University of Southern California Dissertations and Theses

Investigating coordination of lexical and structural information cross-linguistically

(USC Thesis Other)

Investigating coordination of lexical and structural information cross-linguistically

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

INVESTIGATING COORDINATION OF LEXICAL AND STRUCTURAL
INFORMATION CROSS-LINGUISTICALLY

by

Heeju Hwang

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(LINGUISTICS)

August 2012

Copyright 2012 Heeju Hwang

ii
Dedication

To my family and friends

iii
Acknowledgements

This dissertation would not have been possible without the encouragement,
support and help of many people.
First and foremost, I would like to thank my advisor and dissertation chair Elsi
Kaiser. Through our countless meetings and emails, she helped me to be a better scholar.
I am extremely grateful for her support and guidance.
I would also like to thank my committee members – Andrew Simpson and Irving
Biederman for their helpful comments and support. In particular, Andrew was the chair
for my second screening and a committee member for the first screening and a qualifying
exam. I thank him for his encouragement and long-term dedication to my work.
I’d like to thank Stephanie Harves for introducing me to the field of linguistics. If
it had not been for her teaching, I would have never found the joy of linguistics.
Many others helped me with the work that went into this dissertation. Ed
Holsinger helped me design eye-tracking experiments and analyze the data. I thank him
for taking the time to help me. I also thank Erika Varis for recording English distractor
words in the picture-word interference experiment. I am very grateful to Juho Bae and
Byeri Hwang for drawing many of the pictures for the experiments.
I would not have been able to get through the graduate school without the support
of my friends in the program. David Li has been a great source of academic, and
emotional support. I thank him for being a great listener and helping me to look on the
bright side of life. To others in the program – Soyoung Park, Hyuna Byun Kim, Dongsik

iv
Lim, Heui-Joo Jeoung, Yi-Hsien Walker, Stephanie Huang, Xiao He, Erin Tavano and
Michal Temkin Martinez, I am also grateful for their help and companionship. I thank
Joyce Perez for helping me with all sorts of administrative problems.
Many other friends at USC made graduate school more fun than it would have
been otherwise. I thank Jinyoung Lee, Eunyoung Song, Hyunsoo Ha, Gukhui Han, Chan
Lim, Kyumin Kim, Sukwon Kim, and Tony Hong for brining cheer to my academic life.
My friends outside the USC community have also provided me continuing
support and encouragement through the years. I am extremely thankful to Jaeyoung Yoon,
EJ Lee and Kyunga Park for their care and love. They have been a constant source of
support in all I do and helped me to become a better person. I am also grateful for Charles
Kim for his resourcefulness.
And, last but not least, I thank my parents for their unconditional love. I also
thank Eunbi and Toby for accompanying me through the journey.

v
Table of Contents

Dedication ii

Acknowledgements iii

List of Tables vii

List of Figures viii

Abstract x

Chapter 1 Introduction 1
1.1 Coordination problem 1
1.2 Need for the cross-linguistic study of sentence production 4
1.3 Typological properties of English and Korean 6
1.4 Overview of the dissertation 9

Chapter 2 Is Verb Selection Essential for Lexical-Structural Coordination? 12
2.1 Introduction 12
2.2 Experiment 1 19
2.3 Discussion 33

Chapter 3 Lexical-Structural Coordination in English and Korean:
Evidence form Semantic Priming 37
3.1 Introduction 37
3.2 Experiment 2 49
3.3 Discussion 55

Chapter 4 Lexical-Structural Coordination in English and Korean:
Evidence form Attention-capture Cueing 57
4.1 Introduction 57
4.2 Experiment 3 58
4.3 Discussion 64

Chapter 5 Structural Incrementality in Korean:
Evidence form Syntactic Priming 69
5.1 Introduction 69
5.2 Experiment 4 70
5.3 Discussion 77

Chapter 6 Lexical-Structural Coordination in English and Korean:
Evidence form Perspective Priming 79

vi
6.1 Introduction 79
6.2 Experiment 5 82
6.3 Discussion 86

Chapter 7 Lexical-Structural Coordination in English and Korean:
Evidence form Eye Movements 90
7.1 Introduction 90
7.2 Experiment 3: Attention-capture cueing 98
7.3 Experiment 2: Semantic priming 107
7.4 General discussion 117

Chapter 8 General Discussion 120
8.1 Summary of the findings 120
8.2 Implications and questions for future research 122

References 129

Appendix: Baseline Rates of Actives vs. Passives in the Norming Study 139

vii
List of Tables

Table 1 Paired t-tests on the mean proportion of looks to objects in the
object-related and unrelated conditions during 800-1200 ms 32

Table 2 Effectiveness of attention-capture manipulation in Experiment 3 61

Table 3 Paired t-tests on the mean proportion of looks to the cued character
within the first 200 ms of the image display 102

Table 4 Paired two-tailed t-tests on the mean proportion of looks to the first
vs. second referent within the first 200 ms of the image display 103

Table 5 Paired t-tests on the mean proportion of looks to the primed vs.
unprimed character 110

viii
List of Figures

Figure 1 Stages of mental processing engaged in the picture-word interference
paradigm in Roelof (1992) 18

Figure 2 An example of target images 21

Figure 3 Utterance onset latencies of subjects, verbs, and objects in the related
and unrelated condition in English and Korean 25

Figure 4 An example of a verb region 27

Figure 5 Proportion of looks to subjects, objects, and actions in the unrelated and
the verb-related conditions in English 28

Figure 6 Proportion of looks to subjects, objects, and actions in the unrelated and
the verb-related conditions in Korean 30

Figure 7 A scene depicting a biting event 39

Figure 8 English display sequence for Experiment 2 51

Figure 9 Effects of semantic primes on English and Korean speakers’ choice of
active and passives 53

Figure 10 Utterance latencies for active and passive sentences in English and
Korean in the semantic priming experiments 55

Figure 11 Display sequence for attention-capture curing experiment 60

Figure 12 Effects of attention-capture cues on English and Korean speakers’
production of actives 62

Figure 13 Utterance latencies for active and passive sentences in English and
Korean in the attention-capture experiments 64

Figure 14 Display sequence for Experiment 4 72

Figure 15 Effects of syntactic primes on Korean speakers’ choice of actives and
passives 74

Figure 16 Utterance latencies for active and passive sentences in the syntactic
priming experiment 75

ix
Figure 17 Utterance latencies for active and passive sentences by prime condition
in the syntactic priming experiment 76

Figure 18 A scene depicting a chasing event 79

Figure 19 Example of a pictured event (biting) 83

Figure 20 Proportion of active sentences (left) and agent-initial sentences (right) in
English when primed with an agent or a patient perspective 85

Figure 21 Proportion of active sentences (left) and agent-initial sentences (right) in
Korean when primed with an agent or a patient perspective 86

Figure 22 Proportion of looks to the cued vs. uncued scene character relative to
picture onset in English and Korean from Experiment 1 101

Figure 23 Proportion of looks to the first vs. the second referent relative to speech
onset in English and Korean from Experiment 3 104

Figure 24 Proportion of looks to primed vs. unprimed character relative to picture
onset in English and Korean from Experiment 2 108

Figure 25 Proportion of looks to N1 vs. N2 relative to speech onset in English and
Korean from Experiment 2 111

Figure 26 Proportion of looks to N1 vs. N2 relative to speech onset in active
utterances in English and Korean from Experiment 2 113

Figure 27 Proportion of looks to N1 vs. N2 relative to speech onset in passive
utterances in English and Korean from Experiments 2 and 3 114

x
Abstract
Sentence production requires selection of lexical items and structural frames, and
coordination of the two in accordance with the grammar. There are two accounts of how
lexical-structural coordination occurs: a lexical account and a structural account, which
differ in the relative contributions that words and syntax make to early sentence
formulation. A lexical account suggests that lexical items control the formulation of
sentence structures. On the other hand, a structural account suggests that structural
frameworks control lexical items; speakers generate a rudimentary syntactic plan from
their construal of an event, and use this structural information to control the timing of
subsequent lexical retrieval.
My dissertation research investigates how English and Korean speakers integrate
lexical and structural information. The three primary questions that my research
investigates are: Is selection of verbs necessary for sentence planning?; What factors
affect speakers’ choice of subject and sentence structures?; How do eye-movements
relate to speakers’ linguistic choice? My investigation of these questions makes use of
experimental methods such as real-time eye-tracking paradigm and statistical techniques
such as mixed-effects models.
My results show that the mechanisms of lexical-structural coordination interact
with grammatical properties of a language. In English – a language with fixed word order
where grammatical functions such as the subject/object are highly correlated with word
order (e.g., the first noun is almost always the subject in English) – lexical items can
preside over the coordination process. In contrast, in Korean – a language with flexible

xi
word order where grammatical functions are indicated by case particles – structural
frameworks guide the coordination process.
By integrating two distinct fields in the study of language – grammar and
production via cross-linguistic research – my research aims to improve our understanding
of language processing beyond what studies of English alone could achieve, and
contributes to the development of psycholinguistic theories that can accommodate
language-specific processes.

1
Chapter 1
Introduction

1.1 Coordination problem
Language production has not been as intensively investigated as language
comprehension, primarily due to methodological challenges (Ferreira, 2009). In
comprehension research, the input stimulus can be precisely controlled. In production
research, however, it is not obvious how to elicit the sentences that are necessary for
theory development and testing. Although recent methodological advances have led to a
better understanding of sentence production, many aspects of the process from thoughts
to speech remain unknown or are subject to debate.
One of the central issues in the encoding of thoughts into linguistic form is how
speakers coordinate lexical and structural information. In order to produce well-formed
sentences, speakers need to retrieve lexical items, generate a syntactic plan, and
coordinate the two in the way that a particular language requires. How lexical and
structural information are integrated is a linguistic version of the binding problem and
often referred to as the coordination problem (Bock, 1987) (Outside of language
production, the canonical example of the binding problem is how visual features such as
shape are associated with another feature such as their location, Rosenblatt, 1961).
There are two accounts as to how lexical-structural coordination occurs: a lexical
account and a structural account (Bock, Irwin, & Davidson, 2004; Konopka, 2009;
Konopka & Bock, 2009 among others). The two accounts are not mutually exclusive but

2
differ in the relative contributions that they attribute to words and syntax in the process of
sentence formulation (e.g., both lexical and structural incrementality are in working, but
one process could take precedence over the other in the early sentence formulation). Aa
lexical account suggests that lexical items exert a strong influence on the formulation of
sentence structure. This account is grounded in part in the fact that the lexical properties
of verbs often constrain syntactic structures. For example, the verb donate can only occur
in the prepositional object structure (e.g., The man donated some money to the charity/
*The man donated the charity some money), whereas the (virtually) synonymous verb
give can occur in both the prepositional object and the double object structure (e.g., The
man gave some money to the charity/The man gave the charity some money). Syntactic
structures are also contingent upon the accessibility of nouns. Given a choice between
semantically equivalent structures, speakers tend to produce the structure that enables
earlier accommodation of the more accessible lexical item (see Ferreira & Slevc, 2007 for
a review). For example, if the patient entity is more accessible than the agent entity,
speakers are more likely to produce a passive sentence, mentioning the patient entity first
(e.g., A policeman is being bitten by a dog as compared with A dog is biting a policeman).
The immediate influence of lexical accessibility on sentence structures makes sense if
speakers create structures piecemeal, processing the more accessible items sooner. This
mapping mechanism in which lexical retrieval drives sentence structures is referred to as
lexical incrementality (Konopka, 2009; Kuchinsky, Bock, & Irwin, 2011).
On the other hand, a structural account assumes structural incrementality, in
which an early structural plan controls lexical retrieval, rather than lexical items

3
controlling sentence structures. According to this view, speakers generate a rudimentary
syntactic plan on the basis of the relational or causal structure of an event, and use this
structural information to control the timing of subsequent lexical retrieval (Bock et al.,
2004; Konopka, 2009; Kuchinsky et al., in press). The structural framework is not fully
specified, yet it includes information about the dependencies between lexical items and
their ordering – i.e., where to start and how to continue an utterance (Konopka, 2009;
Kuchinsky et al., in press). The structural view accounts for structural variation (e.g.,
active vs. passive) in terms of subtle differences in relational meanings between
structures, rather than in terms of accessibility of individual lexical items (Bock et al.,
2004). For example, if speakers construe the patient entity as ‘figure’ and the agent entity
as ‘background’, they are more likely to produce a passive sentence (e.g., A policeman is
being bitten by a dog). Evidence for structural incrementality comes from studies
showing the existence of an initial period of scene apprehension, during which speakers
are assumed to formulate a structural plan (Griffin & Bock, 2000; Bock, Irwin, Davidson,
& Levelt, 2003; Bock et al., 2004; Kuchinsky et al., in press).
My dissertation research aims to explore the validity of these two accounts of
lexical-structural coordination in typologically different languages, namely English and
Korean. Although the mechanisms of language production are often assumed to be
universal and not dependent on a speaker’s language, most, if not all, of the evidence for
this claim comes from English, Dutch, and German, which are typologically similar (for a
discussion of the importance of typologically diverse work on sentence production, see
Jaeger & Norcliff, 2009). This makes the assumption of universal production

4
mechanisms potentially problematic. In fact, cross-linguistic work on typologically
different languages shows that language-specific properties interact with production
mechanisms (e.g., Arnold, Wasow, Losongco, & Ginstrom, 2000 on English vs. Choi,
2007 on Korean for constituent ordering preferences). In light of the cross-linguistic
evidence for language-specific production mechanisms, the present study investigates
whether typological differences between English and Korean affects the process of the
lexical-structural integration.

1.2 Need for the cross-linguistic study of sentence production
Psycholinguistic research often assumes that the mechanisms of language
production are universal and not dependent on a specific language (e.g., Bock et al. 2001;
Bock et al. 2006). However, existing cross-linguistic work on typologically different
languages shows that language-specific properties interact with production mechanisms.
For example, it has been demonstrated that English speakers exhibit a strong preference
to produce shorter constituents before longer ones. Thus, they are more likely to produce
‘I gave [the book] [to the man riding a bicycle]’ than ‘I gave [the man riding a bicycle]
[the book]’ (e.g., Hawkins, 1994; Arnold et al., 2000; Wasow, 2002). Such short-before-
long preferences are also observed in the ordering of prepositional phrases (PPs) (e.g., I
saw him [yesterday] [in the park] vs. I saw him [there] [a couple of hours ago]), verb
particle shift (e.g., Give [it] up vs. Give up [any hope]), and heavy NP shift (e.g., Put [the
apple] [in the basket] vs. Put [in the basket] [the apple that’s on the napkin]). Because
short-before-long tendencies are also observed in languages other than English such as

5
German (e.g., Uszkoreit, 1987) and Dutch (e.g., Haeseryn, 1997), it was assumed to be a
universal feature of the production mechanism resulting from the Principle of Immediate
Mention. According to this principle, short phrases are formulated faster and thus, are
selected earlier for production (de Smedt, 1994; Wasow, 1997; 2002; Ferreira & Dell,
2000).
However, investigations of typologically different languages such as Japanese and
Korean have challenged the universality of the short-before-long preferences. For
example, Choi (2007) found that unlike English speakers, Korean speakers prefer to
produce long constituents before short ones as illustrated in ex (1) (See Yamashita &
Chang, 2001 for similar findings in Japanese).

(1) Long-before-short preferences in Korean
Mary-ka [kkoli-lul huntunun kay-eykey] [ppye-lul] cwuessta.
Mary-NOM [ tail wagging dog-DAT] [bone-ACC] gave
‘Mary gave [a dog wagging his tail] [a bone]’.

The long-before-short preferences in Korean cannot be explained by the Principle of
Immediate Mention. Rather, Hawkins (2004) suggests that the asymmetry in English and
Korean stems from a need to keep the heads of the two constituents close so that the two
heads are recognized more quickly (see Yamashita & Chang, 2001; 2006 for an
alternative explanation). For example, in English placing the short phrase before the
longer phrase as in ‘I gave [the book] [to the man riding a bicycle]’ allows the heads of

6
the two constituents (underlined) to be recognized more quickly than the reverse ordering,
‘I gave [the man riding a bicycle] [the book]’. In Korean, the opposite is true. As can be
seen in ex (1a), when Korean speakers produce a long phrase before a short one, the
heads of the two constituents are kept closer than when a short phrase precedes a long
one as in ex (2). Thus, Korean speakers exhibit long-before-short preferences.

(2) Kunye-ka [ppye-lul] [kkoli-lul huntunun kay-eykey] cwuessta.
Mary-NOM bone-ACC tail wagging dog-DAT gave
‘Mary gave a dog wagging his tail a bone’.

The constituent-ordering preferences in English and Korean bring out the need for
cross-linguistic work. Because languages differ in their grammatical properties and thus,
can differ quite strikingly in their production preferences, cross-linguistic work is
necessary for achieving a better understanding of the mechanisms of sentence production.
Looking at typologically different languages will thus allow us to test existing hypotheses
and can guide the development of new hypotheses.

1.3 Typological properties of English and Korean
As seen in Choi (2007), Korean is useful for evaluating existing psycholinguistic
theories, because its typological properties are different from those of English.
Specifically, English is a head-initial language, where the heads of syntactic constituents
occur before their complements. Thus, in VPs headed by verbs, verbs precede

7
complement phrases (e.g.,
VP
[
VERB
[ride]
COMPLEMENT
[a bicycle]]). Similarly, in relative
clauses, head nouns precede modifying clauses (e.g.,
RELATIVE CLAUSE
[
HEAD
[the man]
MOD
[who is riding a bicycle]]). English also has a fairly strict word order of Subject–
Verb–Object, and the grammatical functions of arguments are usually defined in terms of
their relative order. Thus, although English allows non-canonical word orders such as
preposing (e.g., carrots, I like, see Ward, 1988; Birner & Ward, 1998), these structures
are relatively rare and the sentential-initial noun phrase usually serves as the subject in
English.
Unlike English, Korean is a head-final language, where complements precede
heads. Thus, complement phrases come before verbs in VPs and modifying clauses
precede head nouns in relative clauses, which can be seen in ex (1). Korean also marks
grammatical functions by case particles rather than word order. In Korean, a subject is
marked by the nominative marker –i/ka, and a direct object is marked by the accusative
marker –(l)ul. Because case particles provide information about the grammatical
functions, word order is relatively free in Korean; canonical sentences in the active voice
have the order of Subject–Object –Verb, but arguments can be reordered as long as verbs
occupy the sentence final position as can seen in ex (3).

(3) ‘She gave a dog a bone’
a. Mary-ka kay-eykey ppye-lul cwuessta. [S-IO-DO-V]
Mary-NOM dog-DAT bone-ACC gave

8
b. Mary-ka ppye-lul kay-eykey cwuessta. [S-DO-IO-V]
Mary-NOM bone-ACC dog-DAT gave

c. Kay-eykey Mary -ka ppye-lul cwuessta. [IO-S-DO-V]
dog-DAT Mary-NOM bone-ACC gave

d. Kay-eykey ppye-lul Mary-ka cwuessta. [IO-DO-S-V]
dog-DAT bone-ACC Mary-NOM gave

e. Ppye-lul Mary-ka kay-eykey cwuessta. [DO-S-IO-V]
bone-ACC Mary-NOM dog-DAT gave

f. Ppye-lul kay-eykey Mary-ka cwuessta. [DO-IO-S-V]
bone-ACC dog-DAT Mary-NOM gave

Reordering of arguments without changes in grammatical functions, as shown in ex (3a-f),
is called ‘scrambling’.
By investigating how two typologically different languages, namely English and
Korean integrate lexical and structural information, I aim to improve theoretical
understanding of language processing beyond what studies of English alone could
achieve, and to contribute to the development of psycholinguistic theories that can
accommodate language-specific as well as universal processes.

9
1.4 Overview of the dissertation
To investigate how English and Korean speakers integrate lexical and structural
information, I examine the following questions:

(i) Is selection of verbs necessary for lexical-structural coordination?
(Chapter 2)
(ii) What factors affect English and Korean speakers’ choice of subject and
choice of sentence form? (Chapters 3, 4, 5, and 6)
(iii) How do English and Korean speakers’ eye movements unfold in terms of
their linguistic choice? (Chapter 7)

My investigation of these questions uses core experimental techniques such as visual-
world eye-tracking and reaction-time measuring, and newly developed statistical methods
such as mixed-effects models (Bates & Sarkar, 2007; Baayen, 2008).
More specifically, Chapter 2 investigates whether English and Korean speakers
select verbs before they start to speak. A strong version of a lexical account emphasizing
the role of lexical items in lexical-structural integration posits that sentence structures can
be derived from syntactic and semantic specifications of verbs. After all, the
subcategorization frames and argument structures of verbs govern the mapping between
thematic roles and syntactic functions (e.g., Grimshaw, 1990). For example, the verb
‘send’ assigns the recipient role to the indirect object (e.g., ‘I sent John the letter’), but
the verb ‘receive’ assigns the recipient role to the subject (e.g., ‘John received the letter’).

