic system: What role do cognitive predispositions play.
In The teachability of language, ed. by M.L. Rice, R.L. Schiefelbusch, & R.K.
Hoyt Jr., 133–169. Baltimore, MD: Paul H. Brookes.
Braine, M.D., R.E. Brody, S.M. Fisch, M.J. Weisberger, & M. Blum. 1990.
Can children use a verb without exposure to its argument structure? Journal of
Child Language 17.313–342.
——, & P.J. Brooks. 1995. Verb argument structure and the problem of avoiding
an overgeneral grammar. In Beyond names for things: Young children’s acqui-
sition of verbs , ed. by M. Tomasello & W.E. Merriman, 353–376. Hillsdale, NJ:
Lawrence Erlbaum Associates.
Bresnan, J., A. Cueni, T. Nikitina, & H. Baayen. 2007. Predicting the da-
tive alternation. In Cognitive foundations of interpretation, ed. by G. Boume,
I. Kraemer, & J. Zwarts, 69–94. Amsterdam: Royal Netherlands Academy of
Science.
——, & M. Ford. 2010. Predicting Syntax: Processing dative constructions in
American and Australian varieties of English. Language 86.168–213.
BIBLIOGRAPHY 179
——, & T. Nikitina. 2009. The gradience of the dative alternation. In Reality explo-
ration and discovery: Pattern interaction in language and life, ed. by L. Uyechi
& L.H. Wee, 161–184. Stanford, CA: CSLI Publications.
Brinkmann, U. 1995. Acquiring the locative alternation: How can children tell
alternators from nonalternators? In The proceedings of the twenty-seventh annual
child language research forum, ed. by E.V. Clark, 111–121. Stanford, CA: CSLI
Publications.
Brooks, P.J., M. Tomasello, K. Dodson, & L.B. Lewis. 1999. Young children’s
overgeneralizations with fixed transitivity verbs. Child Development 1325–1337.
Brown, M., V. Savova, & E. Gibson, Submitted. Syntax encodes information
structure: Evidence from on-line reading comprehension.
Brown, R. 1973. A first language: The early stages . Cambridge, MA: Harvard
University Press.
Budwig, N. 1990. The linguistic marking of nonprototypical agency: An exploration
into children’s use of passives. Linguistics 28.1221–1252.
Cameron-Faulkner, T., E. Lieven, & M. Tomasello. 2003. A construction
based analysis of child directed speech. Cognitive Science 27.843–873.
Campbell, A.L., P. Brooks, & M. Tomasello. 2000. Factors affecting young
children’s use of pronouns as referring expressions. Journal of Speech, Language
and Hearing Research 43.1337–1349.
——, & M. Tomasello. 2001. The acquisition of English dative constructions.
Applied Psycholinguistics 22.253–267.
180 BIBLIOGRAPHY
Chafe, W. 1976. Givenness, contrastiveness, definiteness, subjects, topics, and
point of view. In Subject and Topic, ed. by C.N. Li, 27–55. New York: Academic
Press.
Chapman, R.S., & J.F. Miller. 1975. Word order in early two and three word ut-
terances: Does production precede comprehension? Journal of Speech, Language
and Hearing Research 18.355–371.
Chenu, F., & H. Jisa. 2006. Caused motion constructions and semantic generality
in early acquisition of French. In Constructions in acquisition, ed. by E.V. Clark
& B.F. Kelly, 233–261. Stanford, CA: CSLI Publications.
Choi, H.W. 2008. Beyond grammatical weight: A corpus study of information
structure effect on dative-accusative order in Korean. Discourse and Cognition
15.127–152.
Clark, E.V. 1990. Speaker perspective in language acquisition. Linguistics 28.1201–
1220.
—— 1997. Conceptual perspective and lexical choice in acquisition. Cognition 64.1–
37.
—— 2009. First language acquisition. Cambridge, UK: Cambridge University Press,
2nd edition.
——, & J. Bernicot. 2008. Repetition as ratification: How parents and children
place information in common ground. Journal of Child Language 35.349–371.
——, & K.L. Carpenter. 1989a. On children’s uses of from, by and with in oblique
noun phrases. Journal of Child Language 16.349–364.