10
Thus, it seems reasonable to hypothesize that speakers access verb lemmas before speech
onset (Bock & Levelt, 1994). Using a picture-word interference paradigm, Chapter 2
investigates whether retrieval of verb lemmas is essential for lexical-structural
coordination in English and Korean. The investigation of verb retrieval in typologically
different languages yields a stronger claim as to whether verb retrieval is necessary for
lexical-structural coordination than the investigation of English alone; if verbs are
retrieved in both English and Korean before speakers start to speak, it suggests that verbs
are essential in lexical-structural coordination regardless of whether they occur early
(English) or late (Korean) in the sentence, which provides support for the strong lexical
account. But if verbs are not retrieved in either of the languages, it suggests that factors
other than verbs may be responsible for lexical-structural coordination, which is
compatible with a structural account.
Chapters 3-6 investigate factors affecting speakers’ choice within active vs.
passive structures in English and Korean. According to a lexical account, sentence
structures are contingent upon the accessibility of noun. That is, more accessible lexical
items tend to be assigned to sentence-initial subject positions, with direct consequences
for the syntactic structures that speakers end up producing. For example, when describing
a scene of a fox chasing a chicken, English speakers beginning with an acted-upon
patient entity (e.g., ‘chicken’ in ‘a fox is chasing a chicken’) is more likely to produce a
passive sentence with the chicken in the subject position (e.g., ‘a chicken is being chased
by a fox’). Thus, if lexical incrementality drives lexical-structural coordination, lexical
accessibility should have a significant influence on speakers’ choice of the subject nouns

11
and consequently, on the structures that they produce. On the other hand, according to a
structural account, speakers first choose a structural framework which controls the timing
of subsequent lexical retrieval. That is, speakers determine the subject or the structure of
their utterance based on their structural plans rather than lexical accessibility.
In order to evaluate these two accounts of lexical-structural coordination in
English and Korean, I conducted production experiments where I manipulated lexical
accessibility via semantic priming (Bock, 1986a) and visual attention-capture cues
(Gleitman, January, Nappa & Trueswell, 2007). These studies are reported in Chapter 3
and Chapter 4. In Chapters 5 and 6, I report experiments where I manipulated speakers’
structural frameworks via syntactic priming (e.g., Bock, 1986b), and perspective priming
(e.g., Bower, 1976; 1977).
Chapter 7 investigates the time courses of eye movements in connection to
English and Korean speakers’ linguistic choice. Because eye tracking provides a
continuous measure of the temporal relationship between looking and speaking, it is well
suited to assess the mechanisms of lexical-structural coordination.
Chapter 8 summarizes the findings in Chapters 2-7 and addresses the theoretical
implications and questions for future research.

12
Chapter 2
Is Verb Selection Essential for Lexical and Structural Coordination?

2.1 Introduction
This chapter investigates whether verb selection is necessary for the coordination
of lexical and structural information. The two main accounts of lexical-structural
coordination discussed in Chapter 1, the structural and lexical accounts (Bock et al.,
2004; Konopka, 2009), make different assumptions concerning whether verbs are
accessed before speakers start to speak. According to the structural account, speakers
derive a structural plan or structural framework directly from the relational or causal
structure of a pre-linguistic message (e.g., Griffin & Bock, 2000; Bock et al., 2004). This
predicts that speakers do not need to access a particular verb before they start to speak –
i.e., verb access is not necessary for lexical-structural coordination. For example, if an
event involves two entities, speakers formulate a transitive structural plan that can
accommodate two arguments without the need to access a particular verb’s
subcategorization frame or argument structure.
Although the structure of an event may serve as a rudimentary structural frame for
utterance, crucially a verb’s subcategorization frame and argument structure determine
which slot of the frame is filled by which element of a message (e.g., ‘replace X with Y’
and ‘substitute Y for X’). This suggests that selection of verbs may dictate sentence
structures. In fact, a strong version of the lexical account posits that sentence structures
are derived from the syntactic and semantic specifications of verbs (Bates &

13
McWhinney, 1982; Bock, 1982; Bock & Levelt, 1994; Tomasello, 2000). For example,
Bock and Levelt (1994) suggest that the subject noun cannot be accessed until the verb
has been accessed, because the assignment of thematic roles to functional roles is
controlled by verbs. For example, the verbs ‘send’ and ‘receive’ both require the recipient
role but it is mapped to different syntactic functions; ‘send’ assigns the recipient role to
the indirect object as in ‘I sent John the letter’, whereas ‘receive’ assigns the recipient
role to the subject as in ‘John received the letter’. The fact that the mapping between
functional roles and syntactic functions is determined by the requirements of specific
verbs suggests that verbs may play a central role in lexical-structural coordination and
thus, predicts that verb lemmas should be obligatorily selected. In sum, the structural
account predicts that verb lemmas are not accessed before speech, whereas the lexical
account suggests that the retrieval of verb lemmas is necessary for integrating lexical and
structural information.

2.1.1 Previous research on verb access in English, Dutch, and German
Previous research provides mixed evidence on whether verbs are retrieved before
speech onset. Lindsley (1975; 1976) suggests that English speakers may process some
aspects of verbs before speech onset. When participants were asked to produce subject-
verb sentences (e.g., ‘the man greets’) and actor-only utterances (e.g., ‘the man’), it took
longer for speakers to initiate subject-verb sentences than actor-only utterances. Lindsley
suggests that the delayed latencies of subject-verb sentences are due to retrieval of verb
lemmas in subject-verb sentences. However, when the utterance latencies of subject-verb

14
sentences were compared to the latencies of verb-only utterances (e.g., ‘greeting’),
speakers initiated subject-verb sentences faster than verb-only utterances. Lindsley
suggests that this is because the processing of verbs begins before speech onset but is not
entirely completed before speech onset in subject-verb sentences.
Kempen and Huijbers (1983) provide evidence for retrieval of verb lemmas
before speech onset in Dutch. Similar to Lindsley (1975; 1976), Kempen and Huijbers
had speakers describe pictures of an actor performing an action (e.g., greeting).
Participants were asked to name (i) only the actor (e.g., ‘man’), (ii) only the action (e.g.,
‘greet’), or (iii) both the actor and the action in the Subject-Verb or Verb-Subject order
(e.g., ‘man greets…’ or ‘…greets man’). They found that latencies for naming only the
action were the same as latencies for naming both the actor and action, and thus
concluded that both subject and the verb lemmas were retrieved prior to articulation.
Schriefers, Teruel, and Meinshausen (1998), however, provide results suggesting
that verb lemmas are not obligatorily retrieved before articulation in German. In picture-
word interference experiments, German speakers were asked to describe intransitive
actions (e.g., laugh) and transitive actions (e.g., open) with varied word order (Subject–
Verb or Verb–Subject), while ignoring distractor words. The distractor words were either
semantically related or unrelated to the target verbs. Semantically related distractor verbs
are known to delay utterance latencies for naming actions, because they interfere with
selection of the target verbs (e.g., Roelofs, 1993). Thus, if German speakers select verb
lemmas before speech, semantically related distractor verbs should result in delayed
utterance latencies as compared to semantically unrelated distractor words. Schriefers et

15
al. found that utterance onset latencies were delayed only when transitive verbs occurred
in the utterance-initial position; no distractor effect was found for instransitive verbs and
transitive verbs in the non-utterance-initial position. This suggests that verb lemmas are
not always retrieved before speech onset.
To summarize, evidence concerning retrieval of verb lemmas is mixed. Whereas
Lindsley (1975; 1976) and Kempen and Huijbers (1983) suggest at least some planning
of verbs is necessary for preparing utterances, Schriefers et al. (1998) suggest that verb
selection is not obligatory. None of the studies, however, provide conclusive evidence on
whether verb selection is critical for sentence formulation or not. Rather what they may
show is that verb selection could depend on the experimental designs. That is, the results
reported in the previous experiments could be potentially contributed to strategic
planning. Strategic effects on planning scope resulting from experimental design have
been reported by several researchers (F. Ferreira & Swets, 2002; Korvorst et al., 2006;
Meyer, 1997; Meyer, Roelofs, & Levelt, 2003; Schriefers, 1993). In fact, the experiments
of Lindsley (1975; 1976), Kempen and Huijbers (1983) and Schriefers et al. (1998) had
designs in which participants could adopt a strategy. For example, in studies of Lindsley
(1975; 1976) and Kempen and Huijbers (1983), participants repeatedly produced only a
limited set of nouns and verbs without filler items. Thus, it is possible that the frequent
repetition of a limited set of targets might have caused access to verb lemmas in their
studies.
In Schriefers et al. (1998), the verb interference effects were only found for
transitive verbs in the sentence-initial position. However, transitive actions always

16
involved animate agents and inanimate patients (e.g., man opening a door). The use of
animate agents and inanimate patients might have reduced or eliminated speakers’ need
for verb selection for function assignment. That is, the grammatical function of the
subject or the object could be assigned by conceptual properties (animate – subject,
inanimate – object) rather than by a verb’s subcategorization frame or argument structure.
Thus, the results of the previous experiments could be strategic due to the
frequent repetition of limited words or animacy bias. The present experiment aims to
examine whether verb selection is necessary for sentence formulation in English and
Korean by minimizing repetitive frames and controlling animacy in a picture-word
interference paradigm. The target verbs are transitive verbs, which require two arguments
– a subject and an object. If grammatical functions are assigned through the verb’s
subcategorization frame and argument structure (e.g., “give” vs. “send”) as Bock and
Levelt (1994) suggests, verb interference effects should be found in English and Korean.
Investigating Korean in addition to English could yield stronger evidence as to
whether verb selection is necessary for sentence formulation; Korean is a verb-final
language in which verbs appear in the sentence-final position. Thus, if Korean speakers
retrieve verbs before speech despite the verb's late occurrence in the sentence, this would
provide support for a lexical account that verbs play a crucial role in integrating lexical
and structural information by showing that verb selection is not related to its early or late
occurrence in the sentence. On the other hand, if verbs are not retrieved before speech
onset in either English or Korean, this would suggest that verb retrieval is not necessary
for sentence formulation, providing evidence against the lexical account. In addition to

17
verb interference effects, the present study also investigates object interference effects to
see whether object lemmas are retrieved before speech onset.

2.1.2 Picture-word interference paradigm
In standard picture-word interference experiments in which speakers are
presented with a pictured object and asked to name it while ignoring a distractor word,
the reaction times are typically longer when the target and the distractor belong to the
same semantic category (e.g., Glaser & Dungelhoff, 1984; Lupker, 1979, 1982; Rayner &
Posnansky, 1978; Rayner & Springer, 1986; Underwood, 1976; Underwood & Briggs,
1984). The delayed latencies caused by semantically related distractor words are referred
to as semantic interference effects.
The interference effects are regarded as originating at the lemma level (i.e., during
the selection of the target lemma) rather than at the conceptual level of processing. This is
because longer latencies are obtained only when subjects are asked to name the target
objects – a task which requires lemma access – but not when they perform a categorizing
or recognition task, which accesses the conceptual level (e.g., Glaser & Dungelhoff,
1984; Smith & Magee, 1980; Schriefers, 1993). The interference effects are also
observed for verbs. For simple action naming (e.g., ‘cry’), Roelofs (1993) showed that
semantically-related distractor verbs (e.g., ‘laugh’) prolonged naming latencies relative to
unrelated distractors (e.g., ‘jump’).
According to Roelofs (1992), the interference effects occur as a result of the
competition between the target lemma and the distractor lemma; when the distractor word

18
is heard, the distractor lemma is activated and competes with the target lemma. The target
and the distractor receive more activation when they are related because they activate
each other via connections to shared conceptual nodes. However, the distractor lemma is
more strongly activated than the target lemma partly because the path from picture to
distractor lemma node is shorter than from distractor to target lemma node (Fig. 2). As a
result, a distractor word delays the selection of the target lemma.

Fig. 1. Stages of mental processing engaged in the picture-word interference paradigm
in Roelof (1992)

The experiment in this chapter uses the semantic interference effects to determine
whether verb lemmas are selected before speech onset in English and Korean. As in
Schriefers et al. (1998), each target picture was accompanied by an auditory distractor

19
word. The distractors were semantically related to either (i) the verb in the target image
(e.g., sip for a picture depicting a licking event) or to (ii) the grammatical object in the
target picture (e.g. bandit for a picture in which the pirate is the patient/grammatical
object). If interference effects are obtained for verbs in English and Korean, this would
indicate that verb lemmas are retrieved before utterance onset, suggesting that they are
crucial for lexical-structural coordination. But if verb lemmas are not retrieved before
utterance onset, no interference effect should be found, suggesting that they are not
essential for the coordination process.

2.2 Experiment 1

2.2.1 Method
Participants
Eighteen native speakers of English and 16 native speakers of Korean from the
University of Southern California participated in the experiment for $10 per hour.

Materials
Picture norming study. In order to construct target images, a norming study was
conducted on a separate group of 16 English speakers and 14 Korean speakers.
Participants were asked to name 20 characters. Based on naming latencies and error rates,
eight characters within similar naming latencies with low error rates were selected; in
English, the mean naming latency of the characters was 1124 ms and the error rates were

20
less than 2%. In Korean, the mean naming latency was 1386 ms and the error rates were
4%. The characters also had similar frequencies and syllable length. In English, the target
characters had a word frequency of under 5 per million and the mean word frequency was
4 per million (determined based on the CELEX database, Baayen, Piepenbrock, &
Guikers, 1995). All character names were either one or two syllables in length: four were
monosyllabic and four were disyllabic. In Korean, the target characters had a word
frequency of under 7 per million and the mean frequency was 5 per million (calculated
based on Kang & Kim, 2009). The characters were either disyllabic (four characters) or
trisyllabic (four characters).
The norming study also ensured that the distractor verbs used in the main
experiment were capable of creating interference effects for the target verbs. The
participants in the norming study named 10 depicted actions while ignoring semantically
related or unrelated distractor words. Semantically related distractors were chosen from
cohyponyms (e.g., ‘laugh’) of the actions (e.g., ‘cry’) (Roelofs, 1993), which share the
same superordinate as the action verbs. Only the four actions that exhibited the
interference effects both in English and Korean were selected as target verbs; the average
interference effect was 99 ms in English and 170 ms in Korean, which is greater than the
average interference effect of 74 ms obtained in Schriefers et al. (1998).

Experimental stimuli. Eight target images were constructed using eight characters and
four actions. Each action verb was used twice, but the same characters were never
involved in the same action more than once. Each character occurred once as an agent

21
(the doer of the action) and once as a patient (the acted-upon entity of the action) over the
course of the experiment. To prevent a distractor word priming both subject and object
characters, the two characters in each picture belonged to different semantic categories:
animal and human. The location of the subject and the object (left or right) were
counterbalanced. Twelve fillers were constructed using the characters and the actions in
the norming study which were not used to construct the target images.

Fig. 2. An example of target images

Design
The eight experimental items and twelve fillers were assigned to three
blocks of 60 trials (20 trials per block). Each target and filler item occurred only once in
every block. Because each block contained the eight targets and the 12 fillers and there
were three blocks, every target and filler occurred three times over the course of the
experiment. In total, each participant completed a total of 24 target trials (and 36 filler
trials).

22
The target pictures were presented with three types of distractors: (i) a verb
distractor semantically related to the target verb, (ii) a noun distractor semantically
related to the target object, and (iii) a semantically unrelated adjective distractor to both
the verb and the object. For target characters (e.g., pirate in Figure 2), the related
distractor was a member of the same semantic category as the character (e.g., bandit). For
target verbs (e.g., lick in Figure 2), the distractor was a cohyponym of the verb (e.g., sip).
The three different conditions (verb-related distractor, object-related distractor,
baseline/unrelated distractor) were tested within subjects; Each subject saw each of the 8
target pictures three times: once in the verb-related distractor condition, once in the
object-related distractor condition and once in the unrelated distractor condition. The
repetition of targets within subjects allows us to evaluate the effects of the different
distractor types within subjects. The order of items within each block was
pseudorandomized so that (i) Experimental items did not appear as the first item in any
block, (ii) more than two experimental items did not appear consecutively, and (iii)
experimental items that shared a character or action were separated by at least three trials.
The ordering of block representation was rotated across participants. After each block,
participants performed a word recognition task, in which they were presented with
auditory words and asked if they heard those words in the immediately preceding block.
This was to make sure that participants paid attention to the auditory distractor words.
Before experimental blocks, participants were presented with five practice trials and two
word recognition trials.

23
Procedure
Following Schriefers et al., (1998), participants were given a booklet including
the instructions and the target pictures prior to the experiment. Beside each picture, the
words to be used in the picture descriptions were printed. Participants were instructed to
use only these words for their naming responses.
As soon as participants indicated that they had read the instruction and studied the
picture names, the experimenter asked them to name some of the characters and actions
of the target pictures to ensure that they knew the correct names and actions for the
pictures. After that, participants were seated in front of a 21-inch computer screen and
presented with five practice items before proceeding to the main experiment.
On each trial, participants were first presented with a crosshair for 1000 ms,
which was neutrally located between the characters. Participants were asked to look at the
cross and to press a button on a game controller. Immediately after the button press, the
picture and the auditory distractor were presented simultaneously. Participants described
the picture as quickly as they could and their descriptions were recorded. Participants’
eye movements were also recorded during the scene description. An SR Research
Eyelink II head-mounted eye-tracker was used to collect and store eye-tracking data. At
the end of the description, participants pressed a controller button to proceed to the next
trial. The entire experiment lasted approximately 30-40 min.

Analyses
Following Griffin (2001), the onset of the subject noun (e.g., pirate) was used to

24
mark the beginning of speech instead of the onset of the first function word (e.g., the or
a/an) or fillers (e.g., mmm). Responses where participants did not use the expected
picture names or verbs were categorized as errors and removed from the analyses. Onset
latencies of subjects, verbs, and objects were manually determined using the phonetic
software package Praat (Boersma & Weenink, 1992). Using the MAD-median rule (e.g.,
Wilcox, 2012), trials with outlier latencies were also removed. This resulted in the
removal of about 4 % of the trials in English (13 out of 324 trials) and in Korean (12 out
of 288 trials). The two target items depicting ‘tickle’ were further removed from the
subsequent analyses in English and Korean. This was done because the distractor ‘fiddle’
had a fairly prominent noun meaning, namely, a kind of violin.

2.2.2 Results
Utterance onset latencies
Fig. 3 shows the mean onset latencies of subjects, verbs, and objects in the related
distractor condition and the unrelated condition in English and Korean.

25

Fig. 3. Utterance onset latencies of subjects, verbs, and objects in the related and
unrelated condition in English and Korean

1000#
1500#
2000#
2500#
3000#
3500#
4000#
None% Verb%Related% Object%Related%
Onset%latencies%(ms)%
English%
Object%onset%
Verb%onset%
Subject%onset%
1500$
2000$
2500$
3000$
3500$
4000$
4500$
None% Verb%related% Object%Related%
Onset%latencies%(ms)%
Korean%
Verb%onset%
Object%onset%
Subject%onset%

26
Verb-related distractors caused significant interference effects in English; compared to
unrelated distractors, verb-related distractors delayed utterance (subject) onset latencies
by 140 ms (β=148.01, t=2.504, p<.05, SE=59.10). They also subsequently delayed verb
latencies by 651 ms (β=670.02, t=8.204, p<.001, SE=81.67) and object latencies by 196
ms ((β=213.75, t=2.386, p<.05, SE=89.58). Although verb distractors in Korean caused
greater interference effects than in English in the norming study, they did not delay
utterance (subject) onset latencies in Korean (β=124.53, t=1.328, p=.18, SE=93.81). They
also did not have any influence object latencies (β=126.1, t=1.188, p=.236, SE=106.2) or
verb latencies (β=115.0, t=.955, p=.340, SE=120.5).
In both English and Korean, object-related distractors marginally delayed
utterance onset latencies (β=98.15, t=1.655, p=.098, SE=59.29 in English, β=73.69,
t=.787, p=.43, SE=93.69 in Korean). In English, however, object-related distractors
significantly delayed verb and object onset latencies; they delayed verb onsets by 457 ms
(β=480.53, t=5.865, p<.001, SE=81.94) and object onsets by 184 ms (β=208.24, t=2.317,
p<.05, SE=89.87). In Korean, object-related distractors delayed neither object (β=145.2,
t=1.369, p=.172, SE=106.1) or verb onsets (β=172.6, t=1.435, p=.152, SE=120.3).

Eye movements
Looks to verbs. The verb regions were defined as small regions of a scene that
depicted the action as seen in Fig. 4.

27

Fig. 4. An example of a verb region

Fig. 5 plots proportion of looks to subjects, objects and verbs relative to picture
onset in English. As can be seen in Fig. 5, English speakers tended to look more at the
action region early in the image display in all conditions despite its relatively small areal
proportion. The analyses of two-tailed paired t-tests revealed that English speakers
looked significantly more at the action region during 200 - 400 ms after image display in
the verb-related condition than in the unrelated condition (t
1
(17)=-3.2873, p<.01; t
2
(5)=-
1.7183, p=.146).

28
Fig. 5. Proportion of looks to subjects, objects, and actions in the unrelated and the verb-
related conditions in English.

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Picture'Onset'
English:'Unrelated'
Subject"
Object"
Verb"
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Pciture'Onset'
English:'Verb:related'
Subject"
Object"
Verb"

29
Fig. 5, Continued

Fig. 6 plots proportion of looks to subjects, objects and verbs relative to picture
onset in Korean. Unlike English speakers, Korean speakers did not look more at the verb
regions than subjects and objects early in the image display. However, just like English
speakers, they looked more at the verb regions during 200 – 400ms in the verb-related
condition than in the unrelated condition (t
1
(15)=-3.375, p<.01; t
2
(5)=-2.10, p=.088).

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Pciture'Onset'
English:'Object:related'
Subject"
Object"
Verb"

30
Fig. 6. Proportion of looks to subjects, objects, and actions in the unrelated and the verb-
related conditions in Korean.

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Picture'Onset'
Korean:'Unrelated'
Subject"
Object"
Verb"
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Picture'Onset'
Korean:'Verb8related'
Subject"
Object"
Verb"

31
Fig. 6, Continued

The fact that verb distractors attracted more looks to the verb regions in the verb-
related condition than in the unrelated condition is consistent with Huettig and Altmann’s
(2005) finding that the objects semantically related to the previously presented words
drew more looks. This suggests that verb distractor words in English and Korean were
associated with depicted actions strongly enough to draw significantly more looks to the
verb regions. But the verb interference effects in English and the lack thereof in Korean
suggest that only English speakers retrieved verb lemmas. The fact that looks to the verb
regions exceeded looks to the subjects or objects early in the image display in English but
not in Korean further suggests that verbs play a more important role in English
production than in Korean production.

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
2200"
2400"
2600"
2800"
3000"
Propor%on'of'Looks'
ms'from'Picture'Onset'
Korean:'Object8related'
Subject"
Object"
Verb"

32
Looks to objects. Unlike verb distractors, object distractors did not significantly
attract more looks to the objects early in the image display as compared to unrelated
distractors. Two-tailed paired t-tests revealed that English speakers tended to look more
at objects during 2000 – 2200 ms in the object-related condition than in the unrelated
condition (t
1
(17)=1.9358, p=.069; t
2
(5)=1.5755, p=.176). Korean speakers looked
significantly more at objects during 800 – 1200 ms after picture onset in the object-
related condition (Table 1).

Table 1
Paired t-tests on the mean proportion of looks to objects in the object-related and
unrelated conditions during 800-1200 ms

Additionally, looks to objects in the object-related condition do not follow U-shaped
patterns in other conditions; in both English and Korean, looks to objects tend to increase
during the first 3000 ms after picture onset. The fact that object distractors did draw more
looks to objects than unrelated distractors and looks to objects continued to increase
suggests that object distractors were effective in English and Korean.
However, unlike Korean speakers, English speakers fixated objects earlier in the
object-related condition (1800 ms) than in the unrelated condition (2100 ms). Korean
df
1
t
1
p
1
df
2
t
2
p
2
800–1000 ms 15 -2.690 <.05 5 -3.837 <.05
1000–1200 ms 15 -2.942 <.05 5 -2.386 =.062

33
speakers fixated objects at around 1900 ms regardless of the condition. The fact that
English speakers did not utter objects earlier despite their earlier fixation on objects in the
object-related condition further shows that object distractors interfered with English
speakers’ selection of object lemmas.

2.3 Discussion
Using a picture-word interference paradigm, the present experiment examined
whether transitive verb lemmas and their object lemmas were retrieved before speech
onset in English and Korean. Because distractor words semantically related to depicted
actions are known to delay utterance latencies, the picture-word interference task can
determine whether target verbs are accessed at the time of the word presentation.
The analyses of utterance onset latencies showed that verb-related distractors
significantly delayed utterance onset latencies in English. The verb interference effects in
English are consistent with a lexical account which suggests that lexical items such as
verbs can determine syntactic structures. Consistent with the analyses of utterance onset
latencies, English speakers also tended to look more at the verb regions early in the mage
onset even in the unrelated condition. But importantly, English speakers fixated the verb
regions before fixating the subjects. The early fixation on the verb regions may be
associated with the fact that the verbs occur right after the subjects in English (SVO word
order), but it alone does not fully account for English speakers’ eye movement patterns.
More specifically, if the SVO word order of English is solely responsible for the eye
movement patterns, English speakers should fixate the subjects before the verb regions,

34
which was not the case. Thus, the significant verb interference effects and early looks to
the verb regions taken together suggest that verbs play an important role in English
production. On the other hand, verb-related distractors did not delay utterance onset
latencies in Korean. Unlike English speakers, Korean speakers did not look more at the
verb region than subjects or objects in any condition. Yet, they looked significantly more
at the verb regions in the verb-related condition than in the unrelated condition. This
suggests that the absence of verb interference effects is not due to the ineffectiveness of
verb distractors but rather suggest that verbs may not be crucial for lexical-structural
coordination in Korean compatible with a structural account.
Unlike verb distractors, object distractors did not attract early looks to objects, but
looks to objects continued to increase in the object related condition unlike in the
unrelated condition in both English and Korean. This suggests that semantic relatedness
between object distractors and objects had an influence on eye movements. However,
object distractors did not delay utterance (subject) onset latencies in English and Korean,
suggesting that object lemmas are not retrieved before English and Korean speakers
started to speak. The fact that they significantly delayed verb and object onset latencies
only in English suggests that object lemmas are retrieved after English speakers produced
subjects. Given that English speakers started to fixate objects earlier in the object-related
condition, this suggests that English speakers’ production of objects were interfered by
object distractors. On the other hand, Korean speakers did not fixate objects earlier in the
object related condition. The absence of verb and object interference effects on Korean
speakers’ utterance latencies combined with the fact that looks to verb regions did not

35
exceed looks to subjects and objects early in image onset suggests that Korean speakers
may formulate a structural plan based on the relations between scene elements and may
coordinate eye movements and lemma retrieval according to their plan, compatible with a
structural account.
However, the absence of verb and object retrieval in Korean is also consistent
with highly incremental, word-sized grammatical planning found in Griffin (2001) and
Brown-Schmidt and Konopka (2008). Griffin (2001) monitored speakers’ eye movements
while they produced an utterance such as ‘The clock and the television are above the
needle’ in response to displays of three objects. By varying codability (the number of
alternative names) and frequency of the nouns, Griffin examined whether speakers
retrieved lemmas or phonological forms of the objects; codability is known to affect
retrieval of the lemma, whereas frequency is known to affects retrieval of the
phonological form. Griffin found that only the frequency of the first word affected the
utterance latency or gaze duration to the first picture. Effects of the codability and
frequency of the second and third nouns emerged only when the gaze moved to the
second and the third pictures. This suggests that speakers incrementally select and
phonologically encode nouns.
Similarly, Brown-Schmidt and Konopka (2008) showed that a noun phrase can be
planned in word-sized units by investigating the coordination of gaze and speech during
the production of nouns and adjectival modifiers in English (the small butterfly) and
Spanish (la mariposa pequeña, lit. ‘the butterfly little’). In their experiment, participants
were asked to describe a target picture which was presented with a contrast picture that

36
differed in size, color or number. When fluent pre-nominal adjectives in English and
post-nominal adjectives in Spanish were compared, fixations to the size-contrast picture
occurred later in Spanish than in English, suggesting that adjectives are retrieved later in
Spanish. This suggests that speakers can encode and retrieve each lexical item
sequentially.
The lexically-sized production in Griffin (2001) and Brown-Schmidt and
Konopka (2008) suggests that speakers could begin utterances with a noun even without
a structural plan and incrementally build sentence structures from it. That is, speakers’
formulation of sentence structure could depend on retrieval of nouns rather than verbs,
which is in line with a lexical account. Thus, the lack of verb access in the present
experiment per se does not entirely exclude a lexical account in lexical-structural
coordination in Korean.
In the following chapters, I investigate whether lexical-structural coordination in
English and Korean are driven by retrieval of a noun consistent to a lexical account or
speakers’ structural frame by manipulating lexical (Chapters 3 and 4) and structural
accessibility (Chapters 4 and 5). I derive hypotheses from grammatical properties of
English and Korean and test them by examining English and Korean speakers’ choice of
subject and syntactic choice and their eye movements.

37
Chapter 3
Lexical-Structural Coordination in English and Korean:
Evidence from Semantic Priming

3.1 Introduction
The results of the picture-word interference experiments on English and Korean
in Chapter 2 suggest that selection of verb lemmas is not necessary for sentence
formulation. That is, verbs do not necessarily guide coordination of lexical-structural
coordination in English or Korean although in principle they could by mapping thematic
roles to grammatical functions. Then, in the absence of verb access, how do English and
Korean speakers integrate lexical and structural information? In this chapter, I present
semantic priming experiments that investigated the effects of noun accessibility on
English and Korean speakers’ choice of subject and syntactic structure. The results
suggest different mechanisms of lexical-structural coordination in English and Korean; in
English where grammatical functions are correlated with word order, lexical accessibility
manipulated via semantic priming significantly influenced speakers’ choice of subject
and syntactic structure, consistent with the lexical account. On the other hand, in Korean
where grammatical functions are indicated by case particles, lexical accessibility did not
affect speakers’ choice of subject or structure, which is compatible with a structural
account.

38
3.1.1 Properties of English and lexical-structural coordination
Coordination of lexical items and sentence structures is mediated by grammatical
function assignment (e.g., Ferreira & Slevc, 2007, Jaeger & Norcliff, 2009). For example,
speakers can use either an active or a passive structure to represent an event where a dog
is biting a policeman in Fig. 7. However, an active structure involves linking an agent
entity dog to the subject function (e.g., A dog is a biting a policeman), whereas a passive
structure involves linking a patient entity policeman to the subject function (e.g., A
policeman is being bitten by a dog).
Languages differ in how they assign grammatical functions. As noted in Chapter
1, English defines grammatical functions of arguments in terms of their relative order.
Because grammatical functions are highly correlated with word order in English, a
speakers’ structural choice entails an order of mention; speakers producing a passive
structure for their description of Fig. 7 will almost always start their utterances with the
patient entity policeman. This effect of a sentence structure on the order of mention
suggests that English speakers may use a structural plan to control lexical retrieval.

39

Fig. 7. A scene depicting a biting event

In fact, Griffin and Bock (2000) provides evidence for structurally incremental
production in English. They presented participants with depicted events as Fig. 7, which
could be described with either an active or a passive structure, and compared the eye
movements when participants performed linguistic (picture description) and non-
linguistic (patient detection) tasks. They found that in the patient detection task,
participants fixated the patient entity approximately after 300 ms of image display.
Because the patient detection requires the understanding of relations among scene entities
(e.g., who did what to whom), Griffin and Bock suggest that the initial viewing serves to
extract the event structure. The initial viewing period was also observed in the picture
description task; during the first 300 ms after picture onset, participants did not fixate the
entity that they would mention first more than the entity that they would mention later.
The fact that the first mentioned entity did not draw the most or the earliest fixations
suggests that an individual element does not serve as a starting point for production.
Given that the selection of the first noun phrase occurred within the same time frame as
the detection of the patient entity, Griffin and Bock suggest that participants formulate a

40
structural plan from the causal or relational structure of an event and coordinate lexical
retrieval accordingly (for further evidence for structural incrementality in English, see
Bock, Irwin, Davidson, & Levelt, 2003; Bock et al., 2004; Kuchinsky et al., in press).
However, studies have also shown that lexical incrementality can take precedence
over structural incrementality in English. Because grammatical functions are highly
correlated with word order in English, a speaker’s choice of what to mention first can
influence a sentence structure; speakers who mention the patient entity policeman first in
their description are more likely to produce a passive sentence than those who mention
the agent entity dog first. That lexical items indeed play a powerful role in constraining
structures in English is shown by the fact that English speakers tend to build up sentences
starting with the most accessible lexical item. If a word is made more accessible, it is
more likely to be mentioned early and assigned to the subject position. For example, in a
picture-description task, Bock (1986a) found that speakers tended to choose a structure
which allowed a semantically primed word to be mentioned as the subject of the sentence.
Thus, if speakers whose task is to describe Fig. 7 are primed with criminal, a semantic
associate of policeman, they are more likely to produce a passive sentence such as The
policeman is being bitten by a dog. This shows that lexical accessibility, as manipulated
by means of semantic priming, has an effect on sentence construction.
Lexical accessibility can be also manipulated by perceptual salience. Using a
picture-description task similar to Bock (1986a), Gleitman, January, Nappa and
Trueswell (2007) showed that if a character is made more accessible by an attention-
capturing flash, that character is more likely to be mentioned first in the sentence and to

41
be realized as the grammatical subject than an unprimed character. For example, when
speakers are presented with a subliminal attention-capturing flash that cues the location
where the policeman appears immediately afterwards, they are more likely to utter a
passive sentence than when the flash cues the location of the agentive character, dog.
Gleitman et al. suggest that – despite their non-linguistic nature – attention-capturing
cues affect the choice of sentence forms by increasing the lexical accessibility of the
primed item, akin to the process triggered by lexical primes: By drawing initial attention
and looks to an item, a perceptual prime immediately increases accessibility of the
corresponding lemma and lexeme. In Gleitman et al. (2007), fixation patterns within the
first 200 ms of image display also predicted what participants subsequently said first: The
character that was looked at the most during the initial 200 ms was also the character who
participants tended to say first. This suggests that an individual lexical item can also
serve as a starting point for production in English. In sum, the studies of Bock (1986a)
and Gleitman et al. (2007) provide evidence that lexical incrementality can dominate
lexical-structural coordination in English, particularly when accessibility of an individual
lexical item is increased.
That both structural and lexical incrementality can control lexical-structural
coordination in English is consistent with the typological properties of English; the close
relation between grammatical function and word order in English enables both sentence
structures and lexical items to exert influence on each other. The findings in English raise
the question of whether lexical-structural coordination in typologically different
languages such as Korean is similarly influenced by lexical and structural incrementality.

42
In light of the cross-linguistic evidence for language-specific production mechanisms, the
present study investigates the mechanisms of lexical-structural coordination in Korean by
manipulating lexical accessibility via semantic priming.

3.1.2 Properties of Korean and lexical-structural coordination
Unlike English, Korean marks grammatical functions by case particles. Because
case particles provide information about the grammatical functions, word order is
relatively free in Korean. For example, in sentences ex (4a-d) representing a chasing
event where a dog is the agent entity and a cat is the patient entity, either the agent dog or
the patient cat can take the sentence-initial position regardless of the sentence structure
(active/passive).

(4) a. Kay-ka koyangi-lul ccochnunta. [Active: Agent-Patient-Verb]
dog-NOM cat-ACC chase
‘A dog is chasing a cat’

b. Koyangi-lul kay-ka ccochnunta. [Active: Patient-Agent-Verb]
cat-ACC dog-NOM chase
‘A dog is chasing a cat’

c. Koyangi-ka kay-hantey ccochkinta. [Passive: Patient-Agent-Verb]
cat-NOM dog-BY be being chased

43
‘A cat is being chased by a dog’

d. Kay-hantey koyangi-ka ccochkinta. [Passive: Agent-Patient-Verb]
dog-BY cat-NOM be being chased
‘A cat is being chased by a dog’

Due to the relatively free word order afforded by the case-marking system, Korean
speakers have at least two structural choices when they describe Fig. 7 beginning with the
patient entity, policeman. First, just like English speakers, they could produce a passive
sentence in (5a), where the patient policeman occurs in the grammatical subject position.
Secondly, due to word order flexibility, Korean speakers can produce a scrambled
sentence with Object-Subject-Verb order, as in (5b).

(5) a. Passive sentence
Kyengchal-i kay-hantey mwul-li-ess-ta.
policeman-NOM dog-DAT bite-PASS-PAST-DC
‘A policeman was bitten by a dog.’

b. Scrambled active sentence (Object-Subject-Verb)
Kyengchal-ul kay-ka mwul-ess-ta.
policeman-ACC dog-NOM bite-PAST-DC
‘A dog bit a policeman’

44

Because Korean speakers are allowed to produce either structure, lexical accessibility
may have a greater influence on Korean speakers’ choice of the sentence-initial noun or
syntactic structure.
However, note that word order does not provide reliable clue about sentence
structure in Korean unlike in English. For example, in (4a), the agent entity dog is the
sentence-initial noun phrase, whereas the sentence-initial noun phrase in (4b) is the
patient entity cat. Yet both sentences have an active structure, because dog is marked
with nominative case and cat is marked with accusative case in both sentences. Similarly,
(4c) and (4d) have a different initial noun. But they both have a passive structure, because
the nouns carry the same case particles. In contrast, although sentences (4b) and (4c) have
the same initial noun cat, they differ in structure; (4b) has an active structure, whereas (4c)
has a passive structure. This is because dog carries different case markers in (4b) and
(4c): in (4b) it is marked with accusative, whereas in (4c) it is marked with nominative.
The fact that the order of lexical items does not determine sentence structure in
Korean as shown in ex (4a-d) suggests that lexical incrementality is not likely to guide
lexical-structural coordination in Korean. On the contrary, based on the interdependence
of case particles and sentence structures, we hypothesize that structural incrementality
may predominate over lexical incrementality in Korean lexical-structural coordination.
The primary reason for hypothesizing structural incrementality for Korean is that a
structural plan is necessary for case assignment (e.g., what case to assign to which
entity). Because structural plans include information about functional or hierarchical

45
dependencies between lexical items and their ordering (i.e., where to start an utterance
and how to continue, Kuchinsky et al., in press; Konopka, 2009), they provide guidance
for case assignment. If Korean speakers need a structural plan to assign grammatical
functions via case, they are predicted to formulate structural plans and control subsequent
lexical retrieval accordingly.
There are at least two other reasons to believe that Korean speakers formulate
structural plans to control subsequent lexical retrieval. First, immediately starting with
whatever lexical item is more accessible without construing a structural plan is likely to
impede fluent speech in Korean, particularly if it results in the use of a marked passive
structure. This is because passive sentences are more semantically restricted and
structurally marked than active sentences in Korean (more so than in English). For
instance, only a subset of transitive verbs can be passivized in Korean. The sentence in
ex (6) shows that verbs like speak which can be easily passivized in English cannot be
passivized in Korean (see Wu, 1997 for a list of verbs that can be passivized in Korean).

(6) * Mikuk-eyse yenge-ka malhacinta.
the U.S.-in English-NOM be spoken
‘English is spoken in the U.S.’

In addition, the subject of Korean passive is often required to be animate, as shown by the
asymmetry between (7a) and (7b) (see Park, 2005; Oshima, 2006, among others for
discussion of the syntactic and semantic or pragmatic constraints for passive structures).

46
English passives are not nearly as constrained, as shown by the fact that English
translations of both sentences are completely acceptable.