BIBLIOGRAPHY 181
——, & K.L. Carpenter. 1989b. The notion of source in language acquisition.
Language 65.1–30.
——, & H.H. Clark. 1978. Universals, relativity, and language processing. In
Universals of human language, ed. by J.H. Greenberg, volume 1: Method and
theory, 225–277. Stanford, CA: Stanford University Press.
Clark, H.H. 1994. Managing problems in speaking. Speech Communication 15.243–
250.
——, & J.E. Fox Tree. 2002. Using uh and um in spontaneous speaking. Cognition
84.73–111.
——, & S.E. Haviland. 1977. Comprehension and the given-new contract. Discourse
Production and Comprehension 1.1–40.
——, & C.R. Marshall. 1981. Definite reference and mutual knowledge. In Ele-
ments of discourse understanding , ed. by A.K. Joshi, B.L. Webber, & I.A. Sag,
10–63. Cambridge, UK: Cambridge University Press.
——, & T. Wasow. 1998. Repeating words in spontaneous speech. Cognitive Psy-
chology 37.201–242.
Clifton, C., & L. Frazier. 2004. Should given information come before new? Yes
and no. Memory & cognition 32.886–895.
Collins, P. 1995. The indirect object construction in English: An informational
approach. Linguistics 33.35–49.
Conwell, E., & K. Demuth. 2007. Early syntactic productivity: Evidence from
dative shift. Cognition 103.163–179.
Cook, VJ. 1976. A Note on Indirect Objects. Journal of Child Language 3.435–437.
182 BIBLIOGRAPHY
de Marneffe, M., S. Grimm, I. Arnon, S. Kirby, & J. Bresnan, Submitted.
A statistical model of the grammatical choices in child production of dative
sentences.
de Villiers, J.G. 1985. Learning how to use verbs: Lexical coding and the influence
of the input. Journal of Child Language 12.587–595.
Demuth, K. 1989. Maturation and the acquisition of the Sesotho passive. Language
65.56–80.
——, M. Machobane, & F. Moloi. 2003. Rules and construction effects in learning
the argument structure of verbs. Journal of Child Language 30.797–821.
——, ——, ——, & C. Odato. 2005. Learning animacy hierarchy effects in Sesotho
double object applicatives. Language 81.421–447.
Diessel, H. 2007. Frequency effects in language acquisition, language use, and
diachronic change. New Ideas in Psychology 25.104–123.
Dimroth, C., & B. Narasimhan, 2010. Accessibility and topicality: How informa-
tion structure influences children’s early word order in different discourse con-
texts. Paper presented at the Workshop on Information Structure in Language
Acquisition (DGfS 2010).
Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67.547–
619.
Emonds, J., & R. Whitney. 2006. Double object constructions. In The Blackwell
companion to syntax , ed. by M. Everaert & H. van Riemsdijk, volume 2. Oxford:
Blackwell.
BIBLIOGRAPHY 183
Emslie, H.C., & R.J. Stevenson. 1981. Pre-school children’s use of the articles in
definite and indefinite referring expressions. Journal of Child Language 8.313–
328.
Erteschik-Shir, N. 1979. Discourse constraints on dative movement. In Syntax
and semantics: Discourse and syntax , ed. by T. Givón, volume 12, 441–467. New
York: Academic Press.
Fava, E. 1978. Grammatical relations and word order in italian child discourse.
Proceedings of the 4th Annual Meeting of the Berkeley Linguistics Society 512–
523.
Ferreira, V.S., & H. Yoshita. 2003. Given-new ordering effects on the production
of scrambled sentences in Japanese. Journal of Psycholinguistic Research 32.669–
692.
Fillmore, C.J. 1997. Lectures on deixis . Stanford, CA: CSLI Publications.
Firbas, J. 1964. On defining the theme in Functional Sentence Analysis. Travaux
linguistiques de Prague 1.267–280.
—— 1966. Non-thematic subjects in contemporary English. Travaux linguistiques de
Prague 2.239–256.
—— 1983. On some basic issues of the theory of Functional Sentence Perspective.
Brno Studies in English 15.9–36.
Fischer, S.D., 1971. The acquisition of verb-particle and dative constructions . MIT
dissertation.