(7) a. So-ka nuktay-hantey mek-hi-ess-ta.
cow-NOM wolf-DAT eat-PASS-PAST-DC
‘A cow was eaten by a wolf.’

b. *Sokoki-ka nuktay-hantey mek-hi-ess-ta.
beef- NOM wolf-DAT eat-PASS-PAST-DC
‘Beef was eaten by wolf.’ [The subject ‘beef’ is not an animate entity]

Korean speakers could produce object-initial active sentences with non-canonical
scrambled word order (OSV) rather than passive sentences, but scrambled sentences are
also marked as compared to canonical active sentences; they are much less frequent
(Kwon, Polinsky, & Kluender, 2006) and harder to process than canonical active
sentences, and only licensed in certain discourse contexts (Jackson, 2008). Given the
constrained nature of passives and scrambled OSV sentences in Korean, beginning an
utterance with whatever lexical item is more accessible in the absence of a structural plan
may well hinder the continuation or completion of the utterance.
If lexically incremental production is likely to cause disruptions in Korean and
speakers tend not to start an utterance that they cannot continue or finish (Bock et al.,
2004), then it is reasonable to hypothesize that Korean speakers first generate a structural

47
plan that provides information about where to start and how to continue an utterance.
Another reason to believe that Korean speakers formulate structural plans to
control subsequent lexical retrieval comes from existing work on comprehension. Choi
and Trueswell (in press) showed that during real-time comprehension, Korean speakers
use structural (morpho-syntactic) cues to predict upcoming verbs, using sentences like (8)
containing a temporal ambiguity: the case particle –ey is ambiguous between the genitive
and the locative particle. Thus, the sentence-initial napkin-ey can be interpreted either as
a modifier of frog (‘the frog on the napkin’) or a goal of the upcoming verb (‘put the frog
on the napkin’). The temporal ambiguity is resolved either in favor of the modifier
interpretation upon hearing pick up (‘pick up the frog on the napkin’) or the goal
interpretation upon hearing put (‘put the frog on the napkin’).

(8) Naypkhin-ey kaykwuli-lul nohu-sey-yo / cipu-sey-yo.
Napkin-ey frog-ACC put / pick up
Put / Pick up the frog on the napkin. (from Choi & Trueswell, in press)

Choi and Trueswell (in press) showed that Korean speakers often initially consider a
locative construction and anticipate verbs like put which require a goal thematic role
because –ey is much more frequently used as the locative particle than the genitive
particle. This finding that case particles are used to constrain upcoming lexical items
provides yet another piece of evidence for structural control of lexical items in Korean.

48
3.1.3 Overview of the experiments and predictions
The typological properties of Korean and prior work on comprehension strongly
suggest that structural incrementality will predominate over lexical incrementality during
sentence production in Korean; if Korean speakers formulate a rudimentary structural
plan and use it to control lexical retrieval, they should determine the starting point for
utterance or syntactic structure based on their structural plan rather than based on lexical
accessibility. Thus, I hypothesize that priming a patient entity is not expected to increase
the use of passive sentences or utterances starting with patient entities in Korean. On the
other hand, in English where lexical incrementality can guide production, lexical
incrementality should have significant influence on speakers’ choice of subject and
syntactic structure as reported in Bock (1986a) and Gleitman et al. (2007). To test the
hypothesis, I conducted a picture-description experiment in English and Korean. The
participants’ task was to describe pictured events, which could be described with either
active or passive sentences. I manipulated lexical accessibility in English and Korean by
presenting a semantic prime word before each picture. Prime words were semantically
associated with one of the two scene characters in the picture. The use of a semantic
prime is similar to Bock (1986a), but participants in the present experiment looked at the
prime word without reading it aloud.

49
3.2 Experiment 2

3.2.1 Method
Participants
Twenty-four native speakers of English and twenty-four native speakers of
Korean) from the University of Southern California participated in the experiment for
$10 per hour.

Stimuli
Picture Norming Study. In order to select target images for the experiment, a norming
study was conducted over the internet on a separate group of 24 native English speakers
and 25 native Korean speakers. Participants wrote a single-sentence description for 27
images. Each image displayed a simple color drawing depicting an adverse transitive
event involving two animate characters (e.g., biting, kicking).
Following Gleitman et al. (2007), an image was selected to be a target in the main
experiments if both active and passive constructions occurred at least once among the
descriptions of the pictures (for baseline rates of each item, see Appendix: Baseline Rates
of Actives vs. Passives in the Norming Study). In addition, when selecting the items, we
made sure that the pair of characters in each item was sufficiently distinct from each
other in order to prevent a semantic prime for one character from priming the other. This
resulted in the selection of 16 target images.
Experimental stimuli: The main experiments used the 16 target images chosen in the

50
norming study, as well as 26 filler images. The target images were always preceded by a
semantic prime word which was associated with one of the two scene characters. English
prime words were selected based on the University of South Florida Free Association
Norms (Nelson, McEvoy, & Schreiber, 1998) and Korean primes were chosen based on
Park (2004). The filler images were similar to the targets in style but they could only be
described with active sentences. The fillers were preceded by either words or non-words
(11 fillers were preceded by words and 15 by non-words). The words that preceded filler
images were not related to any of the characters in the scene (e.g.,‘watch’ for an image of
a nun reading a book) and the words and non-words were distributed so that half of them
occurred in the first half of the list and the other in the latter half of the list.

Procedure and design
Participants were seated in front of a 21-inch CRT monitor and instructed to
orally describe the pictured event in one sentence using all characters present in the
picture. Before proceeding to the main experiment, an example item and four practice
items were presented.
On each trial, participants were first presented with a crosshair for 500 ms. A semantic
prime word immediately followed the crosshair and stayed on the screen for 200 ms. The
primes were presented in 120-point Batang font in English or Korean and after 200 ms
they were replaced by a blank screen. To make sure that participants pay attention to and
process the primes, they were asked to make a lexical decision by pressing the right
button of a game controller when they saw a real word and the left button when they saw

51
a non-word. Immediately after the lexical decision task, a target picture appeared and the
participants’ speech was recorded with a desk microphone. Participants’ eye movements
were also recorded with an SR Research Eyelink II headmounted eye-tracker, but they
are discussed in Chapter 7. At the end of the description, participants pressed a controller
button to proceed to the next trial. See Fig. 8 for a demonstration of how the stimuli were
presented.

500 ms 200 ms
Lexical decision Describe the scene

Fig. 8. English display sequence for Experiment 2. Participants saw the crosshair for 500
ms, the semantic prime for 200 ms followed by a blank screen, and then the scene to be
described.

The priming condition of a scene character (agent or patient) and the location of
the agent and patient (left or right) were counterbalanced across two stimulus lists so that
agents and patients on the right and left were primed equally in each list. There were two
additional lists in which the order of the items was reversed. This was to make sure that

52
the order of presentation did not have any effects on the description of the pictures.
Participants were randomly assigned to one of these four lists.
After the experiment, participants were asked what they thought the experiment
was about. Most participants said that they had noticed that some words were related to
the immediately following pictures, but no participant was able to guess the purpose of
the study.

Coding and Analyses
Participants’ speech was transcribed and analyzed for structural choice (active
SOV sentences, passive sentences, or scrambled OSV sentences). Trials containing
disfluencies (fillers or repairs) were included unless they altered the referential and
structural choices (e.g., “A doctor, uh, a bear is hitting a doctor” was not included
because the speaker corrects the choice of the first referent). We excluded utterances with
conjoined NP subjects (e.g., “a bear and a doctor”), as NP conjunctions are processed
differently from active and passive sentences (e.g., Branigan, Pickering, & Tanaka, 2008;
Onish, Murphy, & Bock, 2008). All utterances not containing two NPs were also
excluded. In total, less than 3 % of the trials (11 out of 384 in English and 11 out of 384
in Korean) were removed for one of these reasons. For the remaining trials, utterance
onset latencies were manually determined using the phonetic software package Praat
(Boersma & Weenink, 1992). Thirty-six trials with outlier latencies were further removed
from the analyses using the MAD-median rule (e.g., Wilcox, 2012).
The results were analyzed with mixed-effects logit models (Bates & Sarkar, 2007;

53
Baayen, 2008). To specify the structure of random effects, fully crossed and fully
specified random effects were reduced until the model converged. Then, following
Baayen (2008), only those effects which were found to contribute significantly to the
model were included in the final analyses. All final models contained prime condition
(agent vs. patient) as a fixed effect and random intercepts for subject and item.

3.2.2 Results
Fig. 9 shows the proportion of utterances in the active voice (e.g., “a dog is biting
a policeman”) in English and Korean as a function of whether the agent or the patient
scene character was semantically primed.

Fig. 9. Effects of semantic primes on English and Korean speakers’ choice of actives and
passives.
0.7$
0.75$
0.8$
0.85$
0.9$
0.95$
1$
English( Korean(
Propor0on(of(Ac0ve(Sentences(
Agent(Prime(
Pa0ent(Prime(

54

When patient entities were primed, English speakers produced significantly less
active sentences (and more passive sentences) than when agent entities were primed
(97% vs. 93%) (β=2.343, |z|=2.228, p<.05, SE=1.051). This small but reliable effect of
semantic priming (4%) is consistent with the findings of Bock (1986a). For Korean
speakers, however, semantic priming did not have a significant effect on syntactic choice
(p>.1): Priming the patient did not increase the production of passive sentences.
Participants produced active sentences most of the time regardless of which entity was
primed (81% in the agent prime condition and 81% in the patient prime condition).
Additionally, we also found that semantic priming did not influence word order in
Korean (p>.1); priming the patient did not increase the production of utterances starting
with the patient. Participants began with agent entities most of the time (79% in the agent
prime condition and 81% in the patient prime condition). Non-canonical OSV sentences
were also rare, occurring only 4 times out of 373 utterances (1%).
Consistent with what Gleitman et al. (2007) and Flores d’Arcais (1975) found for
English, when speakers produced passive sentences, they initiated utterances more slowly
(see Fig. 10): The average latencies were significantly longer for passives than actives in
both English (2434 ms for passives and 1894 ms for actives, β=365.5, t=2.493, p<.05,
SE=146.6) and Korean, (3194 ms for passives and 2462 ms for actives, β=357.6, t=3.273,
p<.01, SE=109.3). Given that the active structure is more frequent, less complex,
acquired earlier and more accessible than the passive structure, the greater latencies of
passive utterances are expected.

55

Fig. 10. Utterance latencies for active and passive sentences in English and Korean in the
semantic priming experiments.

3.3 Discussion

The present chapter investigated the effects of semantic priming on English and
Korean speakers’ syntactic choices in order to assess the role that lexical and structural
incrementality play in two typologically different languages. Lexical incrementality
emphasizes the roles of individual lexical items in the choice of sentence structures. If
speakers are engaged in lexically incremental production, they should produce a structure
that allows early accommodation of more accessible lexical items. Thus, semantically
priming patient entities should increase the use of passives structures or sentences
starting with patient entities. In contrast, structural incrementality suggests that speakers
determine a starting point of utterance or syntactic structure based on their structural plan
0"
500"
1000"
1500"
2000"
2500"
3000"
3500"
English( Korean(
U/erance(Latency((ms)(
Ac8ve(
Passive(

56
rather than based on lexical accessibility. Thus, it predicts that speakers’ choice of subject
or syntactic structure is not influenced by lexical accessibility.
Our findings suggest that English speakers’ production is significantly influenced
by lexical accessibility, whereas Korean speakers’ is not. When patient entities were
primed, English speakers were more likely to begin their utterances with patient entities,
producing passive sentences. This suggests that English speakers can build a sentence
structure by starting with more accessible lexical items, corroborating earlier work (e.g.,
Bock, 1986a; Gleitman et al., 2007). In contrast, Korean speakers’ syntactic choice was
not influenced by lexical accessibility; priming patient entities did not increase the use of
passive sentences or sentences starting with patient entities. This is expected if Korean
speakers formulate structural plans and control lexical retrieval accordingly.
It is worth noting that although word order in Korean is flexible and allows
speakers to put patient entities in sentence-initial position without passivization, Korean
speakers rarely produced scrambled sentences. Given that the experiment did not include
any discourse contexts, the rare occurrences of scrambled sentences fit with prior work
that scrambled orders are only licensed in particular discourse context (e.g. Jackson,
2008).
In sum, the effects of semantic priming in English suggests that lexical items can
guide lexical-structural coordination, whereas the absence of semantic priming in Korean
suggests that the coordination process does not depend on the accessibility of lexical
items but may depend on a structural plan.

57
Chapter 4
Lexical-Structural Coordination in English and Korean:
Evidence from Attention-capture Cueing

4.1 Introduction
The finding that semantic priming significantly influenced English speakers’
choice within active vs. passive structures but not Korean speakers’ syntactic choice in
Chapter 3 suggests that lexical incrementality can guide lexical-structural coordination in
English, whereas structural incrementality may guide the coordination process in Korean.
Chapter 4 extends the investigation of lexical-structural coordination in English and
Korean by manipulating lexical accessibility via attention-capture cues. One may suspect
that the absence of semantic priming in Korean may be because semantic prime words
were not as strongly associated to Korean as in English. This is unlikely given that most
Korean speakers reported of being aware that some preceding words were related to the
following pictures, and they were as likely to misname scene character such as fox with
prime words such as wolf as English speakers. However, to eliminate the possibility of
semantic association strength between words and pictures, the present chapter uses the
attention-capture cues to manipulate lexical accessibility. Additionally, comparing the
effects of semantic priming and attention-capture cueing under the same experimental
set-up allows us to estimate the magnitude of linguistic and non-linguistic manipulation
of lexical accessibility on production.

58
4.2 Experiment 3

4.2.1 Method

Participants
Forty-eight participants (24 native speakers of English and 24 native speakers of
Korean) from the University of Southern California participated in the experiment.
Participants received $10 per hour.

Stimuli
The stimuli were the same as those used in Chapter 3. In Chapter 4, however,
lexical accessibility is manipulated by visual attention-capture cues rather than
semantically associated words. Target images were preceded by a attention-capture cue
which was located where one of the scene characters (either the agent or the patient of the
depicted action) would appear. Filler images were also preceded by an attention-capture
cue but it was located either to the right, left or center of the screen and was not
informative of the location of scene characters.

Procedure and design
The procedure closely matched that of Gleitman et al. (2007). As in the semantic
priming experiments (reported in Chapter 3), participants were seated in front of a
computer monitor and instructed to orally describe the pictured event in one sentence

59
using all characters in the picture. On each trial, participants first focused on a crosshair
which was placed neutrally between the two characters. Then they were presented with
an attention-capture cue which consisted of a 0.5 x 0.5-inch black square against a white
background. The attention-capture cue appeared for 60 ms, and was immediately
followed by a picture. . The duration of the attention-capture cue was chosen to be long
enough to attract eye-movements but short enough to not be consciously perceptible.
shows how the stimuli were presented. The participants’ speech was recorded with a desk
microphone. At the end of each description, participants pressed a controller button to
proceed to the next trial.
Participants’ eye gaze data were also collected with an SR EyeLink II
eye-tracker. However, we do not discuss the eye data in this chapter, other than analyses
that we conducted on participants’ initial gaze patterns to assess effectiveness of the
attention-capture cues. The details of the eye gaze data are discussed in Chapter 7. At the
end of the description, participants pressed a controller button to proceed to the next trial.

60

500 ms 60 ms Describe the scene
Fig. 11. Display sequence for attention-capture curing experiment. Participants saw the
crosshair for 500 ms, the attention-capture cue for 60 ms, and then viewed the scene and
described the event.

The priming condition of a scene character (agent or patient) and the location of
the agent and patient (left or right) were counterbalanced across two stimulus lists. There
were also two additional lists with reversed item order.
After the experiment, participants were asked what the experiment was about and
whether they noticed any flash or disruption in the presentation of the scenes. No
participant correctly guessed the purpose of the study or reported being aware of the
attention-capture cue.

Coding and Analyses
The criteria for coding and analyses were identical to those used in the semantic
priming experiments reported in Chapter 3. This resulted in the exclusion of less than 4%
of the trials (13 out of 384 trials in English and 8 out of 384 in Korean). Trials with
outlier latencies were removed using the MAD-median rule (29 trials in English and 31

61
trials in Korean).

4.2.2 Results

To assess the effectiveness of the attention-capture cues, analyses were performed
on participants’ eye gaze data. The attention-capture manipulation was effective both in
English and Korean; participants initially fixated the cued character significantly more
than would be expected by chance, consistent with the findings of Gleitman et al. (2007).
The analyses of one-sample, two-tailed t-tests shows that participants’ initial fixations to
the cued character (rather than the uncued character) are reliably greater than chance
(Table 2) (mean proportion of trials on which participants fixated the cued character first
= 0.70, 95% CI=±0.06 in English, 0.73, 95% CI=±0.04 in Korean).

Table 2
Effectiveness of attention-capture manipulation in Experiment 3: One-sample, two-tailed
t-tests on the mean proportion of trials where participants first looked at the cued
character (Compared to chance level of 0.5)
df
1
t
1
p
1
Df
2
t
2
p
2
English 23 6.46 <.001 15 6.66 <.001
Korean 23 10.65 <.001 15 7.50 <.001

62

The attention-capture manipulation, however, did not have significant effects on
the production of active vs. passive sentences in either English or Korean (p>.1). Fig. 12
presents the proportion of utterances in the active voice (e.g., “a dog is biting a
policeman”) in English and Korean respectively when agent and patient scene characters
were cued.

Fig. 12. Effects of attention-capture cues on English and Korean speakers’ production of
actives.

As can be seen in Fig. 12, English and Korean speakers mostly produced active
sentences regardless of which character was cued. English speakers produced active
sentences 96% of the time in the agent and the patient prime condition, and Korean
speakers produced active sentences 87% of the time in the agent prime condition and
0.7$
0.75$
0.8$
0.85$
0.9$
0.95$
1$
English( Korean(
Propor0on(of(Ac0ve(Sentences(
Agent(Prime(
Pa0ent(Prime(

63
83% in the patient prime condition. Although Korean speakers exhibited a numerical
trend to produce less active sentences (more passive sentences) when patient entities were
cued, this trend was not statistically significant. Korean speakers also did not produce any
non-canonical OSV sentences.
Fig. 13 plots the average onset latencies for active and passive sentences in
English and Korean. As in the semantic priming experiments, producing passive
sentences led to significantly delayed utterance onsets in both English and Korean: The
onset latencies for active and passive sentences were 1954 ms and 2428 ms in English
(β=507.77, t=3.468, p<.001, SE=146.37) and 2522 ms and 2873 ms in Korean (β=227.0,
t=1.945, p=.05, SE=116.7).

64

Fig. 13. Utterance latencies for active and passive sentences in English and Korean in the
attention-capture experiments.

4.3 Discussion

Chapter 4 extended the investigation of the coordination of lexical and structural
information by examining effects of attention-capture cueing on English and Korean
speakers’ syntactic choice. If speakers are engaged in lexically incremental production
where lexical items can play an important role in determining a syntactic structure,
cueing patient entities should increase the use of passives structures as reported in
Gleitman et al. (2007). In contrast, if speakers are engaged in structurally incremental
production where a structural plan controls lexical retrieval rather than the other way
around, their syntactic choice should not be influenced by attention-capture cueing.
0"
500"
1000"
1500"
2000"
2500"
3000"
3500"
English( Korean(
U/erance(Latency((ms)(
Ac8ve(
Passive(

65
I found that the attention-capture manipulation of lexical accessibility did not
influence English or Korean speakers’ syntactic choice. My finding that neither semantic
nor visual attention-capture cueing affected Korean speakers’ syntactic choice provides
evidence against lexical accessibility in Korean but is compatible with the hypothesis that
structural incrementality predominates over lexical incrementality in the production of
Korean.
As for English, the lack of attention-capture effects diverges from the findings of
Gleitman et al. (2007), and also contrasts with the semantic priming effects observed for
English in Chapter 3. However, this may be because perceptual salience has been claimed
to have a relatively weak influence on word order (see Bock et al., 2004 for a review).
For example, Osgood (1971) and Flores d’Arcais (1975) found that speakers hardly ever
used passives to initiate utterances with what were presumably the perceptually more
salient elements. Additionally, Kuchinsky (2009) found that attention-capture effects
interacted with event codability (i.e., the ease of naming events with verbs); speakers
were more likely to start with the cued characters when the events were harder to name.
These findings suggest that the initial attentional advantage offered by attention-capture
cueing might be easily washed away or overturned by linguistic factors during sentence
formulation. Then, the absence of the attention-capture manipulation in English is
perhaps not as surprising. In light of Kuchinsky (2009)’s finding, the different findings
between the current experiment and Gleitman et al. (2007) may be due to differences in
the codability of events. If our images contained more codable events than those of
Gleitman et al., our participants might have been unwilling to begin with cued characters.

66
The fact that attention-capture cueing did not have any significant effects on English
speakers’ syntactic choice – whereas semantic priming did in Chapter 3 – suggests that
the influence of attention-capture cueing on syntactic choice, if it exists, is relatively
weak compared to semantic priming.
In terms of the lexical-structural coordination mechanism, the results of Chapters
3 and 4 as a whole suggest that English speakers can engage in lexically incremental
production; when a lexical item becomes readily accessible via semantic priming, they
are more likely to start utterances with available entities (Chapter 3). But English
speakers were not more likely to start their utterances with the cued entities when lexical
accessibility was manipulated with the attention-capture cues (Chapter 4). Considering
that English speakers were more susceptible to attention-capture cueing when the events
were harder to name, Kuchinsky (2009) suggests that perceptual salience is a last resort
that speakers adopt when they cannot formulate their own structural plan. If the lack of
attention-capture cueing in Chapter 4 is because the events were easy to name as
compared to those used in Gleitman et al. (2007), the semantic priming effects in Chapter
3 provide even stronger support for the lexical incrementality in English; although
English speakers could easily code the event structure and formulate their own structural
plan, the fact that they were still more likely to start with semantically primed entities
suggests that the accessibility of lexical items could guide lexical-coordination in English.
Korean speakers’ syntactic choice, on the other hand, were not susceptible lexical
accessibility; it was not influenced by either semantic priming or attention-capture cueing.
This suggests that Korean speakers are likely to determine the starting point of an

67
utterance or syntactic structure based on their structural plan and do not easily give up the
plan for lexical accessibility.
The claim that structural incrementality guides lexical-structural coordination in
Korean helps us understand why Myachykov and Tomlin (2008) and Myachykov, Garrod,
and Scheepers (2010) did not find attention-capture priming effects in Russian and
Finnish. Russian and Finnish are similar to Korean in that they are case-marking
languages with flexible word order. Furthermore, similar to Korean, the use of passive
sentences is constrained in these languages (e.g., Myachykov, Thompson, Scheepers, &
Garrod, 2011 for a review). Thus, beginning utterances without a structural plan in
Russian and Finnish is also likely to disrupt speech or cause errors. If Russian and
Finnish speakers formulate a syntactic plan to afford a continuance or completion of an
utterance – i.e., engage in structurally incremental processing, it is expected that their
structural choice is not affected by the attention-capture manipulation, just like Korean
speakers’ syntactic choice was not affected by the attention-capture manipulation.
In addition to providing insights into the mechanisms of the lexical-structural
integration in English and Korean, the studies in Chapters 3 and 4 also allow us to
compare the effects of linguistic primes and attention-capture cues on syntactic choice
under the same experimental set-up. The fact that English speakers’ syntactic choice was
significantly influenced by semantic priming but not by attention-capture cueing suggests
that linguistic priming exerts a greater influence on sentence production. The greater
effects of semantic priming can be attributed to the advantage it has in linguistic
processing; because semantic prime words share semantic contents (also phonemes in

68
some cases) with the target words, they could activate the target lemmas or lexemes more
strongly than attention-captures cues. Unlike semantic primes, attention-capture cues do
not share any semantic or phonemic features with the targets but merely draw initial
looks to the targets. The initial looks, however, do not guarantee the activation of target
lemmas and lexemes. Prior evidence suggests that participants do not necessarily retrieve
an object lemma by merely looking at it (e.g., Schriefers, Meyer, & Levelt, 1990; Smith
& Magee, 1980). Furthermore, Kuchinsky (2009)’s finding suggests that the initial
attention-capture advantage on early eye movements can be easily overturned by
linguistic factors such as event codability (i.e., the ease of naming events). This suggests
that linguistic priming has a greater influence on language processing than non-linguistic
salience.
In summary, the findings of Chapters 3 and 4 as a whole suggest that in English,
lexical retrieval can drive sentence structure, whereas in Korean, a structural plan
controls subsequent lexical retrieval. ‘

69
Chapter 5
Structural Incrementality in Korean:
Evidence from Syntactic Priming

5.1 Introduction
The lack of semantic and attention-capturing priming in Korean reported in
Chapters 3 and 4 suggests that lexical information does not drive lexical-structural
coordination in Korean. In order to further investigate lexical-structural coordination in
Korean, Chapter 5 manipulates structural accessibility via syntactic priming.
Previous findings on English suggest that English speakers could formulate a
structural plan either from (i) their perception of event structures (e.g., Griffin & Bock,
2000) or (ii) structural accessibility (e.g., Bock, 1986b). This suggests that Korean
speakers could formulate a sentence structure either based on their perception of event
structures or structural accessibility if structural incrementality guides lexical-structural
coordination in Korean. If syntactic priming affects Korean speakers’ syntactic choice
(e.g., if Korean speakers produce more passive sentences after passive sentence primes
than active sentence primes), it would indicate that Korean speakers could formulate
structural plans using more accessible sentence structures, like English speakers. On the
other hand, the absence of syntactic priming would be compatible with the claim that
Korean speakers’ formulation of structural plans depends on their coding of relational or
causal structure of events rather than structural accessibility.
However, regardless of whether syntactic priming affects Korean speakers’

70
structural choice, we expect that syntactic priming should have significant influence on
Korean speakers’ utterance onset latencies if Korean speakers formulate structural plans
before speech onset. In particular, if a primed structure corresponds to speakers’ syntactic
plan, it should facilitate production, significantly reducing utterance onset latencies. On
the other hand, if a primed structure conflicts with speakers’ structural plan, it should
interfere with production, increasing utterance latencies. For example, passive utterance
latencies should be shorter and active utterance latencies longer when passive structures
are primed than when active structures are primed.

5.2 Experiment 4

5.2.1 Method

Participants
Twenty-two native speakers of Korean from Southern California community
participated in the experiment. Participants received $10 per hour.

Stimuli
The experimental stimuli were the same as those used in semantic priming and
attention-capturing cueing in Chapters 3 and 4. In the syntactic priming experiment,
however, participants were presented either an active or a passive sentence before each
picture. Half of the target images were preceded by an active prime sentence and the

71
other half were preceded by a passive prime sentence. The pairings of sentences with
pictures were random, but care was taken to ensure that the descriptions of pictures did
not employ similar content words from the prime sentences. Fig. 14 shows an example of
a pairing between a prime sentence and a target image. Filler images were also preceded
by sentences but the sentences were composed of predicate adjective or intransitive
constructions that could not elicit the active/passive alternation (e.g., Mary’s hands are
pretty).

Procedure and design
The experiment was run with Paradigm developed by Perception Research
Systems. As in Experiments 2 and 3, participants were instructed to orally describe the
pictured event in one sentence using all characters in the picture. To encourage
participants to process the prime sentences, they were also told that they would perform a
memory task, which would measure how accurately they could later recognize previously
presented sentences.
On each trial, participants first focused on a crosshair for 500 ms. Prime sentences
immediately followed the crosshair and were presented in 28-point Batang font in Korean.
Participants were instructed to pay close attention to the prime sentences and to read them
aloud in order to be able to recognize them later. After reading each sentence, participants
pressed the space bar, which replaced the prime sentence with a blank screen. They then
indicated whether they had encountered the particular sentence before in the session by
pressing a key marked as yes (the F key) or no (the J key). Immediately after the response,

72
a picture appeared and stayed on the screen until participants pressed the space bar at the
end of their description. Fig. 14 shows a trial with a passive prime sentence, Minswu was
caught by Sengho in the second frame. The participants’ speech was recorded with a
head-mounted USB microphone.

500 ms Read aloud Recognition Describe the scene
Fig. 14. Display sequence for Experiment 4. Participants saw the crosshair for 500 ms,
and a prime sentence followed by a blank screen, and the scene to be described.

As in semantic and attention-capture experiments in Chapters 3 and 4, the
location of the agent and patient (left or right) were counterbalanced across two stimulus
lists. There were also two additional lists with reversed item order.
After the experiment, participants were asked what the experiment was about and
whether they noticed any relationships between the sentences and the pictures. No
participant correctly guessed the purpose of the study or reported being aware of the
effects that the sentences they had repeated might have had on their descriptions of the
pictures.

민수가 성호한테 붙들렸다.

73
Coding and Analyses
The criteria for coding and analyses were identical to those used in Experiments 2
and 3. This resulted in the exclusion of less than 9 % of the trials (30 out of 352 trials).
Thirteen trials with outlier latencies were further removed from the analyses using the
MAD-median rule.

5.2.2 Results

Fig. 15 shows the proportion of utterances in the active voice (e.g., a dog is biting
a policeman) in Korean when active and passive structures were primed. As expected,
syntactic priming did not influence production of active versus passive sentences in
Korean (p>.1). Korean speakers mostly produced active sentences regardless of which
sentence structure was primed; they produced active sentences 88% of the time in the
active prime condition and 89% in the passive prime condition.

74

Fig. 15. Effects of syntactic primes on Korean speakers’ choice of actives and passives.

Fig. 16 plots the mean utterance latencies for active and passive sentences. As in
the first two experiments, the latencies for active and passive sentences differ
significantly. However, unlike Experiments 2 and 3, passive sentences were now
produced significantly faster than active sentences; the mean latency of passive sentences
was 2554 ms, whereas the mean latency of active sentences was 2878 ms (β=-491.0 t=-
2.553, p<.05, SE=192.3). If the rapid production of passives and the slowdown of actives
were caused by the passive primes, this would suggest that passive primes facilitated
speakers’ passive structural plans but interfered with active plans, as we hypothesized.

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
Ac#ve&Prime& Passive&Prime&
Propor#on&of&U2erances&in&Ac#ve&

75

Fig. 16. Utterance latencies for active and passive sentences in the syntactic priming
experiment.

To see whether passive primes indeed decreased the utterance latency of passive
sentences and increased that of actives, Fig. 17 plots utterance latencies for active and
passive sentences broken down by the prime condition (active prime vs. passive prime).
As we hypothesized, the prior presence of a passive structure significantly reduced the
mean utterance latency of passive utterances (β=-414.7, t=-2.281, p<.05, SE=181.8), as
compared to trials which were preceded by an active structure. The mean utterance
latency of passive utterances was 2240 ms in the passive prime condition and 2833 ms in
the active prime condition. Passive primes also marginally increased the mean latency of
active sentences as compared to active primes (β=206.9, t=1.906, p=.05, SE=108.6): The
mean active utterance latency was 3046 ms in the passive prime condition and 2708 ms
in the active prime condition.
0"
500"
1000"
1500"
2000"
2500"
3000"
3500"
Ac#ve& Passive&
U,erance&Latency&(ms)&

76

Fig. 17. Utterance latencies for active and passive sentences by prime condition in the
syntactic priming experiment.

Additionally, if Korean speakers produced active and passive sentences faster
when primed with active and passive sentences respectively, we expect that faster trials
should exhibit syntactic priming effects. More specifically, when we look only at the
trials whose speech onset latencies were faster than the median, we find significant
syntactic priming effects (β=4.013, z=-2.642, p<.01, SE=1.519); when primed with active
sentences, Korean speakers produced active sentences 92% of the time, but when primed
with passive sentences they produced active sentences 86% of the time.

0"
500"
1000"
1500"
2000"
2500"
3000"
3500"
Ac#ve&Prime& Passive&Prime&
U.erance&Latency&(ms)&
Ac)ve"
Passive"

77
5.3 Discussion

Chapter 5 extended the investigation of lexical-structural coordination by
examining effects of syntactic priming on Korean speakers’ syntactic choice and
utterance latencies. We predicted that syntactic priming should have a significant effect
on Korean speakers’ active/passive choice if Korean speakers formulate their structural
plan using more accessible sentence structures. However, if Korean speakers formulate
their plans primarily on the basis of their perception of events, syntactic priming should
not influence their syntactic choice. Regardless of its effect on syntactic choice, we
predicted that syntactic priming should significantly affect utterance onset latencies by
either facilitating or interfering with speakers’ syntactic plan.
The results showed that syntactic priming did not influence Korean speakers’
syntactic choice; priming a passive structure did not increase the production of passive
sentences. But syntactic priming did have significant influence on utterance latencies;
when compared to active primes, passive primes significantly reduced utterance latencies
of subsequent passive utterances and marginally increased the latencies of active
utterances (i.e., active primes marginally reduced active utterance latencies and
significantly increased passive utterance latencies). The fact that we obtained only a
marginal effect of syntactic priming on the latencies of active utterance could be due to
ceiling effects resulting from high frequency and accessibility of active structures.
The patterns of latencies suggest that primed structures either facilitated or
interfered with the structural plan that speakers had formulated. If Korean speakers did

78
not have structural plans, we should not observe any kind of facilitation or interference
effects of syntactic priming; passive utterances should be produced more slowly in the
active and the passive condition as they are more marked than active utterances, and the
latencies of passive utterances should not differ between the two conditions, contrary to
our findings. Rather the clear facilitation or interference patterns of utterance latency
indicate that Korean speakers do formulate structural plans. Additionally, the trials faster
than the median showed significant syntactic priming effects; Korean speakers produced
significantly less active sentences (i.e., more passive sentences) when primed with
passive sentences. This further suggests that syntactic priming indeed has a significant
influence on Korean sentence production.
In sum, the results as a whole suggest that Korean speakers formulate a structural
plan before speech, but the plan is not likely to derive from structural accessibility. The
results rather suggest the possibility that Korean speakers may formulate a structural
frame based on their perception of an event structure. Chapter 6 explores this possibility
by manipulating speakers’ way of viewing an event.

79
Chapter 6
Lexical-Structural Coordination in English and Korean:
Evidence from Perspective Priming

6.1 Introduction
The results of Chapter 5 suggest that Korean speakers formulate a structural plan
before speech, but they do not derive their plan from structural accessibility. To
investigate whether Korean speakers formulate a structural plan based on their coding of
an event structure, Chapter 6 aims to manipulate speakers’ perception of event structures.
Different ways of describing a particular event are assumed to reflect different
perspectives. For example, although sentences ex (9a-c) have the same core meaning,
their syntactic frames and function assignment are assumed to reflect different
perspectives on the event – i.e., which semantic or relational aspects of the event are
more or less important (e.g., Bock, Irwin & Davidson, 2004; Ferreira & Slevc, 2007;
Gleitman, January, Nappa, & Trueswell, 2007).

Fig. 18. A scene depicting a chasing event

80

(9)
(a) A fox is chasing a chicken. (active, agent-initial)
(b) A chicken is being chased by a fox. (passive, patient-initial)
(c) A chicken is running away from a fox. (active, patient-initial)

This can be construed in terms of a gestalt-type Figure-Ground distinction: The
perspectivally-prominent entity, construed as figure, tends to be assigned to the subject
role, whereas the entity understood as background is placed in the predicate (Talmy,
1978; Gleitman, Gleitman, & Ostrin, 1996). Thus, sentence (9a) gives prominence to
what the fox is doing, and is taken to assume the perspective of fox – the agent
perspective. In contrast, sentences (9b-c) describe the event in terms of the animal being
chased, chicken, and thus, are assumed to take the patient perspective. Given that
different ways of describing a particular event reflect different perspectives, this raises
the question of whether making a certain perspective more prominent (i.e., ‘perspective
priming’) would affect speakers’ choices regarding subject (agent-initial/patient-initial)
or syntactic form (active/passive). For example, if the patient perspective is primed,
would speakers be more likely to produce sentences starting with patient characters?
Under a structural account, perspective priming should have significant influence
on sentence structures, because priming a certain perspective can influence how speakers
view the relational structure of an event, thus, affecting speakers’ structural plan. For
example, if the patient perspective is made more prominent by perspective priming, it

81
should cause speakers to focus on patient entities, leading them to eventually mention
them first. On the other hand, if lexical accessibility controls sentence structures,
perspective priming should not influence production. This is because a perspective is not
associated with specific words and thus, does not increase accessibility of a particular
lexical item.
In order to investigae effects of perspective priming on English and Korean
speakers’ choice of sentence forms, we used a picture-description task. In the prime task,
participants read and summarized a short story written from the perspective of the
aggressor (agent) or the victim (patient) (adapted from Bower, 1976; 1977). The prime
task was immediately followed by the picture-description task, in which participants
described pictured events either with active or passive sentences.
If English and Korean speakers formulate structural plans based on their construal
of event structures, perspective priming should affect their choice of subject or syntactic
structure (e.g., priming a patient perspective should increase utterances starting with
patient entities). This is because priming a certain perspective can influence how Korean
speakers view a relational or causal structure of an event, thereby affecting their
production.

82
6.2 Experiment 5

6.2.1. Method

Participants
Fourteen native speakers of English and fourteen native speakers of Korean at the
University of Southern California participated in the experiment for $10 per hour.

Materials
The experiment consisted of a priming task followed by a picture-description task.
In order to select target images for the picture-description task, a norming study was
conducted on a separate group of twenty English speakers and twenty Korean speakers;
participants wrote down single-sentence descriptions of 50 images. To be selected as a
target, an image had to elicit at least one occurrence of agent-initial active and patient-
initial passive sentences in English and Korean, following the criteria of Gleitman et al.
(2007). This resulted in the selection of 26 items. The experiment also included 70 filler
images, which were similar to target images in style but could only be described with
agent-initial active sentences.

83

Fig. 19. Example of a pictured event (biting)

Procedure
The study was run with Experiment Builder software (SR Research). Participants
were told that they would perform two unrelated tasks, (i) reading and summarizing a
story for the purposes of a future study (prime task) and (ii) describing pictured events
(experimental task). In the prime task (adapted from Bower; 1976, 1977), participants
read a short story about a robbery (about one double-spaced page), written from the
perspective of the burglar/aggressor (i.e., agent perspective) or from the perspective of
the victim (i.e., patient perspective)
1
. After reading the agent-perspective or the patient-
perspective story (between-subjects manipulation), participants provided a written
summary of it. The priming task took 5-10 minutes. Immediately after the prime task,
participants performed a picture-description task. In the picture-description task,
participants were instructed to describe pictured events in one sentence using all the

1

There were no passive sentences in the agent-perspective story in English or Korean. In the patient-
perspective story, the English version contained one passive sentence, but the Korean version did not have

84
characters present in the scene. Participants’ speech was recorded with a desk
microphone.

Coding and Analyses
Participants’ speech was analyzed for (i) subject choice (agent-initial versus
patient-initial) and (ii) syntactic structure (actives versus passives). In analyses of subject
choice, ‘agent’ and ‘patient’ were defined in terms of the depicted event rather than
participants’ utterances. For example, for Fig. 19 in which we depicted a biting event, the
dentist was adversely affected by the biting event and thus, was defined as the patient
entity regardless of what utterance speakers produced. We excluded utterances where the
two scene characters were not lexically distinguished (e.g., ‘two children’). Utterances
with conjoined NP subjects were also removed (e.g., ‘a boy and a girl’), because order
variation within NP conjunctions is processed differently from structural variations (e.g.,
Branigan, Pickering, & Tanaka, 2008; Onishi, Murphy, & Bock, 2008). In total, less than
11% of the trials were removed in English and Korean (English: 38/364, Korean: 33/364).
The results were analyzed with mixed-effects logit models. All final models
retained random intercepts for subject and item.

6.2.2 Results

As seen in Fig. 20, perspective priming did not have significant influence on
English speakers’ choice of subject (agent-initial/patient-initial, β=0.10, |z|=0.203,

85
p=.839) or syntactic structure (β=0.468, |z|=0.76, p=.447): English speakers mostly
produced agent-initial (94%) and active sentences (92%) regardless of what perspective
had been primed.

Fig. 20. Proportion of active sentences (left) and agent-initial sentences (right) in English
when primed with an agent or a patient perspective.

In Korean, however, priming the patient perspective significantly increased the
proportion of patient-initial sentences (i.e., decreased the proportion of agent-initial
sentences) (18% vs. 4%, β=1.91, |z|=2.95, p<.01) (Fig. 21): Participants primed with the
patient perspective were more likely to start utterances with patient characters than those
primed with the agent perspective. They also tended to produce more passive sentences
than active sentences, but the effect was marginal (12% vs. 4%, β=4.12, |z|=1.82, p=.07).
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
Agent&Prime& Pa,ent&Prime&
Propor,on&of&Ac,ve&Sentences&in&English&
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
Agent&Prime& Pa,ent&Prime&
Propor,on&of&Agent7ini,al&Sentences&in&English&

86

Fig. 21. Proportion of active sentences (left) and agent-initial sentences (right) in Korean
when primed with an agent or a patient perspective.

In sum, perspective priming influenced speakers’ choice of subject in Korean but
not in English
2
. It did not have any influence on English and Korean speaekrs’ choice
within active versus passive structures.

6.3. Discussion

Chapter 6 tested whether English and Korean speakers’ choice of subject and
syntactic structure was susceptible to perspective priming. We found that Korean
speakers’ choice of subject was significantly affected by the perspective manipulation;
patient-perspective primed Korean speakers were more likely to start their sentences with
the patients of the depicted action than agent-perspective primed speakers. They also

2

The sentence-initial noun did not always correspond to the subject in Korean due to its flexible word
order, but sentences with scrambled word order were rare (about 1%, 5 out of 364 utterances).