Fisher, C. 2000. From form to meaning: A role for structural alignment in the
acquisition of language. Advances in Child Development and Behavior 27.1–54.
184 BIBLIOGRAPHY
——, D.G. Hall, S. Rakowitz, & L. Gleitman. 1994. When it is better to receive
than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua
92.333–375.
Fox Tree, J.E. 2001. Listeners’ uses of um and uh in speech comprehension.
Memory & Cognition 29.320–326.
Fraser, C., U. Bellugi, & R. Brown. 1963. Control of grammar in imitation,
comprehension, and production. Journal of Verbal Learning and Verbal Behavior
2.121–135.
Galati, A., & S.E. Brennan. 2010. Attenuating information in spoken communi-
cation: For the speaker, or for the addressee? Journal of Memory and Language
62.35–51.
Gallo, C.G., T.F. Jaeger, & R. Smyth. 2008. Incremental syntactic planning
across clauses. Proceedings of the 30th Annual Meeting of the Cognitive Science
Society (CogSci08) 845–850.
Gentner, D. 1978. On relational meaning: The acquisition of verb meaning. Child
Development 49.988–998.
Givón, T. 1983. Topic continuity in discourse: A quantitative cross-language study .
Amsterdam: John Benjamins.
—— 1984. Direct object and dative shifting: Semantic and pragmatic case. In Objects:
Towards a theory of grammatical relations , ed. by F. Plank, 151–182. London:
Academic Press.
—— 1988. The pragmatics of word-order: Predictability, importance and attention.
In Studies in syntactic typology , ed. by M. Hammond, E.A. Moravcsik, & J.R.
Wirth, volume 17, 243–284. Amsterdam: John Benjamins.
BIBLIOGRAPHY 185
Gleitman, L. 1990. The structural sources of verb meanings. Language Acquisition
1.3–55.
Goldberg, A.E. 1995. Constructions: A construction grammar approach to argu-
ment structure. Chicago: University of Chicago Press.
—— 2006. Constructions at work: The nature of generalization in language. New
York: Oxford University Press.
——, D.M. Casenhiser, & N. Sethuraman. 2005. The role of prediction in
construction-learning. Journal of Child Language 32.407–426.
Goldman-Eisler, F. 1972. Pauses, clauses, sentences. Language and Speech 15.103–
113.
Graff, D., J. Kong, K. Chen, & K. Maeda, 2007. English gigaword third edition.
Philadelphia, PA: The Linguistic Data Consortium.
Green, G.M. 1971. Some implications of an interaction among constraints. In
Proceedings of the 7th Regional Meeting of the Chicago Linguistic Society , 85–
100.
—— 1974. Semantics and syntactic regularity . Bloomington, IN: Indiana University
Press.
Greenfield, P. 1979. Informativeness, presupposition and semantic choice in single
word utterances. In Developmental Pragmatics , ed. by E. Ochs & B.B. Schieffelin,
159–166. New York: Academic Press.
Gropen, J., 1989. Learning locative verbs: How universal linking rules constrain
productivity . MIT dissertation.
186 BIBLIOGRAPHY
——, S. Pinker, M. Hollander, & R. Goldberg. 1991a. Affectedness and
direct objects: The role of lexical semantics in the acquisition of verb argument
structure. Cognition 41.153–195.
——, ——, ——, & ——. 1991b. Syntax and semantics in the acquisition of locative
verbs. Journal of Child Language 18.115–151.
——, ——, ——, ——, & R. Wilson. 1989. The learnability and acquisition of the
dative alternation in English. Language 65.203–257.
Grünloh, T., E. Lieven, & M. Tomasello, In prep. Young children’s intonational
marking of new and given referents.
Gundel, J.K. 1988. Universals of topic-comment structure. In Studies in syntactic
typology , ed. by M. Hammond, E.A. Moravcsik, & J.R. Wirth, volume 17, 209–
239. Amsterdam: John Benjamins.
——, & T. Fretheim. 2004. Topic and focus. In Handbook of pragmatics , ed. by
L.R. Horn & G. Ward, 175–196. Oxford: Blackwell.
——, N. Hedberg, & R. Zacharski. 1993. Cognitive status and the form of
referring expressions in discourse. Language 69.274–307.