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
Agent&Prime& Pa,ent&Prime&
Propor,on&of&Ac,ve&Sentences&in&Korean&
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
Agent&Prime& Pa,ent&Prime&
Propor,on&of&Agent5ini,al&Sentences&in&Korean&

87
tended to produce more passive sentences than active sentences, but the effect did not
reach significance. The fact that perspective priming had a significant effect on the
subject choice but marginal effect on the syntactic choice in Korean suggests that priming
the patient perspective could foreground patient characters, thereby increasing the
availability of sentences in which patients are prominent (i.e., sentences that mention
patient entities first), but not passive structures in particular. This not only reflects the
fact that word order does not constrain syntactic structures in Korean, but also provides
support for the autonomous view of syntactic structure that syntax is represented
independent of other forms of knowledge such as specific features of meaning (e.g., Bock,
1987; Frazier & Fodor, 1978).
On the other hand, English speakers’ choice of subject or syntactic structure was
not influenced by perspective priming. This findings is at odds with Griffin and Bock
(2000)’s proposal that English speakers’ production depends on their structural plan. The
fact that semantic and syntactic priming have a singificnat influence on English speakers’
production whereas perspective priming does not rather suggests that English speakers
produce utterances using more accessible lexical items or sentence structures. If English
speakers can produce utterances using whatever lexical items or sentences structures
happen to be more accessible at a particular point in time, perspective priming is not
likely to have significant influence on English speakers’ subject or structural choice for
two main reasons. First, the manipulation of perspective prominence does not increase
lexical or structural accessibility because perspectives are not associated with specific
lemmas or syntactic structures. Second, if English speakers have little need to rely on

88
their construal of events to produce utterances because they can produce sentences using
more lexical items or sentence structures, they are also likely to be less sensitive to
perspectival information.
Overall, we suggest that the different findings for English and Korean are related
to how they define grammatical functions. In English, where grammatical functions are
highly correlated with word order, lexical accessibility facilitates production by allowing
speakers to build structures using more accessible lexical items as a starting point.
English speakers also tend to use more accessible sentence structures to control lexical
retrieval. The absence of perspective priming, on the other hand, suggests that event-
structural information may not be central for speech formulation in English. However, in
Korean where grammatical functions are indicated by case particles, perspective
manipulation had a significant effect on speakers’ choice of subject, whereas
manipulation of lexical accessibility did not influence their production (Hwang & Kaiser,
2009). This suggests that Korean speakers’ production is more susceptible to event-
structural information.
One may wonder whether the priming effects in Korean could result from
thematic role priming rather than perspective priming. It is possible that the victim and
the aggressor primes have primed the patient and the agent thematic roles respectively,
causing priming effects in the present study. This seems unlikely for two main reasons,
however. First, the priming task did not establish a reliable connection between the
victim and the patient thematic role, or between the aggressor and the agent thematic role.
More specifically, in the texts that the participants read in the prime task, both the

89
aggressor and the victim were realized as the agent (i.e., asumed the agent thematic role)
in about 90% of the sentences. Second, our analyses defined ‘agent’ and ‘patient’ in
terms of the pictured event. For example, in Fig. 18, chicken was defined as the patient
entity that was affected by the chasing event regardless of what utterance participants
produced. However, thematic roles are traditionally defined in terms of verbs. Thus,
chicken assumes the patient role in (8a) and (8b), when the event is described with
‘chase’, but the agent role in (8c) when the event is described with ‘run away’. But in our
analyses, both (8b) and (8c) are analyzed as patient-initial, since they start with the
patient of the pictured event. Because there is no direct mapping between thematic roles
and our analyses, our results cannot be reduced to thematic role priming.
To conclude, our findings show that perspective prominence can guide speakers’
starting point for an utterance, but its effect is modulated by language-specific properties.
More generally, the study suggests that the grammatical properties of a language can
have fundamental effects on what information speakers prioritize during production (e.g.,
lexical information versus event-structure representation). The precise interplay of
different kinds of information, however, is not yet well understood. For example, given
that effects of attention-capture cues increased when participants described hard-to-
interpret events (Kuchinsky & Bock, 2010), would English speakers prioritize event-
structural information when events are easier to describe? How production interacts with
different kinds of information as well as grammatical properties is an important question
for future research.

90
Chapter 7
Lexical-Structural Coordination in English and Korean:
Evidence from Eye Movements

7.1 Introduction
This chapter aims to investigate lexical-structural coordination in English and
Korean by examining eye gaze data collected in the experiments presented in Chapters 3
and 4. Because eye tracking provides a continuous readout of the temporal relationship
between looking and speaking, it is well suited to assess the mechanisms of lexical-
structural coordination in terms of how speakers determine a starting point for an
utterance and establish structural or hierarchical dependencies (i.e., grammatical
relationships between lexical elements) during on-line production. In the following, we
make predictions regarding how lexical and structural information may be integrated in
English and Korean based on their typological properties, and then test these predictions
by examining the temporal relationships between looking and speaking patterns during
scene description in the two languages.

7.1.1 English
Because grammatical functions are highly correlated with word order in English,
speakers’ choice regarding sentence structure influences the order of lexical items and
vice versa; speakers producing a passive structure for their description will almost always
start their utterances with the patient entity and speakers beginning with the patient entity

91
are most likely to produce a passive sentence.
In fact, existing online production research on English supports both structural
and lexical guidance. Evidence for structural incrementality comes from Griffin and Bock
(2000). They had participants describe pictured events and monitored their eye
movements during the task. They reasoned that if sentence production could begin with
retrieval of a lexical item as entailed by lexical incrementality, initial fixations to one
scene character or the other should predict the subject character. On the other hand, if
sentences are formulated upon consideration of the relations between scene characters as
entailed by structural incrementality, initial fixations should not predict the subject
character. The results of Griffin and Bock (2000) provided support for the latter; initial
fixations were not predictive of the subject character. That is, participants did not fixate
the subject character more than the object character during the first 300 ms of picture
display. Looks to the subject and the object characters significantly diverged beyond 300
ms after picture onset. Similar patterns of eye movements were found when participants
performed a non-linguistic patient-detection task; when participants were asked to
identify the patient of the depicted action, they fixated the patient character
approximately after 300 ms of image display. Given that the patient detection task
requires the understanding of relations among scene entities (e.g., who did what to
whom), Griffin and Bock (2000) suggests that the first 300 ms of viewing serves to
extract the event structure and speakers formulate sentence structures upon consideration
of the event structure rather than lexical retrieval.
Using a picture-description task similar to Griffin and Bock (2000), Gleitman et al.

92
(2007) showed that lexical incrementality could also guide lexical-structural coordination
in English, in particular when accessibility of an individual lexical item is increased.
When a character was made more accessible by a subliminal attention-capturing flash, it
was more likely to be chosen as the subject and looked more during the first 200 ms of
the display. For example, when the attention-capturing flash cued the location where the
patient character would appear, speakers were more likely to produce a passive sentence
than when the flash cued the location of the agent character. Gleitman et al. suggest that
attention-capture cues draw initial attention and looks to the cued character, which in turn
affects the choice of sentence forms by increasing the accessibility of the corresponding
lemma and lexeme. Although Gleitman et al. did not test the claims of lexical vs.
structural incrementality, the immediate influence of lexical retrieval on sentence
structures suggests that lexical incrementality can drive lexical-structural coordination in
English production.
In sum, the close relationship between grammatical functions and word order in
English allows either structural or lexical control of production, which receives empirical
support from Griffin and Bock (2000) and Gleitman et al. (2007).

7.1.2 Korean
Unlike in English, Korean indicates grammatical functions by case particles. As
discussed in Chapter 2, we hypothesize that in Korean structural incrementality will
dominate lexical-structural coordination because structural plans are necessary for case
assignment.

93
Although there are no existing online experiments on lexical-structural
coordination in Korean, existing work on real-time comprehension provides evidence for
structural control of lexical items in Korean. Choi and Trueswell (in press) showed that
Korean speakers used structural information – i.e., case particles to estimate upcoming
verbs. For example, ‘napkin-ey’ in (10) is temporally ambiguous between a modifier of
‘frog’ or a goal of the upcoming verb ‘put’, because the case particle –ey can mark both
the genitive and the locative construction.

(10) Naypkhin-ey kaykwuli-lul nohu-sey-yo / cipu-sey-yo.
Napkin-ey frog-ACC put / pick up
‘Put / Pick up the frog on the napkin.’

However, –ey is more commonly used as the locative marker and Choi and Trueswell
found that Korean speakers tended to anticipate verbs like ‘put’, which occurs in the
locative construction. Specifically, when presented with an array with a frog on a napkin,
a giraffe in a basket, a frog in a bowl, and an empty napkin, Korean speakers were more
likely to initially look at the empty napkin, which was a plausible goal/location (see
Kamide et al., 2003 for Japanese speakers’ use of case particles to anticipate forthcoming
arguments). The fact that structural information such as case particles is used to estimate
forthcoming linguistic input in parsing Korean suggests that structural information may
also dominate lexical-structural coordination in Korean production.

94
7.1.3 Predictions
We hypothesize that in English where grammatical functions are highly correlated
with word order, either lexical or structural incrementality could guide lexical-structural
coordination. However, in Korean where grammatical functions are indicated by case
particles, structural incrementality would take precedence over lexical incrementality.
Online production research on English provides support for both structural and lexical
incrementality in English; Griffin and Bock (2000) showed that English speakers
determined a starting point based on their structural plan consistent with structural
incrementality, whereas Gleitman et al. (2007) showed that English speakers’ starting
point was influenced by lexical retrieval consistent with lexical incrementality. The
studies of Griffin and Bock (2000) and Gleitman et al. (2007), however, did not examine
how structural dependencies were established during production. For Korean, there is no
online production evidence that I know of to show how speakers determine a starting
point or structural dependencies. Thus, the experiments presented in this chapter aims to
evaluate lexical and structural incrementality in English and Korean by examining how
English and Korean speakers determine a starting point and structural dependencies. We
manipulated lexical accessibility via attention-capturing visual cues (Experiment 3) or
semantic prime words (Experiment 2), and monitored English and Korean speakers’ eye
movements during scene description. If English speakers opt for lexical incrementality
when accessibility of a lexical item is increased as found in Gleitman et al. (2007) unlike
Korean speakers who do not give up structural incrementality, we expect to find different
patterns of eye movements between English and Korean. This is because lexical and

95
structural incrementality make different predictions on how speakers determine a starting
point or structural dependencies between lexical items.

Predictions about a starting point for an utterance
In terms of a starting point for an utterance (i.e., subjects or in general the first
referents), lexical incrementality posits that more accessible lexical items serve as a
starting point, whereas structural incrementality posits that a starting point is determined
by speakers’ structural plan rather than by the accessibility of lexical items. If the
selection of a starting point is sensitive to lexical accessibility as suggested by lexical
incrementality, speakers should look more at the lexically accessible referent at the outset
of sentence formulation. On the other hand, if a starting point is determined by a
structural plan as suggested by structural incrementality, we do not expect speakers to
look more at the lexically accessible referent in particular.
Looks to the referents, however, can be driven by either reflexive or voluntary
attention. It is well known that attention-capture cues as used in Gleitman et al. (2007)
reflexively draw looks to the characters (e.g., Posner, 1980; Jonides, 1981; Muller &
Rabbitt, 1989). This predicts that English and Korean speakers should look more at the
cued character early after image display regardless of whether they determine a starting
point from lexical accessibility or a structural plan. However, if English speakers choose
a more accessible and thus, cued referent as a starting point, whereas Korean speakers do
not, the first referent should draw more looks than the second referent under the influence
of attention-capture cues only in English as reported in Gleitman et al. (2007).

96
Speakers can also deliberately direct their looks to the characters, reflecting what
is on their mind (e.g., Rayner, 1998; Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy,
1995). Given that semantic prime words direct looks to the primed referent but not
reflexively (Huettig & Altmann, 2005), looks to the semantically primed referent are a
good barometer to assess how speakers determine a starting point under lexical and
structural incrementality. If English speakers choose a lexically more accessible referent
as the starting point for their utterance as reported by Bock (1986), they should look more
at the semantically primed scene characters. On the other hand, if Korean speakers
determine a starting point based on their structural plan, they are not expected to look
more at the semantically primed scene character.

Predictions about structural dependencies
Lexical and structural incrementality also differ in how structural dependencies
(i.e., grammatical relationships) between lexical elements are established. Lexical
incrementality implies that hierarchical dependencies between lexical items are
established following lexical retrieval. That is, speakers begin an utterance without
necessarily knowing where it is headed due to the absence of an overarching plan. On the
other hand, structural incrementality proposes that speakers have established structural
dependencies by the time they start to speak; speakers have a structural scheme of where
their utterance is headed when they start to speak, and use the plan to prepare more of the
upcoming elements. Therefore, lexical and structural incrementality differ in how looking
patterns unfold relative to utterance onset.

97
Both kinds of incrementality predict that speakers fixate a referent prior to naming
it; they fixate more on the first referent (e.g., ‘fox’ in ‘a fox is chasing a chicken’) before
speech onset, but more on the second referent (e.g., ‘chicken’) after speech onset. Thus,
both English and Korean speakers fixate more on the second referent after they speak,
although we expect looks to the second referent to emerge earlier in Korean because the
second referent immediately follows the first referent; in English, the second referent
follows the verb.
But importantly, lexical and structural incrementality predict differently how
looks to the second referent unfold. If speakers direct eye movements and prepare
upcoming referents under the guidance of structural plans, they should be able to fixate
the second referent more quickly. That is, looks to the second referent should diverge
faster from looks to the first referent when eye movements are directed under the control
of structural plans than when they are under no guidance. We expect the contrast between
lexical and structural incrementality to be revealed more clearly in passive utterances
than active utterances; because active structures are more accessible and less complex
than passive structures, we predict that both English and Korean speakers could quickly
fixate the second referent regardless of whether they have structural plans or not. When
speakers produce passive utterances, on the other hand, their structural plans or the lack
thereof could have a significant effect on fixation on the second referent. Given that
sentence structures are not necessarily entirely prepared at the outset of sentence
formulation (Bock et al, 2004; Schriefers, Teruel, & Meinshausen, 1998), the marked
nature of passive structures is expected to delay fixation on the second referent in both

98
English and Korean. But we predict that the delay should be aggravated by the absence of
structural guidance. Thus, if English speakers begin their utterance without a structural
plan, they should fixate on the second referent much more slowly in passive utterances.
On the other hand, if Korean speakers have a structural scheme for passive utterances at
the onset of speech, they should still fixate the second referent quickly, although the use
of passive structures is more restrained in Korean than in English.
In sum, if English speakers choose a lexically accessible referent as the starting
point for their utterance and build structural dependencies on the fly, they should be more
likely to look at the lexically accessible referent and fixate the second referent slowly
after speech onset. On the other hand, if Korean speakers determine a starting point from
their structural plan and direct their gaze according to the plan, they are not expected to
particularly look more at the accessible referent, but they should be able to fixate the
second referent more quickly.

7.2 Experiment 3: Attention-capture cueing

7.2.1 Method

The methods were the same as those used in the attention-capture experiments in Chapter
4.

99

7.2.2 Results

In what follows, I first present analyses investigating whether attention-capture
cues were effective, i.e., whether they successfully triggered eye-movements towards the
cued character. Then, I turn to the analyses that show whether looks to the cued
characters are associated with looks to the first referents, which concerns how speakers
choose the first referents. After that, I present analyses on how looks to the second
referents unfold, which concerns how speakers establish grammatical relations between
lexical items.

Looks to the cued vs. uncued scene character from picture onset
As reported in Gleitman et al. (2007), attention-capture visual cues were effective
at attracting initial eye movements toward the cued scene character; the analyses of one
sample, two-tailed t-tests on mean proportions of trials with initial fixation to the cued
character (Table 2, repeated from Chapter 4) showed that both English and Korean
participants were significantly more likely to first fixate the cued character than would be
expected by chance (mean proportion of trials on which participants fixated the cued
character first = 0.69, 95% CI= 0.06 in English, and 0.73, 95% CI= 0.04 in Korean). The
proportion of looks to the cued vs. uncued scene character in Fig. 22 also demonstrates
the effect of visual cues on early looking patterns; participants were more likely to look at
the cued scene character than the uncued character during the first 200 ms of the image

100
display in both English (mean proportion of time that participants spent viewing the cued
character was 0.17, and the uncued character was 0.02, 95% CI = ± 0.06) and in Korean
(mean proportion of time that participants spent viewing the cued character was 0.16, and
the uncued character was 0.01, 95% CI = ± 0.05).

Table 2
Effectiveness of Attention-capture Manipulation in Experiment 3: One-sample two-tailed
t-tests on the mean proportion of trials where participants first fixated the cued character
(compared to chance level of 0.5)

df
1
t
1
p
1
df
2
t
2
p
2
English 23 5.88 <.001 15 7.88 <.001
Korean 23 10.72 <.001 15 7.06 <.001

101

Fig. 22. Proportion of looks to the cued vs. uncued scene character relative to picture
onset in English and Korean from Experiment 1

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
Proportion
of
Looks
ms
from
Picture
Onset
English
Looks
to
Cued
Looks
to
Uncued
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
Proportion
of
Looks
ms
from
Picture
Onset
Korean
Looks
to
Cued
Looks
to
Uncued

102
Table 3
Paired t-tests on the mean proportion of looks to the cued character within the first 200
ms of the image display

In order to examine whether early fixation is related to which referents English
and Korean speakers mention first, we compared the proportion of looks to the first
versus the second referent during the first 200 ms of image display. If English speakers
tend to choose the more accessible scene character as the first referent, the first referent
should be looked more at during the first 200 ms of image display as reported by
Gleitman et al. (2007). Korean speakers, however, should not fixate the first referent
more if they determine the first referent from their structural plan. Analyses of the mean
proportion of looks to the first referent versus to the second referent are shown in Table 4.
As expected, English speakers were significantly more likely to look at the first referent
during the first 200 ms of the image display (mean proportion of looks to the first referent
was 0.12, and the second referent was 0.07, 95% CI = ± 0.02). In Korean, however, looks
to the first versus the second referent did not significantly diverge during the first 200 ms
of the image display (mean proportion of looks to the first referent was 0.10, and the
second referent was 0.07, 95% CI = ± 0.03).

df
1
t
1
p
1
df
2
t
2
p
2
English 23 5.01 <.001 15 10.90 <.001
Korean 23 7.50 <.001 15 10.33 <.001

103
Table 4
Paired two-tailed t-tests on the mean proportion of looks to the first vs. second referent
within the first 200 ms of the image display

Looks to the second referent
Although attention-capture cues were effective at attracting initial looks to the
cued scene character in both English and Korean, only English speakers looked more at
the character that they eventually mentioned first during the first 200 ms of image display.
This suggests that English speakers choose the first referent based on lexical accessibility,
whereas Korean speakers choose the first referent based on their structural plan. Lexical
and structural incrementality also predict differences in how looks to the second referent
unfold; if English speakers begin their utterance without a structural plan, looks to the
second referent should diverge more slowly than when eye movements are directed under
the control of structural plans as in Korean.
In order to examine how looks to the second referent unfold, Fig. 23 plots looks to
the first and the second referent relative to speech onset in English and Korean. As can be
seen in Fig. 23, both English and Korean speakers fixated the first referent before speech
onset. However, fixation on the second referent occurs more quickly and fluently in
Korean. In English looks to the second referent started to diverge from looks to the first
df
1
t
1
p
1
df
2
t
2
p
2
English 23 4.03 <.001 15 2.02 .06
Korean 23 1.71 .10 15 0.73 .48

104
referent at speech onset but reached significance 100 ms after speech onset (t
1
(23)=-2.83,
p<.01, t
2
(15)=-2.81, p<.05, by a paired two-tailed t-test). In Korean, however, looks to
the second referent started to diverge from looks to the first referent about 200 ms before
speech onset and diverged significantly right away (t
1
(23)=-4.67, p<.001, t
2
(15)=-4.28,
p<.001, by a paired two-tailed t-test).

Fig. 23. Proportion of looks to the first vs. the second referent relative to speech onset in
English and Korean from Experiment 3 (dashed lines represent when looks to the second
referent significantly diverge from looks to the first referent)

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
-‐1000
-‐800
-‐600
-‐400
-‐200
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Proportion
of
Looks
ms
from
Speech
Onset
English
Looks
to
N1
Looks
to
N2

105
Fig. 23, Continued

7.2.3 Discussion

In order to investigate the mechanisms of lexical-structural coordination in
English and Korean, Experiment 1 examined the time course of eye movements relative
to image and utterance onset under the influence of attention-capture manipulation. We
predicted that both English and Korean speakers should look more at the cued scene
character because the subliminal flash such as attention-capture cues is known to
reflexively orient looks to the stimuli. However, if lexical incrementality guides lexical-
structural coordination in English, whereas structural incrementality guides lexical-
structural coordination in Korean, English and Korean speakers should show different
time courses of eye movements in terms of a connection to linguistic choice.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-‐1000
-‐800
-‐600
-‐400
-‐200
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Proportion
of
Looks
ms
from
Speech
Onset
Korean
Looks
to
N1
Looks
to
N2

106
As reported by Gleitman et al. (2007), attention-capture cues were effective at
attracting looks to the cued scene character; both English and Korean speakers were more
likely to direct their gaze to the cued scene character than would be expected by chance
and to look more at the cued character during the first 200 ms after the image display.
English and Korean speakers, however, showed different time courses of eye
movements in terms of the first and the second referent. English speakers tended to look
significantly more at the first referent during the first 200 ms of image display, whereas
Korean speakers did not. Korean speakers, however, fixated the second referent more
quickly; fixation on the second referent started to diverge immediately in Korean,
whereas in English significant divergence occurred 100 ms after speakers fixated the
second referent.
The fact that English speakers looked more at the first referent under the influence
of attention-capture cues but fixated the second referent more slowly suggests that
English speakers tended to build sentence structures on the fly starting with a more
accessible lexical item, consistent with lexical incrementality in English. On the other
hand, the fact that Korean speakers did not look more at the first referent early in the
image display but fixated the second referent more quickly and intensely suggests that
Korean speakers were provided with structural guidance to a starting point and the
upcoming referent.

107
7.3 Experiment 2: Semantic priming

Experiment 2 extends the investigation of lexical-structural coordination in
English and Korean by examining the time course of eye movements under the influence
of semantic primes. Because semantic primes do not reflexively direct eye-movements,
looks to semantically primed entities could be a good barometer to access lexical vs.
structural accessibility. In terms of a starting point, if English speakers tend to choose
more accessible lexical items as a starting point for their utterance, they should look more
at the semantically primed scene characters. On the other hand, if Korean speakers
choose a starting point from their structural plan, they are not expected to look more at
the primed scene character. In terms of the looks to the first and the second referents
relative to utterance onset, if English speakers’ eye movements are not under the
direction of structural guidance unlike Korean speakers’, we expect that English speakers
fixate the second referent more slowly than Korean speakers.

7.3.1 Methods

The methods were the same as those used in the semantic priming experiment in Chapter
3.

108
7.3.2 Results

Looks to the primed vs. unprimed scene character from picture onset
To examine how semantic primes affected initial looks to primed and unprimed
scene characters, we plotted the proportion of looks to the primed versus unprimed
characters in English and Korean relative to picture onset (Fig. 24).

Fig. 24. Proportion of looks to primed vs. unprimed character relative to picture onset in
English and Korean from Experiment 2

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
Proportion
of
Looks
ms
from
Picture
Onset
English
Looks
to
Primed
Looks
to
Unprimed

109
Fig. 24, Continued

As can be seen in Fig. 24, semantic primes, unlike attention-capture cues, did not
immediately attract looks to the primed scene character during first 0-200 ms after picture
onset; analyses of one-sample two-tailed t-tests on mean proportions of trials with initial
fixation to the primed character demonstrated that participants were not more likely to
direct their gaze to the primed scene character than would be expected by chance during
the first 200 ms of the image display (mean proportion of trials on which participants
fixated the cued character first = 0.52 in English, and 0.50 in Korean, p>.1).
Although looks to the primed versus unprimed character did not diverge during
the first 200 ms of the image display, the primed scene character was looked significantly
more at during 400 – 800 ms of the image display in English (Table 5). Looks to the first
referent also started to continuously diverge from looks to the second character at 400 ms
after picture onset. In Korean, on the other hand, looks to the primed versus unprimed
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
Proportion
of
Looks
ms
from
Picture
Onset
Korean
Looks
to
Primed
Looks
to
Unprimed

110
character did not diverge during the first 3000 ms after picture onset (p>.1). Yet, Korean
speakers started to continuously fixate the first referent at 400 ms. In sum, looks to the
primed characters were related to looks to the first referent in English, but not in Korean.

Table 5
Paired t-tests on the mean proportion of looks to the primed vs. unprimed character

Looks to the first vs. second referent from speech onset
Fig. 25 plots looks to the first and the second referent relative to speech onset in
English and Korean. As can be seen in Fig. 25, both English and Korean speakers fixated
the first referent before speech onset. However, as in Experiment 1 fixation on the second
referent unfolds more quickly and fluently in Korean. Although looks to the second
referent began to diverge from looks to the first referent 100 ms before speech onset in
English, it reached significance 100 ms after speech onset in English (t
1
(23)=-5.25,
p<.001, t
2
(15)=-3.66, p<.01, by a paired two-tailed t-test). In Korean, looks to the second
referent started to significantly diverge 200 ms before speech onset and immediately
ms df
1
t
1
p
1
df
2
t
2
p
2
0-200 23 .46 .64 15 .59 .55
200-400 23 .78 .43 15 1.09 .29
400-600 23 3.41 <.01 15 2.95 <.01
600-800 23 4.03 <.001 15 3.51 <.01
800-1000 23 .03 .97 15 .11 .90

111
reached significance (t
1
(23)=-2.59, p<.05, t
2
(15)=-2.24, p<.05, by a paired two-tailed t-
test).

Fig. 25. Proportion of looks to N1 vs. N2 relative to speech onset in English and Korean
from Experiment 2 (dashed lines represent when N2 looks significantly diverge from N1
looks)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-‐1000
-‐800
-‐600
-‐400
-‐200
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Proportion
of
Looks
ms
from
Speech
Onset
English
Looks
to
N1
Looks
to
N2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-‐1000
-‐800
-‐600
-‐400
-‐200
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Proportion
of
Looks
ms
from
Speech
Onset
Korean
Looks
to
N1
Looks
to
N2

112

Although the fixation patterns on the second referent suggest that Korean
speakers’ eye movements are directed by structural plans unlike English speakers’, the
influence of structural guidance or its lack thereof should be more prominent when
speakers produce passive sentences. For active utterances, English and Korean speakers
should be able to fixate the second referent early relative to speech onset regardless of
whether they have structural plans or not, due to highly accessible nature of active
structures. But when they produce marked passive sentences, the lack of structural
guidance could have adverse effects on fixation on the second referent; the marked nature
of passive structures should delay speakers’ fixation on the second referent, but the lack
of structural guidance should exacerbate the delay. Thus, we expect that English speakers’
looks to the second referent should diverge much more slowly in passives than in actives.
Korean speakers, however, should be able to fixate the second referent quickly due to
structural guidance, although the use of passive structures is more restrained in Korean.
To examine this possibility, we analyzed the eye gaze data separately for active and
passive utterances. Because eye movement patterns relative to speech onset are similar in
Experiments 2 and 3, and there are relatively few passive utterances in English and
Korean, we collapsed the eye-gaze data across the two experiments.
Fig. 26 plots the eye gaze data relative to speech onset in active utterances. As can
be seen in Fig. 26, fixation on the second referent unfolded fast and fluently in both
English and Korean.

113

Fig. 26. Proportion of looks to N1 vs. N2 relative to speech onset in active utterances in
English and Korean from Experiment 2 (dashed lines represent when N2 looks
significantly diverge from N1 looks)

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
,1000"
,800"
,600"
,400"
,200"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
Propor%on'of'Looks'
ms'from'Speech'Onset'
English:'Ac%ve'
Looks"to"N1"
Looks"to"N2"
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
-1000"
-800"
-600"
-400"
-200"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
Propor%on'of'Looks'
ms'from'Speech'Onset'
Korean:'Ac%ve'
Looks"to"N1"
Looks"to"N2"

114
In passive utterances, however, analyses of eye movements showed that fixation on the
second referent was much delayed in English than in Korean. As can be seen in Fig. 27,
English speakers started to fixate the second referent about 400 ms after speech onset but
it did not reach significance only 800 ms after speech onset (t
1
(13)=-2.83, p<.05, t
2
(12)=-
2.83, p<.05). On the other hand, Korean speakers fixated the second referent at speech
onset and looks to the second referent immediately diverged from looks to the first
referent (t
1
(40)=-2.58, p<.05, t
2
(13)=-4.28, p<.001)

Fig. 27. Proportion of looks to N1 vs. N2 relative to speech onset in passive utterances in
English and Korean from Experiments 2 and 3 (dashed lines represent when N2 looks
significantly diverge from N1 looks)

0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
-1000"
-800"
-600"
-400"
-200"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
Propor%on'of'Looks'
'
ms'from'Speech'Onset''''''''
English:'Passive'
Looks"to"N1"
Looks"to"N2"

115

Fig. 27, continued

7.3.3 Discussion

Unlike attention-capture visual cues, semantic primes did not attract looks to the
primed scene character during the first 200 ms of the image display. However, during the
400 – 800 ms after picture onset, the primed character drew significantly more looks than
the unprimed character in English. In Korean, however, looks to the primed versus
unprimed character did not diverge during the 3000 ms of the image display. Given that
looks to the first referent started to continuously diverge from looks to the second referent
at 400 ms in English and Korean, looks to the primed character were correlated with
looks to the first referent in English, but not in Korean. This suggests that English
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
0.9"
1"
-1000"
-800"
-600"
-400"
-200"
0"
200"
400"
600"
800"
1000"
1200"
1400"
1600"
1800"
2000"
Propor%on'of'Looks'
ms'from'Speech'Onset'
Korean:'Passive'
Looks"to"N1"
Looks"to"N2"

116
speakers tended to choose lexically accessible scene character as the first referent,
whereas Korean speakers tended to choose the first referent based on their structural
plans rather than lexical accessibility.
The time course of eye movements in terms of the second referent further
supports for lexical incrementality in English and structural incrementality in Korean.
Looks to the second referent unfolded more quickly and fluently in Korean than in
English; once looks to the second referent started to diverge from looks to the first
referent, they significantly diverged immediately in Korean, but it took about 100 ms in
English. The contrast between English and Korean is more clearly revealed in eye
movement analyses of passive utterances. Although the use of passive structures is more
restrained in Korean as discussed in Chapter 2, fixations on the second referents unfolded
faster and more fluently in Korean; English speakers started to fixate the second referent
about 400 ms after speech onset and looks to the second referent significantly started to
diverge only about 800 ms after speech onset. But Korean speakers fixated the second
referents on speech onset and looks to the second referents immediately diverged
significantly. The faster and more fluent fixation on the second referent in Korean
suggests that Korean speakers’ eye movements were directed by structural plans, whereas
English speakers’ were not under structural control.

117
7.4 General discussion

In Chapter 7, we set out to investigate how English and Korean speakers
coordinate lexical and structural information by examining the time course of eye
movements in relation to subsequent linguistic choice – i.e., how English and Korean
speakers determine a starting point and structural dependencies when lexical accessibility
was manipulated by attention-capture visual cues (Experiment 1) and semantic prime
words (Experiment 2).
In terms of a starting point, English speakers tended to look more at the lexically
accessible scene characters in Experiments 1 and 2. Analyses of looks to the cued versus
the uncued characters relative to picture onset in Experiment 1 showed that attention-
capture cues significantly attracted looks to the cued characters during the first 200 ms of
the image display in both English and Korean. However, only English speakers tended to
look significantly more at the first referent during the first 200 ms of the image display.
In Experiment 2, English speakers looked more at the semantically primed scene
characters during 400 – 800 ms of the image display and during this time window they
also looked significantly more at the first referent. The close relation between looks to the
accessible scene characters and the first referents in English suggests that English
speakers tend to choose more accessible lexical items as a starting point for their
utterance. On the other hand, the absence of such relation in Korean suggests that Korean
speakers are likely to determine a starting point based on their structural plan.
The fast fixations on the second referent in Korean further provide support for

118
structural guidance in Korean; once Korean speakers fixated the second referent, looks to
the second referent immediately started to diverge from looks to the first referent. Even
when producing marked passive utterances, Korean speakers fixated the second referent
quickly and fluently. But in English, significant divergence between looks to the second
versus the first referent occurred about 100 ms after English speakers fixated the second
referent. The delay is much longer in passive sentences. The fast and fluent fixation on
the second referent in Korean suggests that Korean speakers’ eye movements are under
structural control as suggested by structural incrementality. The slow and delayed
fixation on the second referent in English, on the other hand, suggests more incremental
structure-building process of English speakers as suggested by lexical incrementality.
There is additional evidence that reveals English speakers’ sensitivity to lexical
accessibility. English speakers fixated the second referent earlier in the semantic priming
condition than in the attention-capture cueing condition. This could be because
semantically primed characters are easier to encode linguistically; semantic prime words
which share semantic contents (and possibly morpho-phonology) with the primed scene
character better activate the lemmas (and lexemes) of the primed characters than
attention-capture cues which lack such semantic or morpho-phonological support.
Semantic priming, however, did not have advantage over attention-capture priming in
Korean; in both the attention-capture and semantic priming conditions, Korean speakers
fixated the second referent about 200 ms before speech onset. Korean speakers’
insusceptibility to semantic support suggests that Korean speakers retrieval of lexical
items is under control of their structural plan.

119
Overall, the findings reported in the present chapter indicate that lexical
incrementality tends to predominate over structural incrementality in English, whereas
structural incrementality predominates over lexical incrementality in Korean.

120
Chapter 8
General Discussion

8.1 Summary of the findings
This dissertation set out to investigate how two typologically different languages,
namely English and Korean, coordinate lexical and structural information during on-line
production.
Two accounts of lexical-structural coordination make different assumptions about
the relative contributions that words and syntax make in sentence formulation. A lexical
account suggests that lexical items control the formulation of sentence structures (e.g.,
replace X for Y, substitute Y for X), whereas a structural account suggests that structural
frameworks control lexical items; speakers generate a rudimentary syntactic plan from
their construal of an event, and use this structural information to control the timing of
subsequent lexical retrieval (Konopka, 2009; Konopka & Bock, 2009). Put simply,
speakers ‘latch onto’ a word and build a sentence structure from it, or first choose a
structural frame and fill the structure with the appropriate words.
I hypothesized that the mechanisms of lexical-structural coordination in English
and Korean are related to their function assignment properties: In English, where
grammatical functions are highly correlated with word order (e.g. the first noun is almost
always the subject), I hypothesized that either words or sentence frames could guide
lexical-structural coordination; passives are likely to begin with patient entities, and vice
versa. However, in Korean, where word order is flexible and grammatical functions are

121
marked by case particles, I hypothesized that structural plans should control lexical
retrieval. Because the order of lexical items per se does not determine a syntactic
structure, and structural frames are necessary for case assignment, I predicted that Korean
speakers formulate structural plans and control lexical retrieval according to the plans.
To test these hypotheses, I have investigated three primary questions: Is selection
of verbs necessary for sentence planning? (Chapter 2); Does manipulation of lexical,
structural, and perspectival accessibility affect English and Korean speakers’ sentence
forms? (Chapters 3, 4, 5 and 6); What do the eyes say about lexical-structural
coordination? (Chapter 7). I investigated these questions using core experimental
techniques such as real-time eye tracking and reaction time measuring, and statistical
methods such as mixed-effects models.
I found that English speakers selected verbs before they started to speak, whereas
Korean speakers did not (Chapter 2). This together with previous findings in English
suggest that English speakers produce utterances using more accessible lexical items or
sentence structures; they tended to choose more accessible lexical items as the subject
(i.e., semantic priming in Chapter 3) and produce more accessible sentence structures (i.e.,
structural priming). Importantly, because word order constrains grammatical functions in
English, English speakers could build sentence structures on the fly starting from more
accessible lexical items without pre-determined structural frameworks. On the other hand,
Korean speakers’ choice of subject was not influenced by lexical accessibility (i.e., no
lexical priming in Chapters 3 and 4), but by their perception of event structures (i.e.,
perspective priming in Chapter 6). The fact that structural priming had significant

122
influence on utterance onset latencies (e.g., priming a passive structure significantly
reduced passive latencies, while increasing active latencies in Chapter 5) suggests that
Korean speakers formulate structural frameworks to control retrieval of lexical items, but
their plans are based on their perception of events rather than based on lexical or
structural accessibility. Analyses of eye movements were also consistent with the
hypotheses (Chapter 7); English speakers looked more at lexically accessible referents
early in the image display, whereas Korean speakers did not. This suggests that the
choice of the first referent (i.e., the subject) was influenced by lexical accessibility in
English but not in Korean, compatible with the structural account. The fact that looks to
the second referent unfolded more quickly and fluently in Korean provides evidence that
structural frameworks directed Korean speakers’ eye movements unlike English speakers’
gaze.

8.2 Implications and questions for future research
As a whole, the cross-linguistic research reported here sheds light on the
coordination problem of how speakers combine lexical items and sentence structures. The
results suggest that the coordination of lexical and structural information is mediated by
language-specific grammatical properties; in English where grammatical functions are
correlated with word order, either lexical items or structural frames can guide lexical-
structural coordination, but in Korean where grammatical functions are marked by case
particles, structural frames guide the coordination process.
These findings help to unify contradictory results reported in English as well as

123
between languages. In English, Griffin and Bock (2000) suggested that English speakers
formulate a rudimentary structural scheme to control lexical retrieval, whereas Gleitman
et al. (2007) suggested that speakers could proceed with whatever lexical item is more
accessible without a structural plan. Considering that grammatical functions are highly
correlated with word order in English, however, both lexical and structural control are
possible in English. The claim receives experimental support from priming studies.
Semantic priming (Bock, 1986a) shows that speakers’ syntactic choice can be riven by
lexical accessibility, whereas syntactic priming (Bock, 1986b) shows that speakers’
syntactic choice can be used to control lexical retrieval.
Although syntactic priming is well observed cross-linguistically (see Tanaka,
2011 for Japanese; Park, 2006 for Korean), lexical accessibility effects are not as well
attested in some languages. As reported in Chapters 3 and 4, semantic priming and
attention-capture cueing did not influence Korean speakers’ word order or syntactic
choice. Myachykov and colleagues (2005; 2007) also did not find any effects of
attention-capture manipulation in Russian and Finnish. The absence of lexical
accessibility on these languages is quite surprising given that they allow more syntactic
structures or word order variation that can accommodate more accessible lexical items
sooner. Importantly, however, these languages indicate grammatical functions by case
particles. Thus, speakers of these languages might as well need a structural plan to assign
case or grammatical functions. If Russian and Finnish speakers formulate a structural
frame to control lexical retrieval just like Korean speakers, then the absence of lexical
accessibility on these languages is rather expected.

124
The current research not only accounts for contradictory findings of previous
research but also helps us make predictions on relative contributions of lexical and
structural information on production. For example, in English where grammatical
functions are highly correlated with word order, either lexical item or sentence structure
can guide lexical-structural coordination, but structural influence should be greater in
active-passive alternation. This is because the use of a passive structure almost always
requires beginning with a patient entity (e.g., a chicken is being chased by a fox), but
beginning with a patient entity does not necessarily lead to the use of a patient entity (e.g.,
a chicken is running away from a fox). In fact, Bock (1986a) found 4% of semantic
priming on active/passive alternation, whereas Bock (1986b) found 8% of syntactic
priming on active/passive alternation.
However, the precise interplay between production and lexical and structural
information as well as grammatical properties is not yet understood, although it has far-
reaching implications for theories of language processing. One question for future
research is what the relationship between lexical and structural accessibility is. For
example, what makes speakers prioritize one process over the other? What is the
relationship between the two ways of formulating a structural plan – i.e., if speakers
could formulate a structural plan based on either their perception of event structures or
structural accessibility, are the two ways mutually exclusive or could both have effects?
And when does one process exert more influence than the other? These are questions
which have important implications for theories of production.
Another question for future research is how lexical-structural coordination is

125
affected by other grammatical or linguistic properties, in addition to word order and
grammatical function assignment. For example, languages also differ in whether they are
subject prominent or topic prominent languages. English is a subject prominent language,
which marks the subject in a clear and reliably way (e.g., the subject is sentence-initial),
whereas Korean is a topic prominent language, which marks the topic in a way (e.g., the
topic is morphologically marked). Although Korean does not require the subject to be in
the sentence-initial position, the topic almost always occurs in the sentence-initial
position. The fact that there are topic-prominent and subject-prominent languages raises
the interesting question of how a topic-comment structure would interact with lexical and
structural accessibility. For instance, would Korean speakers be more likely to produce a
passive sentence after a passive prime in a topic-comment structure as in (11) because the
topic must occur in the sentence-initial position? Or would passive production not be
affected by syntactic priming because the topic is rather a discourse property rather than a
syntactic one (e.g., Li & Thompson, 1976)?