Halliday, M.A.K. 1967a. Notes on transitivity and theme in English. Part 1.
Journal of Linguistics 3.37–81.
—— 1967b. Notes on transitivity and theme in English. Part 2. Journal of Linguistics
3.177–274.
—— 1970. Language structure and language function. In New horizons in linguistics ,
ed. by J. Lyons, 140–165. Baltimore, MD: Penguin Books.
BIBLIOGRAPHY 187
Hawkins, J.A. 1994. A performance theory of order and constituency . Cambridge,
UK: Cambridge University Press.
Hickmann, M. 1997. Information status and grounding in children’s narratives:
A crosslinguistic perspective. In Processing interclausal relationships: Studies
in the production and comprehension of text , ed. by J. Costermans & M. Fayol,
221–243. Mahwah, NJ: Lawrence Erlbaum Associates.
—— 2003. Children’s discourse: Person, space and time across languages . Cambridge,
UK: Cambridge University Press.
—— 2004. Coherence, cohesion, and context: Some comparative perspectives in nar-
rative development. In Relating events in narrative: Typological and contextual
perspectives., ed. by S. Strömqvist & L. Verhoven, volume 2, 281–306. Mahwah,
NJ: Lawrence Erlbaum Associates.
——, & H. Hendriks. 1999. Cohesion and anaphora in children’s narratives: A
comparison of English, French, German, and Mandarin Chinese. Journal of
Child Language 26.419–452.
——, H. Hendriks, F. Roland, & J. Liang. 1996. The marking of new information
in children’s narratives: A comparison of English, French, German and Mandarin
Chinese. Journal of Child Language 23.591–619.
——, M. Kail, & F. Roland. 1995. Cohesive anaphoric relations in French children’s
narratives as a function of mutual knowledge. First Language 15.277–300.
Hinrichs, L., & B. Szmrecsanyi. 2007. Recent changes in the function and
frequency of Standard English genitive constructions: A multivariate analysis of
tagged corpora. English Language and Linguistics 11.437–474.
188 BIBLIOGRAPHY
Hornby, P.A. 1971. Surface structure and the topic-comment distinction: A devel-
opmental study. Child Development 1975–1988.
Ingham, J., & I. Reid, 2001. Transcriber. Apple Computer, Inc.
Jaeger, T.F. 2008. Categorical data analysis: Away from ANOVAs (transformation
or not) and towards logit mixed models. Journal of Memory and Language
59.434–446.
——, & T. Wasow. 2006. Processing as a source of accessibility effects on variation.
In Proceedings of the 31st Annual Meeting of the Berkeley Linguistic Society , ed.
by R.T. Cover & Y. Kim, 169–180. Ann Arbor, MN: Sheridan Books.
Jeffries, L., & P. Willis. 1984. A return to the spray paint issue. Journal of
Pragmatics 8.715–729.
Kail, M., & M. Hickmann. 1992. French children’s ability to introduce referents
in narratives as a function of mutual knowledge. First Language 12.73–94.
Karmiloff-Smith, A. 1980. Psychological processes underlying pronominalization
and non-pronominalization in children’s connected discourse. In Papers from the
parasession on pronouns and anaphora, ed. by J. Krieman & E. Ojedo, 231–250.
Chicago: Chicago Linguistics Society.
Kidd, C., K.S. White, & R.N. Aslin. 2009. Children’s use of disfluencies for prag-
matic inference in lexical development. Proceedings of the 31st Annual Meeting
of the Cognitive Science Society 1494–1499.
Kidd, E., S. Brandt, E. Lieven, & M. Tomasello. 2007. Object relatives made
easy: A cross-linguistic comparison of the constraints influencing young children’s
processing of relative clauses. Language and Cognitive Processes 22.860–897.
BIBLIOGRAPHY 189
Kim, M., B. Landau, & C. Phillips. 1999. Cross-linguistic differences in children’s
syntax for locative verbs. Proceedings of the 23rd Boston University Conference
on Language Development 1.337–358.
Klein, D., & C.D. Manning. 2003a. Accurate unlexicalized parsing. In Proceedings
of the 41st Annual Meeting on association for computational linguistics , ed. by
E. Hinrichs & D. Roth, volume 1, 423–430. Morristown, NJ: Association for
Computational Linguistics.
——, & ——. 2003b. Fast exact inference with a factored model for natural language
parsing. In Advances in neural information processing systems 15 (NIPS 2002),
3–10. Cambridge, MA: MIT Press.
Krifka, M. 2004. Semantic and pragmatic conditions for the dative alternation.
Korean Journal of English Language and Linguistics 4.1–32.
Kruijff-Korbayová, I., & M. Steedman. 2003. Discourse and information
structure. Journal of Logic, Language and Information 12.249–259.
Kuno, S. 1972. Functional sentence perspective: A case study from Japanese and
English. Linguistic inquiry 3.269–320.
Laakso, A., & L.B. Smith. 2007. Pronouns and verbs in adult speech to children:
A corpus analysis. Journal of Child Language 34.725–763.
Levelt, W.J.M. 1989. Speaking: From intention to articulation. Cambridge, MA:
MIT Press.
Levin, B. 1993. English verb classes and alternations: A preliminary investigation.
Chicago: University of Chicago Press.
190 BIBLIOGRAPHY
Levin, H., I. Silverman, & B.L. Ford. 1967. Hesitations in children’s speech dur-
ing explanation and description. Journal of Verbal Learning and Verbal Behavior
6.560–564.
Lumsden, J.S. 1994. The load alternation: Semantic shifts and implicit arguments.
Revue québécoise de linguistique 23.79–98.
MacWhinney, B. 2000. The CHILDES project: Tools for analyzing talk . Hillsdale,
NJ: Lawrence Erlbaum Associates.
——, & E. Bates. 1978. Sentential devices for conveying givenness and newness:
A cross-cultural developmental study. Journal of Verbal Learning and Verbal
Behavior 17.539–558.
——, & H. Osser. 1977. Verbal planning functions in children’s speech. Child
Development 48.978–985.
Maratsos, M.P. 1974. Preschool children’s use of definite and indefinite articles.
Child Development 446–455.
Matthews, D., E. Lieven, A. Theakston, & M. Tomasello. 2005. The role
of frequency in the acquisition of English word order. Cognitive Development
20.121–136.
——, ——, ——, & ——. 2006. The effect of perceptual availability and prior dis-
course on young children’s use of referring expressions. Applied Psycholinguistics
27.403–422.
Mazurkewich, I., & L. White. 1984. The acquisition of the dative alternation:
Unlearning overgeneralizations. Cognition 16.261–283.
BIBLIOGRAPHY 191
McDaniel, D., C. McKee, & M.F. Garrett. 2010. Children’s sentence planning:
Syntactic correlates of fluency variations. Journal of Child Language 37.59–94.
Nadig, A.S., & J.C. Sedivy. 2002. Evidence of perspective-taking constraints in
children’s on-line reference resolution. Psychological Science 13.329–336.
Naigles, L.R., & E. Hoff-Ginsberg. 1998. Why are some verbs learned before
other verbs? Effects of input frequency and structure on children’s early verb
use. Journal of Child Language 25.95–120.
Narasimhan, B., & C. Dimroth. 2008. Word order and information status in
child language. Cognition 107.317–329.
——, & ——, 2009. Information status and word order in child and adult language.
Paper presented at Stanford University’s Construction of Meaning Workshop.
——, & M. Gullberg. 2006. Perspective-shifts in event descriptions in Tamil child
language. Journal of Child Language 33.99–124.
Oehrle, R.T., 1976. The grammar of the English dative alternation. MIT disserta-
tion.
—— 1977. Review of Georgia M. Green: Semantics and syntactic regularity. Language
53.198–208.
O’Neill, D.K. 1996. Two-year-old children’s sensitivity to a parent’s knowledge
state when making requests. Child Development 659–677.
—— 2005. Talking about “new” information: The given/new distinction and children’s
developing theory of mind. In Why language matters for theory of mind , ed. by
J.W. Astington & J.A. Baird, 84–105. New York: Oxford University Press.
192 BIBLIOGRAPHY
Osgood, C.E., & A.M. Zehler. 1981. Acquisition of bi-transitive sentences: Pre-
linguistic determinants of language acquisition. Journal of Child Language 8.367–
383.
Ozón, G. 2006. Ditransitives, the Given Before New principle, and textual retriev-
ability: A corpus-based study using ICECUP. In The changing face of corpus
linguistics , ed. by A. Renouf & A. Kehoe, 243–262. Amsterdam: Rodopi.
Pinker, S. 1989. Learnability and cognition: The acquisition of argument structure.
Cambridge, MA: MIT Press.
—— 1994. How could a child use verb syntax to learn verb semantics? Lingua
92.377–410.
Power, R.J.D., & M.F. Dal Martello. 1986. The use of the definite and indefi-
nite articles by Italian preschool children. Journal of Child Language 13.145–54.
Prat-Sala, M., & U. Hahn. 2007. Catalan children’s sensitivity to the discourse
constraints imposed by different kinds of question. Language Learning 57.443–
467.
Primus, B. 1998. The relative order of recipient and patient in the languages of
Europe. In Constituent order in the languages of Europe, ed. by A. Siewierska,
421–473. Berlin: Mouton de Gruyter.
Prince, E.F. 1981. Toward a taxonomy of given-new information. In Radical
pragmatics , ed. by P. Cole, 223–255. New York: Academic Press.
—— 1992. The ZPG letter: Subjects, definiteness, and information-status. In Dis-
course description: Diverse analyses of a fund raising text , ed. by S. Thompson
& W. Mann, 295–325. Amsterdam: John Benjamins.
BIBLIOGRAPHY 193
Quené, H., & H. Van den Bergh. 2004. On multi-level modeling of data from
repeated measures designs: A tutorial. Speech Communication 43.103–121.
Ransom, E.N. 1979. Definiteness and animacy constraints on passive and double-
object constructions in English. Glossa 13.215–240.
Rappaport, M., & B. Levin. 1988. What to do with theta-roles. In Syntax and
semantics: Thematic Relations , ed. by W. Wilkins, volume 21, 7–36. New York:
Academic Press.
Rappaport Hovav, M., & B. Levin. 2008. The English dative alternation: The
case for verb sensitivity. Journal of Linguistics 44.129–167.
Rispoli, M. 2003. Changes in the nature of sentence production during the period
of grammatical development. Journal of Speech, Language and Hearing Research
46.818–830.
Roeper, T., S. Lapointe, J. Bing, & S. Tavakolian. 1981. A lexical approach
to language acquisition. Language Acquisition and Linguistic Theory 35–58.
Rohde, D.L.T., 2001. Tgrep2 user manual.
Rowland, C.F., J.M. Pine, E.V.M. Lieven, & A.L. Theakston. 2003. Deter-
minants of acquisition order in wh-questions: Re-evaluating the role of caregiver
speech. Journal of Child Language 30.609–635.
Saffran, J.R., R.N. Aslin, & E.L. Newport. 1996. Statistical learning by
8-month-old infants. Science 274.1926–1928.
Saylor, M.M., J.A. Baird, & C. Gallerani. 2006. Telling others what’s new:
Preschoolers’ adherence to the given-new contract. Journal of Cognition and
Development 7.341–379.
194 BIBLIOGRAPHY
Schwartz-Norman, L. 1976. The grammar of ‘content’and ‘container’. Journal of
Linguistics 12.279–287.
Serratrice, L. 2008. The role of discourse and perceptual cues in the choice of
referential expressions in English preschoolers, school-age children, and adults.
Language Learning and Development 4.309–332.
Shimpi, P.M., P.B. Gámez, J. Huttenlocher, & M. Vasilyeva. 2007. Syntactic
priming in 3-and 4-year-old children: Evidence for abstract representations of
transitive and dative forms. Developmental Psychology 43.1334–1345.
Slobin, D.I. 1985. Crosslinguistic evidence for the language-making capacity. In
The crosslinguistic study of language acquisition, ed. by D.I. Slobin, volume 2,
1157–1249. Hillsdale, NJ: Lawrence Erlbaum Associates.
Smith, J., M. Durham, & L. Fortune. 2007. “Mam, my trousers is fa’in doon!”:
Community, caregiver, and child in the acquisition of variation in a Scottish
dialect. Language Variation and Change 19.63–99.
Smith, V.L., & H.H. Clark. 1993. On the course of answering questions. Journal
of Memory and Language 32.25–25.
Smyth, R.H., G.D. Prideaux, & J.T. Hogan. 1979. The effect of context on
dative position. Lingua 47.27–42.
Snow, C.E. 1979. The role of social interaction in language acquisition. In Children’s
language and communication: Minnesota symposia on child psychology , ed. by
W.A. Collins, volume 2, 157–182.
Snyder, K.M., 2003. The relationship between form and function in ditransitive
constructions . University of Pennsylvania dissertation.
BIBLIOGRAPHY 195
Snyder, W., & K. Stromswold. 1997. The structure and acquisition of English
dative constructions. Linguistic Inquiry 28.281–317.
Song, H., & C. Fisher. 2005. Who’s “she”? Discourse prominence influences
preschoolers’ comprehension of pronouns. Journal of Memory and Language
52.29–57.
——, & ——. 2007. Discourse prominence effects on 2.5-year-old children’s interpre-
tation of pronouns. Lingua 117.1959–1987.
Steedman, M., & P.N. Johnson-Laird. 1978. A programmatic theory of linguistic
performance. In Recent advances in the psychology of language: Formal and
experimental approaches , ed. by R.N. Campbell & P.T. Smith, volume 2, 171–
192. New York: Plenum Press.
Stephens, N., 2008. Animacy in early child and child-directed transitives: A corpus
study. Stanford University, MS.
Temperley, D. 2007. Minimization of dependency length in written English. Cog-
nition 105.300–333.
Theakston, A.L., E.V.M. Lieven, J.M. Pine, & C.F. Rowland. 2004. Seman-
tic generality, input frequency and the acquisition of syntax. Journal of Child
Language 31.61–99.
Thompson, S.A. 1990. Information flow and dative shift in English discourse. In
Development and diversity: Language variation across time and space, ed. by
J.A. Edmondson, F. Crawford, & P. Mühlhäusler, 239–253. Dallas, TX: Summer
Institute of Linguistics.
196 BIBLIOGRAPHY
Thothathiri, M., & J. Snedeker. 2008. Syntactic priming during language com-
prehension in three-and four-year-old children. Journal of Memory and Language
58.188–213.
Tily, H., S. Gahl, I. Arnon, N. Snider, A. Kothari, & J. Bresnan. 2009.
Syntactic probabilities affect pronunciation variation in spontaneous speech. Lan-
guage and Cognition 1.147–165.
Tomasello, M. 2003. Constructing a language: A usage-based theory of language
acquisition. Cambridge, MA: Harvard University Press.
Turner, E.A., & R. Rommetveit. 1968. Focus of attention in recall of active and
passive sentences. Journal of Verbal Learning and Verbal Behavior 7.543–548.
Uziel-Karl, S. 2002. Acquisition of verb argument structure: Canonical mapping or
verb-by-verb? Proceedings of the 26th Boston University Conference on Language
Development 26.701–711.
Vallduví, E. 1992. The informational component . New York: Garland Publishing.
Viau, J. 2006. Give= CAUSE/GO: Evidence for early semantic decomposition of
dative verbs in English child corpora. Proceedings of the 30th Boston University
Conference on Language Development 30.665–676.
Ward, G., B. Birner, & R. Huddleston. 2002. Information packaging. In
The Cambridge Grammar of the English Language, ed. by R. Huddleston &
G. Pullum, 1363–1447. Cambridge, UK: Cambridge University Press.
Warden, D.A. 1976. The influence of context on children’s use of identifying
expressions and references. British Journal of Psychology 67.101–112.
BIBLIOGRAPHY 197
Waryas, C., & K. Stremel. 1974. On the preferred form of the double object
construction. Journal of Psycholinguistic Research 3.271–280.
Wasow, T. 1997. Remarks on grammatical weight. Language Variation and Change
9.81–105.
—— 2002. Postverbal behavior . Stanford, CA: CSLI Publications.
White, L. 1987. Children’s overgeneralizations of the English dative alternation.
In Children’s language, ed. by K.E. Nelson & A. van Kleek, volume 6, 261–287.
New York: Gardner Press.
top related