(11) Talk-un yewu-hantey ccokiessta.
chicken-TOPIC fox-BY was chased
‘The chicken was chased by a fox’

In addition to cross-linguistic differences, how would lexical-structural
coordination interact with structures or verb types in a language? The experiments
reported here showed that Korean speakers’ active/passive choice was not influenced by

126
lexical accessibility. But would Korean speakers be susceptible to lexical accessibility
when asked to produce different constructions such as perspective predicates (e.g.,
buy/sell) or dative constructions? Also, the current experiments mostly employed pictures
that elicited morphological passives in Korean because they occur more commonly and
sound natural in colloquial speech than syntactic passives (Lee & Lee, 2008). Yet,
syntactic passives are more like English passives in that they allow most transitive verbs
to be passivized. Thus, it would be interesting to test whether syntactic passives would be
more accommodating to lexical accessibility than morphological passives. Furthermore,
although Korean prefers animate subjects, experience verbs such as frighten easily allows
inanimate subjects as in ‘the noise frightened John.’ This raises the question of whether
Korean speakers would exhibit lexical accessibility effects for experience verb
constructions compared to constructions with action verbs (e.g., hit). Another intriguing
question for future work is how factors such as giveness vs. newness would interact with
lexical-structural coordination. An entity given in the immediately preceding sentence is
likely to be more accessible at a lexical level, whereas an entity given in the discourse
such as the topic is likely to be more prominent at the event-structural level. Given this,
how would lexical vs. event-level accessibility interact with production – e.g., which
entity would speakers be more likely to assign to the subject position between the
immediately mentioned entity or the discourse salient entity?
Another important question for future research is how lexical-structural
coordination interacts with factors other than grammatical properties such as complexity
or easy of encoding event structures. Kuchinsky and Bock (2010)’s finding that the

127
effects of attention-capture cues increased when the events were harder to denote with
verbs suggests verb codability may influence lexical-structural coordination. This further
raises the question of what features of message other than verb codability would affect
production. For example, would speakers be dependent less on lexical accessibility and
more on their structural plan when events are more plausible and thus, easy to encode? Or
would speakers describing implausible events be less susceptible to lexical accessibility
because a certain structure may better communicate their message? When speakers are
asked to describe implausible pictured events, how would their production be affected by
linguistic processes such as syntactic priming?
In addition to event or message properties, Kuno and Kaburiaki (1977) suggests
that the choice of syntactic structure interacts with empathy (e.g., passivization is used
when the speaker wants to describe an event in terms of the patient referent) and that such
interaction may be more clearly observable in Japanese because Japanese has more
morphological and syntactic devices for marking speakers’ empathy – an attitude toward
an event. If Korean speakers’ greater production of passive or patient-initial sentences
can be attributed to their greater empathy with patient entities (note that Korean
morphological passives are often used to communicate that the patient entity is adversely
affected by an event), then would Korean speakers be more willing to take lexical or
structural accessibility into considertation when the events are neutral (e.g., a student met
his friend)?
Finally, given that comprehension also requires coordination of lexical and
structural information, another important question is how the findings in comprehension

128
extend to production and vice versa. The fact that more frequent production errors are
also more likely to be overlooked suggests that the mechanisms of production and
comprehension may be closely related. This raises the question of how factors found to
constrain comprehension would affect production. For example, existing research
suggests that pragmatic information rapidly constrains speakers’ parsing decision (e.g.,
Trueswell, Tanenhaus, & Garnsey, 1994). Sentence (12a) tends to cause processing
difficulty because speakers often misanalyse the defendant as the subject of the verb
examine (i.e., main clause analysis). However, speakers do not experience any processing
difficulty in (12b) because inanimacy of the evidence makes the main clause analysis
implausible.

(12) a. The defendant examined by the lawyer turned out to be unreliable.
b. The evidence examined by the lawyer turned out to be unreliable.

Then, would speakers also experience more processing difficulty when producing
sentence (12a) than (12b), consequently preferring to produce (12b)? The fact that
English speakers tend to put the animate entity to the subject position rather makes the
contrary prediction to the question (e.g., Bock & Warren, 1985). Identifying how
production and comprehension are different or similar will contribute to the development
of theories that unify the two processes.

129
References

Arnold, J. E., Wasow, T., Losongco, T., & Ginstrom, R. (2000). Heaviness vs. newness:
the effects of structural complexity and discourse status on constituent ordering.
Language 76(1). 28–55.
Bates, E., & MacWhinney, B. (1982). Functionalist approaches to grammar. In E.
Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art
(pp.173-218). Cambridge, England: Cambridge University Press.
Bates, D. M. & Sarkar, D. (2007). Lme4: Linear mixed-effects models using S4 classes.
R-package version 0.99875-6.
Baayen,, R. H. (2008). Analyzing Linguistic Data. A Practical Introduction to Statistics
Using R. Cambridge University Press.
Baayen, R. H., Piepenbrock, R., & Gulikers, L. (1995). The CELEX Lexical Database.
http://www.celex.mpi.nl.
Birner, B. J. & Ward, B. (1998). Information Status and Non-canonical Word Order in
English. Amsterdam/Philadelphia: John Benjamins.
Bock, J. K. (1986a). Meaning, sound, and syntax: lexical priming in sentence production.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 575-
586.
Bock, J. K. (1986b). Syntactic persistence in language production. Cognitive Psychology,
18, 355–387.

130
Bock, J. K. (1987). Coordination words and syntax in speech plans. In A. W. Ellis (Ed.),
Progress in the psychology of language, 3, 337-390. London: Erlbaum.
Bock, J. K., Irwin, D. E. & Davidson, D. J. (2004). Putting first things first. In J. M.
Henderson & F. Ferreira (Eds.), The integration of language, vision, and action:
Eye movements and the visual world. New York: Psychology Press.
Bock, J. K., Irwin, D. E., Davidson, D. J., & Levelt, W. J. M. (2003). Minding the clock.
Journal of Memory and Language, 48, 653-685.
Bock, J. K., Loebell, H., & Morey, R. (1992). From conceptual roles to structural
relations: Bridging the syntactic cleft. Psychological Review, 99, 150–171.
Bock, J. K., & Warren, R. K. (1985). Conceptual accessibility and syntactic structure in
sentence formulation. Cognition, 21, 47–67.
Boersma, P. & Weenink, D. (1992-2011). Praat: doing phonetics by compiter. Available
at: http://www.praat.org/.
Branigan, H. P., Pickering, M. J., & Tanaka, M. (2008). Contributions of animacy to
grammatical function assignment and word order during production. Lingua, 118,
172–189.
Brown-Schmidt, S., & Konopka, A. E. (2008). Little houses and casas pequeñas: message
formulation and syntactic form in unscripted speech with speakers of English and
Spanish. Cognition, 109. 274–280.
Choi, H.-W. (2007). Length and order: a corpus study of Korean dative-accusative
construction. Discourse and Cognition, 14, 207–227.

131
Choi, Y. & Trueswell, J. C. (in press). Children’s (in)ability to recover from garden-paths
in a verb-final language: Evidence for developing control in sentence processing.
Journal of Experimental Child Psychology.
de Smedt, K. J. M. J. (1994). Parallelism in incremental sentence generation. Parallelism
in Natural Language Processing, ed. by G. Adriaes and U. Hahn, pp. 421–47.
Norwood, NJ:Ablex.
Ferreira, V. S., & Dell, G. S. (2000). Effect of ambiguity and lexical availability on
syntactic and lexical production. Cognitive Psychology 40(4).296–340.
Ferreira, V. S. & Slevc, L. R. (2007). Grammatical encoding. In M. Gareth Gaskell (Ed.),
The Oxford Handbook of Psycholinguistics. Oxford University Press.
Ferreira, F., & Swets, B. (2002). How incremental is language production? Evidence
from the production of utterances requiring the computation of arithmetic sums.
Journal of Memory and Language, 46, 57–84.
Flores d'Arcais, G, B. (1975). Some attention-capturing determinants of sentence
construction. In G. B. Flores d'Arcais (Ed.), Studies in perception: Festschrift for
Fabio Metelli. Milan, Italy: Martello-Giunti.
Glaser, W. R. & Düngelhoff, F. J. (1984). The time course of picture-word interference.
Journal of Experimental Psychology: Human Perception and Performance, 8,
875–894.
Gleitman, L., January, D., Nappa, R. & Trueswell, J. (2007). On the give and take
between event apprehension and utterance formulation. Journal of Memory and
Language, 57, 544-569.

132
Griffin, Z. M. (2001). Gaze durations during speech reflect word selection and
phonological encoding. Cognition, 82, B1–B14.
Griffin, Z. M., & Bock, J. K. (2000). What the eyes say about speaking. Psychological
Science, 11, 274-279.
Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press.Haeseryn, W.
(Ed.). (1997). Algemene Nederlandse Spraakkunst. Groningen: Nijhoff.
Hawkins, J. A. (1994). A Performance Theory of Order and Constituency. Cambridge,
UK: Cambridge University Press.
Hwang, H. & Kaiser, E. (2011). Is Access to a Verb Lemma Essential in Speech Planning
Cross-linguistically? Poster presented at the 24th Annual CUNY Conference on
Human Sentence Processing. Standford University. Standford, CA.
Hwang, H. & Kaiser, E. (2012). Coordinating Lexical and Structural Information during
Language Production: Evidence from Semantic and Structural Priming. Poster
presented at the 25th Annual CUNY Conference on Human Sentence Processing.
CUNY. New York, NY.
Hwang, H. & Kaiser, E. (under review). The Effects of Perspective Priming on Sentence
Production in English and Korean.
Jackson, K. (2008). The Effect of Information Structure on Korean Scrambling. Ph.D.
Dissertation. University of Hawaii at Manoa.
Jaeger, T. F. & Norcliff, E. J. (2009). The cross-linguistic study of sentence production.
Language and Linguistics Compass, 3(4), 866-887.

133
Kaiser, E. & Vihman, V.A. (2006). Invisible Arguments: Effects of Demotion in Estonian
and Finnish. In T. Solstad & B. Lyngfelt (eds), Demoting the agent: passive and
other voice-related phenomena. John Benjamins, 111-141.
Kang, B. & Kim, H. (2009). Korean word frequencies. Seoul, Korea: Hankukmunhwa
Press.
Kempen, G., & Huijbers, P. (1983). The lexicalization process in sentence production and
naming: Indirect election of words. Cognition, 14, 185–209.
Konopka, A. E. (2009). Planning ahead: Effects of experience with messages, structures,
and words on the scope of linguistic planning. Unpublished doctoral dissertation.
University of Illinois at Urbana-Champaign.
Konopka, A. E. & Bock, J. K. (2009). Lexical or syntactic control of sentence
formulation? Structural generalizations from idiom production. Cognitive
Psychology, 58, 68-101.
Korvorst, M., Roelofs, A., & Levelt, W. J. M. (2006). Incrementality in naming and
reading complex numerals: Evidence from eyetracking. Quarterly Journal of
Experimental Psychology, 59, 296–311.
Kwon, N., Polinsky, M. & Kluender, R. (2006) Subject preference in Korean. In D.
Baumer, D. Montero & M. Scanlon (eds), Proceedings of the 25th West Coast
Conference on Formal Linguistics (WCCFL 25), pp 1-14. Somerville, MA:
Cascadilla Press.
Kuchinsky, S. E. (2009). From seeing to saying: Perceiving, planning, producing,
Unpublished doctoral dissertation. University of Illinois at Urbana-Champaign.

134
Kuchinsky, S. E., Bock, J. K., & Irwin, D. E. (in press). Reversing the hands of time:
Changing the mapping from seeing to saying. Journal of Experimental
Psychology: Learning, Memory, and Cognition.
Kuno, S. & Kaburaki, E. (1977). Empathy and syntax. Linguistic Inquiry, 8- 4, 627-672.
LaHeij, W., Drikx, J., & Kramer, P. (1990). Categorical interference and associative
priming in picture naming. British Journal of Psychology, 81, 511–525.
Lee, K-O, & Lee, Y. (2008). An Event-structural account of passive acquisition.
Language and Speech, 51, 133-149.
Charles N. Li & Thompson, S. A. (1976). Subject and Topic: A New Typology of
Language. In: Charles N. Li (ed.). Subject and Topic. London /New York:
Academic Press.
Lindsley, J. R. (1975). Producing simple utterances: how far ahead do we plan? Cognitive
Psychology, 7, 1–19.
Lindsley, J. R. (1976). Producing simple utterances: details of the planning process.
Journal of Psycholinguistic Research, 5(4), 331–354.
Lupker, S. J. (1979). The semantic nature of response competition in the picture-word
interference task. Memory & Cognition, 7, 485–495.
Lupker, S. J. (1982). The role of phonetic and orthographic similarity in picture-word
interference. Canadian Journal of Psychology, 36, 349–367.
Meyer, A. S. (1997). Conceptual influences on grammatical planning units. Language
and Cognitive Processes, 12, 859–864.

135
Meyer, A. S., Roelofs, A., & Levelt, W. J. M. (2003). Word length effects in object
naming: The role of a response criterion. Journal of Memory and Language, 48,
131–147.
Myachykov, A. & Tomlin, R. (2008). Attention-capturing priming and structural choice
in Russian sentence production. Journal of Cognitive Science, 6(1), 31-48.
Myachykov, A., Thompson, D., Scheepers, C., & Garrod, S. (2011). Visual attention and
structural choice in sentence production across languages. Language and
Linguistics Compass, 5(2), 95-107.
Myachykov, A., Garrod, S. & Scheepers, C. (2010). Attention-capturing priming of
structural choice during English and Finnish sentence production. In R. K. Mishra
& N. Srinivasan (Eds.), Language & Cognition: state of the art. Munich: Lincom
Europa.
Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South
Florida word association, rhyme, and word fragment norms.
http://www.usf.edu/FreeAssociation/.
Onish, K. H., Murphy, G. L., & Bock, J. K. (2008). Prototypicality in sentence
production. Cognitive Psychology, 56, 103-141.
Oshima, D. Y. (2006). Adversity and Korean/Japanese passives: Constructional analogy,
Journal of East Asian Linguistics, 15(2),137-166.
Park, B. J. (2006). Cross-linguistic study of syntactic persistence in Korean-English
bilinguals. Talk presented at the High Desert Linguistics Conference. University
of New Mexico, NM.

136
Park, S. D. (2005). Parameters of Passive Constructions in English and Korean. Ph.D
dissertation. Cornell University.
Park, T. (2004). Investigation of association frequency and imagery value of Korean
words. The Korean Journal of Experimental Psychology, 16(2), 237-260.
Pickering, M. & Ferreira, V. S. (2008). Structural priming: A critical review.
Psychological Bulletin, 134(3), 427-459.
Rayner, K. & Posnansky, C. (1978). Stages of processing in word identification. Journal
of Experimental Psychology: General, 107, 64-80.
Rayner, K., & Springer, C. J. (1986). Graphemic and semantic effects in the picture-word
interference task. British Journal of Psychology,77, 207–222.
Roelofs, A. (1992). A spreading-activation theory of lemma retrieval in speaking.
Cognition, 42, 107-142.
Roelofs, A. (1993). Testing a non-decompositional theory of lemma retrieval in speaking:
Retrieval of verbs. Cognition, 47, 59–87.
Rosenblatt, F. (1961). Principles of Neurodynamics: Perception and the Theory of Brain
Mechanisms. Spartan Books, Washington, DC.
Schriefers, H. (1993). Syntactic processes in the production of noun phrases. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 19, 841–850.
Schriefers, H., Teruel, E., & Meinshausen, R. M. (1998). Producing simple sentences:
Results from picture–word interference experiments. Journal of Memory and
Language, 39, 609–632.

137
Smith, M.C., & Magee, L.E. (1980). Tracing the time course of picture-word processing.
Journal of Experimental Psychology: General, 109, 373-392.
Tomasello, M. (2000). Do young children have adult syntactic competence? Cognition,
74, 209-253.
Uszkoreit, H. (1987). Word Order and Constituent Structure in German. Stanford, CA:
Center for the Study of Language and Information.
Underwood, G. (1976). Semantic interference from unattended printed words. British
Journal of Psychology, 67, 327–338.
Underwood, G., & Briggs, P. (1984). The development of word recognition processes.
British Journal of Psychology, 75, 243–255.
Ward, G. (1988). The Semantics and Pragmatics of Preposing. New York: Garland.
Wasow, T. (1997). End-weight from the speaker’s perspective. Journal of
Psycholinguistics Research 26.347–62.
Wasow, T.(2002). Postverbal Behavior. Stanford, CA: Center for the Study of Language
and Information Publications.
Wilcox, R. (2012). Modern Statistics for the Social and Behavioral Sciences: A Practical
Introduction. New York: CRC Press.
Wu, I. (1997). Research on Korean passives. Seoul: Hankwuk Mwunhwasa.

138
Yamashita, H., & Chang, F. (2001). Long before short preference in the production of a
head-final language. Cognition, 81(2).B45–55.
Yamashita, H. (2006). Sentence production in Japanese. Handbook of East Asian
Psycholinguistics, ed. by M. Nakayama, R. Mazuka and Y. Shirai, Volume 2,
Japanese. Cambridge, UK: Cambridge University Press.

139
Appendix: Baseline Rates of Actives vs. Passives in the Norming Study

Verb (agent-patient) Active Passive
1 step on (girl-soldier) 22 2
2 hit (bear-doctor) 23 2
3 hit (tiger-elephant) 20 3
4 take away (chicken-boy) 21 4
5 butt (bull-fireman) 23 1
6 kick (horse-doctor) 20 5
7 pinch (monkey-grandpa) 24 1
8 pinch (lobster-cat) 22 3
9 chase (dog-hen) 20 4
10 eat (snake-mouse) 21 4
11 prick (hedgehog-rabbit) 7 18
12 squirt (penguin-mailman) 23 1
13 sting (bee-pig) 20 5
14 bite (dog-police) 21 4
15 kick (kangaroo-fox) 22 3
16 stab (camel-lion) 22 1

Abstract (if available)

Abstract Sentence production requires selection of lexical items and structural frames, and coordination of the two in accordance with the grammar. There are two accounts of how lexical-structural coordination occurs: a lexical account and a structural account, which differ in the relative contributions that words and syntax make to early sentence formulation. A lexical account suggests that lexical items control the formulation of sentence structures. On the other hand, a structural account suggests that structural frameworks control lexical items

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Exploring the effects of Korean subject marking and action verbs’ repetition frequency: how they influence the discourse and the memory representations of entities and events

PDF

Flexibility in language production

PDF

How to think before you speak: getting from abstract thoughts to sentences

PDF

Prosody and informativity: a cross-linguistic investigation

PDF

Dynamics of multiple pronoun resolution

PDF

Syntactic and non-syntactic factors in reflexive pronoun resolution in Mandarin Chinese

PDF

Discourse level processing and pronoun interpretation

PDF

Interaction between prosody and information structure: experimental evidence from Hindi and Bangla

PDF

Processing the dynamicity of events in language

PDF

Building phrase structure from items and contexts

PDF

Comparative iIlusions at the syntax-semantics interface

PDF

Syntax-prosody interactions in the clausal domain: head movement and coalescence

PDF

The grammar of correction

PDF

Building a knowledgebase for deep lexical semantics

PDF

Cognitive-linguistic factors and brain morphology predict individual differences in form-sound association learning: two samples from English-speaking and Chinese-speaking university students

PDF

When things are left unsaid: existential and anaphoric implicit objects in discourse

PDF

Towards a correlational law of language: three factors constraining judgement variation

PDF

Number marking and definiteness in Bangla

PDF

A unified syntactic account of Mandarin subject nominals

PDF

Considering the effects of disfluent speech on children’s sentence processing capabilities and language development

Asset Metadata

Creator Hwang, Heeju (author)

Core Title Investigating coordination of lexical and structural information cross-linguistically

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Publication Date 07/25/2012

Defense Date 05/10/2012

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag English,Korean,lexical and structural information,OAI-PMH Harvest,sentence production

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kaiser, Elsi (committee chair), Biederman, Irving (committee member), Simpson, Andrew (committee member)

Creator Email everlucy@gmail.com,heejuhwang@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-65914

Unique identifier UC11290352

Identifier usctheses-c3-65914 (legacy record id)

Legacy Identifier etd-HwangHeeju-1002.pdf

Dmrecord 65914

Document Type Dissertation

Rights Hwang, Heeju

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA