:4..RmL$ I .V V V . . V V. V . , . . . V V V).
.5. V. M . V . V V . . uh}. V .
9.2.52.4}. r J... V V . V V . V V .V t. t1\1. \V I
”in”. ﬂaw... . ”i . V . . . V . .V V . . V . V H
ﬁx...“ PH m. a: . . . V . Lh .

é

 

5! .2: .
b '1. V,
.3. o... ‘.
.11... .3...‘ V .
ntu1tﬁn1t}. x I;
. , V V »A...lv\§kw.~cxdrl.¢\at1a1
V . . V .9... i {a .n .
«M: , . 31.2.1.5! V V V V . . with: :H-‘r; I!
. , .1 . . . , .
.1 ._ , . . . . V V n 1
. 11.5.... 5:. 5.". V 1: . . . . z
. 11.1. 2 . V
Hit-(2.111... rﬁ .
{.1321 I
Ll-fil Fiﬁ-IA V'-
. .u .1“ . dd u
‘1 :LJv . a
dd. .1

32......

W.
.......:..a s, z ...
$2! «QWVEnC: z...

I
a

 

ll.» 09
.s‘. .1

 

. , , V V Vt V l. J. .. . . . V 3.. 5.5;. .v . .V
V . . . ﬁlﬁﬂum: . . . «.5 mm

 

 

3 L18?“ ‘RY I
17‘ 3 Michigan State
3 University

 

 

This is to certify that the
dissertation entitled

SECOND LANGUAGE ACQUISITION OF COMPLEX
NOMINALS: THE ARCHITECTURE OF THE SECOND
LANGUAGE LEXICON, INDIVIDUAL DIFFERENCES, AND
SECOND LANGUAGE DEVELOPMENT

presented by
Junkyu Lee

has been accepted towards fulﬁllment
of the requirements for the

PhD. degree in Second Language Studies

Major Professor’s Signature
Dem m M /4 2m 7

Date

 

 

 

 

 

MSU is an Afﬁrmative Action/Equal Opportunity Employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
' TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE
Aw ~~ : “:12

3"
-1. 4L.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5108 K:lProj/Acc&PrelelRC/DateDue.indd

 

 

SECOND LANGUAGE ACQUISITION OF COMPLEX NOMINALS:
THE ARCHITECTURE OF THE SECOND LANGUAGE LEXICON, INDIVIDUAL
DIFFERENCES, AND SECOND LANGUAGE DEVELOPMENT
By

Junkyu Lee

A DISSERTATION

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

DOCTOR OF PHILOSOPHY
Second Language Studies

2009

ABSTRACT
SECOND LANGUAGE ACQUISITION OF COMPLEX NOMINALS:
THE ARCHITECTURE OF THE SECOND LANGUAGE LEXICON, INDIVIDUAL
DIFFERENCES, AND SECOND LANGUAGE DEVELOPMENT
By

Junkyu Lee

This dissertation investigates the question Of how the second language lexicon is
functionally organized and how lexical information is processed in real-time. An answer
to this question is pursued by looking at complex nominals (e.g., coffee cup), a rarely
explored linguistic target in psycholinguistics-based SLA research. While taking into
account ﬁrst language backgrounds and individual cognitive abilities, the focus is on
whether native speakers and non-native speakers interpret a complex nominal via
retrieval and/or computational routes.

Three groups participated in this study: (1) an L2 English group whose Ll
(Chinese and Korean) has a modiﬁer-head complex nominal structure (e.g., orange-juice),
(2) another non-native speaker group whose Ll (Spanish and Thai) has a head-modiﬁer
structure (e.g., juice-orange), and (3) an L1 English group as a control. All the
participants completed three computer-based tasks: (1) a lexical sense decision task, (2)
an inhibitory control test, and (3) a working memory test.

Results showed that native and non-native speakers relied on both retrieval and
computational routes in the recognition of complex nominals taken from a large corpus.

The L1 group tended to use more retrieval mechanisms whereas the L2 groups were

inclined to resort to computational routes. Also, the effects of cognitive abilities were
confounded with processing costs and LI background.

This study emphasizes that two key concepts (storage and computation) of the
language faculty are integral to L2 lexical acquisition. The contribution of empirical L2

data to linguistic inquiry is also highlighted.

Copyright by
JUNKYU LEE
2009

4.

This dissertation is dedicated to my family.

ACKNOWLEDGEMENTS

This dissertation would not have been possible without the help and support of numerous
people. Thus my sincere gratitude goes to all those people who have made this
dissertation possible. I would like to start by acknowleding each member of my
dissertation committee who has offered constant support and encouragement throughout
the processes of this dissertation: Eve Zyzik, who always gave me insightful comments
that have improved this dissertation in numerous ways; Rebecca F oote, whose extremely
helpful feedback made this dissertation psycholinguistically sound and substantial;
Shawn Loewen, whose constructive guidiances on quantitative methods made this
dissertation far more professional; ﬁnally my chair to whom I am most deeply indebted,
Susan Gass, who did not only support me intellectually and academically throughout my
Ph.D. studies but also supported me at every single step Of my doctoral research.

Special thanks must be made to my friends and colleagues who helped me in my
revisions and material development, including Marthe Russell, Jennifer Behney, Luke
Plonsky, Mark Shea, Jamie Thomas, and Robin Roots.

I gratefully acknowledge Joan Reid, Russ Werner, and Anna Davis, the staff of
the Second Language Studies program, for giving me all necessary assistance during my
studies.

The support and encouragement of many friends and colleagues helped me
overcome setbacks and stay focused on my graduate study; I am grateful to all the SL5
students including Shaofeng Li, Tomoko Okuno, Seongmee Ahn, Ben White, Xiaoqing

(Andy) Chen, Chien-I-Isiung (Scott) Chiu, Amy Thompson, Fei Fei, Hyewon Lee, Sun Yi

vi

Hong, Soo Hyon Kim, Grace Lee Amuzie, Jimin Kahng, Hyojung Lim, Yeon Heo, Susan
Berendes Wood, Hiep Chau, Ching-Ni Hsieh, Nobuhiro Kamiya, Kimi Nakatsukasa,
Maren Schierloh, Tetyana Sydorenko, Emma Trentman , Solene Inceoglu, Sara Hillman,
Wen-Hsin Chen, and Baburhan Uzum; I am also thankful to my colleagues at Yale and
Seoul National for their constant encouragement, including Hosung Nam, Hyoungbae
Lee, Seunghun Lee, Youngioo Lee, and Yoonshin Kim.

AISO, I would like to express my appreciation to the many professors in Korea and
the US who have led my academic life fruitful and successful, including Sungbom Lee,
Chungmin Lee, Jong-bok Kim, Seungho Nam, Maria Piﬁango, Larry Horn, Stephen
Anderson, Louis Goldstein, Maria Babyonyshev, Seungja Choi, Debra Friedman, Senta
Goertler, Debra Hardison, Charlene Polio, Diana Pulido, Patti Spinner, Paula Winke and
Patricia Paulsell.

I wish to express my gratitude to my parents (Min-soc Lee & Haeng-ja Lee) who
always supported and encouraged me during this period. I would also like to thank my
parents-in-laws (Myungho Ko & Haejung Byun) and my sister (J isun Lee) and her family
for their endless support and love. Last but not least I am deeply indebted to my son, Rich
(Jonghyun) Lee and my wife, Yoojung Ko, whose love and support provided me with the

energy to attain my goals.

Vii

TABLE OF CONTENTS

LISTS OF TABLES ........................................................................................................... xi
LISTS OF FIGURES ....................................................................................................... xiii
CHAPTER ONE - INTRODUCTION ................................................................................ 1
CHAPTER TWO - BAKGROUND ................................................................................... 8
2.1. Linguistic Target: Complex nominals ..................................................................... 8
2.2. L1 psycholinguistic models of compounds ............................................................ 13
2.2.]. Activation of the whole word versus activation of each constituent versus both
................................................................................................................................... 15
2.2.2. The activation Of relation: The locus of computation ..................................... 22

2.3. The theoretical importance of examining L2 complex nominal data in
understanding the structure of the mental lexicon ........................................................ 27
2.3.1. Within SLA research ....................................................................................... 27
2.3.2. Beyond SLA research ..................................................................................... 30
2.4. Compounding Factors that affect L2 complex nominal processing ...................... 32
2.4.1. Learner variables ............................................................................................. 32
2.4.2. An item variable: Frequency ........................................................................... 45
2.5. General objectives .................................................................................................. 47
CHAPTER THREE — RESEARCH DESIGN AND METHODOLOGY ........................ 48
3.1. Overview of the study ............................................................................................ 48
3.2. Task 1: Complex nominal processing in an interference paradigm ....................... 48
The research questions of Task 1 .............................................................................. 56
3.3. Task 2: Complex nominal processing and working memory capacity .................. 62
The research question of Task 2 ............................................................................... 66
3.4. Task 3: Complex nominal processing and inhibitory control ability ..................... 68
The research question of Task 3 ............................................................................... 71
3.5. Participants ............................................................................................................. 73
3.6. Method — Task 1 .................................................................................................... 76
3.6.1 . Materials ......................................................................................................... 76
3.6.2. Procedure ........................................................................................................ 77

viii

' A Pl-1

3.7. Method —Task 2 ..................................................................................................... 80

3.7.1. Materials and procedures ................................................................................ 80
3.7.2. Scoring Procedures ......................................................................................... 85
3.8. Method — Task 3 .................................................................................................... 86
3.8.1. Materials ......................................................................................................... 86
3.8.2. Procedures and scoring procedures ................................................................. 87
CHAPTER FOUR —RESULTS ........................................................................................ 89
4.1. Task 1 ..................................................................................................................... 89
4.1.1. Research question 1: The difference between the NS and the NNS groups 91
4.1.2. Research question 2: The difference between the LH and the RH groups ..... 94
4.1.3. Research question 3: Within-group differences .............................................. 95
4.1.4. Research question 4: Comparing the NS and the NNS groups with respect to
holistic representations of complex nominals ........................................................... 98
4.2. Task 2 ................................................................................................................... 103
4.2.1. NS group ....................... : ............................................................................... 103
4.2.2. RH group ....................................................................................................... 104
4.2.3. LH group ....................................................................................................... 105
4.3. Task 3 ................................................................................................................... 107
4.3.1. NS group ....................................................................................................... 108
4.3.2. RH group ....................................................................................................... 109
4.3.3. LH group ....................................................................................................... 110
CHAPTER FIVE - DISCUSSION .................................................................................. 1 13
4.1. The second language mental lexicon ................................................................... 113
4.2. The role of cognitive ability on comprehending complex nominals ................... 126
CHAPTER SIX - CONCLUSION .................................................................................. 132
APPENDIX A ................................................................................................................ 139
APPENDIX B ................................................................................................................ 140
APPENDIX C ................................................................................................................. 141
APPENDIX D ................................................................................................................ 142

ix

 

BIBLIOGRAPHY ........................................................................................................... 143

LISTS OF TABLES

Table 1. Levi (l978)’s categorizations Of semantic relations in complex nominals ........ 12
Table 2. Examples of complex nominals in Thai ............................................................. 32
Table 3. Less productive forms of complex nominals in Spanish .................................... 33
Table 4. Productive forms of complex nominals in Spanish ............................................ 34
Table 5. The top 20 high-frequency complex nominals with coffee as a modifier .......... 42
Table 6. The four experimental conditions in Task 1 ....................................................... 51
Table 7. The most frequent uses of juice as a head noun ................................................. 55
Table 8. Gender and age information Of participants ....................................................... 74
Table 9. The summary of the L2 proﬁciency tests .......................................................... 75

Table 10. The frequency and the number of syllables Of complex nominals in Task 1 77

Table 11. A sample of the working memory test ............................................................. 83
Table 12. The response times in the four experimental conditions .................................. 90
Table 13. The accuracy scores in the four conditions ....................................................... 91
Table 14. Bonferroni Post-hoc comparisons of response times ........................................ 93
Table 15. Bonferroni Post-hoe comparisons of accuracy scores ...................................... 93
Table 16. Post-hoc comparisons of accuracy scores of the NS group .............................. 96
Table 17. Post-hoe comparisons of accuracy scores of RH and LH group ..................... 97

Table 18. The descriptive statistics of the 96 test items according to the frequency of the
whole words ...................................................................................................................... 99

Table 19. The correlations between the frequency and the reaction times in the NS group

......................................................................................................................................... 100

xi

Table 20. The correlations between the frequency and the reaction times in the RH group
......................................................................................................................................... 101
Table 21. The correlations between the frequency and the reaction times in the LH group
......................................................................................................................................... 101
Table 22. A summary Of research questions and the corresponding ﬁndings in Task 1 102
Table 23. A summary of working memory scores .......................................................... 103
Table 24. A summary of research questions and the corresponding ﬁndings in Task 2 107
Table 25. A summary of inhibition scores ...................................................................... 108
Table 26. A summary of research questions and the corresponding findings in Task 3.

......................................................................................................................................... 112

xii

~1

LISTS OF FIGURES

Figure 1. An illustration Of a full-listing model ................................................................ 16
Figure 2. An illustration of a full-parsing model .............................................................. 17
Figure 3. An illustration of a dual-route model ................................................................ 21
Figure 4. A revised illustration of a full-parsing model .................................................... 26
Figure 5. A revised illustration of a dual-route model ...................................................... 26

Figure 6. An illustration of the potential relationship between WM capacity and complex
nominal processing ........................................................................................................... 38

Figure 7. The potential role of inhibitory control ability on complex nominal processing

with the framework of CARAN ........................................................................................ 44
Figure 8. The processing of a complex nominal in the SS condition ............................... 52
Figure 9. The processing of a complex nominal in the SM condition .............................. 52
Figure 10. The processing of a complex nominal in the MS condition ............................ 53
Figure 11. The processing of a complex nominal in the MM condition ........................... 53

Figure 12. A prediction for the difference between the RH and the LH group in terms of

reaction times .................................................................................................................... 57
Figure 13. A prediction for the NS group: Uniform interferences ................................... 59
Figure 14. A prediction for a NNS group: Graded interfere ............................................. 61

Figure 15. An illustration of the potential functions Of the WM capacity in the recognition
of a MM complex nominal ............................................................................................... 63
Figure 16. A prediction of the WM effect in case of retrievals or efﬁcient computations67

Figure 17. A prediction of the WM effect in case of the delayed computations .............. 67

xiii

Figure 18. An illustration of the potential functions of inhibitory control ability in the
recognition of 3 MM complex nominal ............................................................................ 69
Figure 19. A prediction of the inhibitory control ability in case of retrievals or efﬁcient
computations ..................................................................................................................... 72
Figure 20. A prediction of the inhibitory control ability in case Of the delayed
computations ..................................................................................................................... 73
Figure 21. The procedure of the lexical sense decision task ............................................. 79

Figure 22. A sample Of a block containing two equations and two recalls in the WM test

........................................................................................................................................... 85
Figure 23. The mean response times Of the three groups in the four conditions .............. 91
Figure 24. The accuracy scores in the four conditions ..................................................... 91

xiv

CHAPTER ONE - INTRODUCTION

In this dissertation, I will investigate the question of how the second language
lexicon is organized and how lexical information is processed in real-time. I pursue an
answer to this question by looking at complex nominals (e.g., coﬂee cup), which are
rarely explored within the psycholinguistic context of second language acquisition;
speciﬁcally, I focus on how native and non-native speakers interpret the meaning Of
complex nominals. At the same time, I take into account other potential variables that
may affect complex nominal processing by second language learners, including the
leamers’ ﬁrst language backgrounds, their individual cognitive abilities, and item-
frequency effects.

One might claim that words are the building blocks of language and language
Ieaming. Within the context Of second language (L2) acquisition, as Gass and Selinker
(2008) go so far as to state, the lexicon is presumably the most signiﬁcant component of
second language acquisition (p. 449). L2 lexical knowledge is, therefore, Of inevitable
importance in both production and comprehension of an L2 (e.g., Bogaards & Laufer,
2004; Gass, 1988; Nation, 2001; Read 2000).

The second language lexicon has been investigated in relation to retention (e.g.,
Hulstijn & Laufer, 2001), incidental learning paradigms (e.g., Pulido, 2003), receptive
and productive knowledge (e.g., Webb, 2005), L2 phonetics (e.g., Barcroﬁ & Sommers,
2005), and writing (e.g., Barcroﬁ, 2004). Despite the wide range of L2 lexicon studies to

date, relatively little attention is paid to the issue of L2 lexical representation (Jiang, 2000;

Wolter, 2001). Jiang (2004) correctly pointed out that “more basic and speciﬁc issues
related to the acquisition process, such as how lexical knowledge is represented in the
learner’s mind what stages a word goes through before it becomes an integrated part
of the learner’s lexicon, have received little attention” (p. 416). Given that the mental
representations are related to L2 learners’ performance, research into the L2 lexicon
should pay more attention to the architecture of the L2 lexicon.

In order to understand the mental architecture of the second language lexicon, I
pay attention tO the dual functions of the language faculty (e.g., Jackendoff, 1997; Pinker,
1999). The dual functions include storage (i.e., memorizing lexical units) and
computation (i.e., manipulating lexical units through rules). In the L1 literature, both
storage and computation underlie many explanations regarding the functional
organization of the lexicon or mental dictionary (e.g., Clashen, 1999; Ullman, Pancheva,
Love, Yee, Swinney, & Hickok, 2005). In other words, Ll studies suggest that an
appropriate characterization of the lexicon is likely to combine both computation and
storage.

The organization of the L2 lexicon is presumably similar to that of the L1 lexicon,
which implies that both storage and computation should be considered as integral to the
study of L2 lexical acquisition. To date, however, in L2 vocabulary research, the
consideration of this dual function is rarely applied. L2 vocabulary learning focuses

mainly on the storage of words (e.g., de Bot, Paribakht, & Wesche, 1997; J iang, 2002)

and large lexical units such as formulaic sequences (e.g., Schmitt, 2004; Wray, 2002)‘.

 

I See Abel (2003) for a notable application of the dual functions in examining L2 idioms.

Given that both storage and computation are integral in the lexicon, the computations of
lexical information should be as important as the storage of words.

In order to explore the dual functions in the second language lexicon, I chose
complex nominals as the linguistic target of this study. Complex nominals in this
dissertation refer to the combination of two nouns such as coﬂee cup. Complex nominals
provide an informative clue to understanding functional organization of the lexicon. A
theoretical question regarding the representation and processing of complex nominals is
whether a complex nominal such as coffee cup is represented as a single word (i.e.,
coffee-cup) or as two words (i.e., coﬂee and cup) or as three words (coffee, cup, and
coffee-cup). This question corresponds to the theoretical ramiﬁcations of complex
nominals: (l) full-listing models in support of holistic representations, (2) full-parsing
models in favor of discrete representation, and (3) dual-route models integrating full-
listing and full-parsing models.

Complex nominals in English have been used as a key linguistic target in L1
research literature to reveal the structure of the mental lexicon, including Ll acquisition
(e.g., Clark, 1993; Nicoladis, 2003), L1 psycholinguistics (e.g., Gagné, 2001; Libben,
1998), L1 neuro-imaging studies (e.g., F iorentino & Poeppel, 2007), and L1 aphasia
studies (e.g., Badecker, 2001). One reason for the importance is related to the linguistic
productivity of complex nominals in English. An intriguing aspect of complex nominals
is that English native speakers interpret the meaning of novel complex nominals such as
beach beverage easily and rapidly (Gerrig & Murphy, 1992; Murphy, 1990; Potter &

Falconer, 1979; Springer & Murphy, 1992). The high productivity of complex nominals

also entails that there always co-exist both familiar and novel complex nominals, which
provides a good venue to examine the dual functions in the lexicon.

Insofar as the computations of complex nominals are concerned, previous Ll
literature (e.g., Gagné, 2002) suggests that the semantic integration of a complex nominal
involves two processes: (1) the meaning activation of each constituent in complex
nominals and (2) the establishment of meaningful relations to combine each constituent.
When interpreting coffee cup, for example, one activates not only the meaning of coffee
and the meaning of cup but also one search for a meaningful relationship of the two
words as a whole. This semantic integration would yield greater processing loads than a
simple retrieval of the stored lexical representation. Given that L2 processing is less
efﬁcient than L1 processing, exploring the semantic integration process can provide an
important insight as to how L1 and L2 speakers differ in terms of processing complex
nominals.

At the same time, individual differences in how complex nominals are stored
and/or computed may be related to diverse factors including the learners’ ﬁrst language
backgrounds and their individual cognitive abilities. First, the headedness of the mother
tongue may affect the recognition processes of L2 complex nominals because languages
differ with respect to the headedness of compounds (right-headed compounds such as
orange-juice vs. left-headed compounds such as juice-orange). Thus, L2 learners whose
ﬁrst language has a left-headed structure may ﬁnd it more difﬁcult to process right-
headed L2 complex nominals than L2 learners whose compound structure in the ﬁrst
language is right-headed. Secondly, linguistic cognitive abilities including working

memory capacity and linguistic inhibitory control ability may play a role in processing

L2 complex nominals. Individuals differ in terms of their cognitive ability to inhibit or
suppress irrelevant information among competing sources. This ability is related tO
inhibitory control ability and working memory capacity, which are both an important part
of executive functioning that controls and manages cognitive processes (Anderson &
Green, 2001; Gass & Lee, in press; Green, 1998; Norman & Shallice, 1986). From the
perspective of the functional architecture of the L1 and L2 lexicon, cognitive abilities
such as working memory capacity and inhibitory control ability are likely to be
associated with computational mechanisms, but not with retrieval systems. A reason for
the potential effects of the cognitive abilities on computing complex nominals is that the
cognitive abilities are observable in higher-order cognitive processes (e.g., Daneman &
Carpenter, 1980; Waters & Caplan, 1996; Juffs, 2004). As illustrated above, semantic
integration involves the resolution of competing information, which is related to higher
processing costs than a simple retrieval of information from long-term memory. Thus,
those who are better at inhibiting irrelevant information or in maintaining relevant
information might be more efﬁcient in semantic integration processes associated with
complex nominal processing than those who are not. Within the broader psycholinguistic
context, the importance of individual variables in general language processing research
has been well-documented (e.g., Daneman & Carpenter, 1980; Just & Carpenter, 1992).
However, the roles of individual cognitive factors have rarely been explored in L1
complex nominal processing, presumably because L1 processing of complex nominals is
too efﬁcient and automatized to be observed. In this regard, L2 processing of complex
nominals could offer one way to examine the function of cognitive variables in L2

processing, given that L2 processing is generally less efﬁcient.

Within the domain of second language acquisition, research needs to account for
the interpretive processes of complex nominals by L2 learners, which includes not only
how word meanings are accessed, but also how meanings are retrieved or created by the
concepts Of known words. In other words, L2 vocabulary research should integrate into
the research agenda both computation and storage, a key in characterizing the attributes
of complex nominals.

Beyond the domain of second language acquisition, researching L2 complex
nominals has great potential for enriching the research basis of the mental lexicon.
Particularly, interlanguage complex nominal data has certain methodological (e.g., the
use of “authentic” or “real” complex nominals) and theoretical advantages (i.e., the less
efﬁcient/automatized processing system of L2 learners), which are difﬁcult to examine
through native speakers’ performance in normal circumstances.

This dissertation is the ﬁrst attempt, to the best of my knowledge, (1) to examine
the representation and processing Of complex nominals by second language learners, (2)

to investigate complex nominal processing by native and non-native speakers within the
context of an interference paradigmz, and (3) to explore the effect of cognitive abilities

(e.g., working memory capacity and inhibitory control capacity) on complex nominal

 

2 I hypothesize that the processing costs of semantic integration would differ in accordance with
the number of meanings in each constituent. In this regard, I put forward four logically possible
combinations of complex nominals; (1) single meaning - single meaning (e.g., coffee cup), (2) single
meaning - multiple meanings (e.g., violin bow), (3) multiple meanings — single meaning (e.g., well water),
and (4) multiple meanings - multiple meanings (e.g., lead mine). In general, I expect that the interferences

related to each condition would differ. See Task 1 for more details.

processing, with two L2 leamer groups whose Ll differs in terms of the headedness Of
compound structure.

The structure of this dissertation is as follows. Chapter 2 will give the necessary
background for the various issues related to complex nominal processing. Chapter 3
consists of three tasks: Task 1 examines the representation and processing of complex
nominals via an interference paradigm; Tasks 2 and 3 are concerned with the effect of
two cognitive variables (i.e., inhibitory control ability and working memory capacity) on
L2 complex nominal processing. Each task includes speciﬁc tasks, research questions,
and results. Chapter 4 discusses the results of the three tasks in relation to the architecture
of the second language mental lexicon. Chapter 5 concludes the dissertation not only with
future research directions but also with the contributions of this dissertation to language

science and to second language acquisition.

CHAPTER TWO - BAKGROUND

2.1. Linguistic Target: Complex nominals

This dissertation is concerned with how native speakers and non-native speakers
interpret a combination of two nouns (i.e., complex nominals) in English with a special
reference to the dual functions in the lexicon. Particularly, the focus of this study is
familiar compounds such as orange juice and coﬂee cup. The familiarity of compounds is
operationalized as the existence of compounds in a large English corpus containing
spoken and written data.

In English, complex nominals are one of the most well known examples of
linguistic productivity as the creation of complex nominals is a highly productive
morphological process (e.g., Adams, 1973; Bolinger & Sears, 1981; Bauer, 2001; Clark,
1993; Katamba, 1993; Stockwell & Minkova, 2001). The high productivity of complex
nominals using two nouns implies that there always exist novel complex nominals that
are understood in a language community (e.g., Downing, 1977; Levi, 1978). Thus,
complex nominals are a shortcut to enriching the lexicon in the sense that single words in
the lexicon are used to create a new complex nominal.

In what follows, I ﬁrst narrow the scope of the targets of this study with respect to
terminology and writing conventions and then, I describe some linguistic characteristics
of complex nominals that are relevant to the focus of this study: (1) headedness, (2) the
semantic transparency Of each noun in compounds, and (3) the semantic relation between

the two nouns.

 

Terminology. Various names for complex nominals have been used in the
literature. In linguistics, complex nominals (e.g., Levi, 1978) are called noun-noun
compounds (e.g., Clark, 1993; Warren, 1978), root-root compounds (e.g., Nicoladis,
2007), nominal compounds (e.g., Zimmer, 1972; van Jaarsveld & Rattink, 1988; Blanken,
2000), or nominal compounding (e.g., Bauer, 1978; Downing, 1977; Zimmer, 1971,
1972). In cognitive psychology, complex nominals often refer to concept combination

(e.g., Costello & Keane, 2001; Estes & Glucksberg, 2000). For the sake Of ease of

f

exposition, I will consistently use “complex nominals” to encompass all the names in the
literature and will not make any theoretical distinction among the various terms.

Writing conventions. English complex nominals have typically three forms: (1) a
space between two nouns (e.g., coffee cup), (2) a hyphen between two nouns (e.g., tape-
record), and (3) no space between two nouns (e. g., football). A common account of how
these three forms develop over time is that an English complex nominal tends to begin
with a space between two nouns, and then the complex nominal appeared to be
hyphenated when the complex nominal became recognized as a single word, and ﬁnally

the complex nominal does not have a space or a hyphen when it became a single word
(BOrjars & Burridge, 2001; Quirk, Greenbaum, Leech, & Svartvik, 1985)3. For example,

Jesperson (1968) pointed out that the word today originated from to and day, and also in

 

3 Yet, the historical explanation of the writing conventions of English complex nominals is not
always clear. Rather, many linguists pointed out that the writing conventions are a random process (Adams,
1973; Katamba, 1993; Napoli, 1996; Wardhaugh, 1995). For example, Quirk, Greenbaum, Leech, and
Svartvik (1985) demonstrated that some words in English can be used in the three forms such as ﬂower pot,

ﬂower-pot, and ﬂowerpot.

the 1920’s the hyphenated form to-day was used. The evolutionary account suggested

that open compounds that do have a space are less likely to be stored as a single word in
the lexicon, considering that compounds without a space are likely to be lexicalized. Thus,
this dissertation limits the scope of linguistic targets to open compounds which have a
space between two nouns.

Headedness. Complex nominals (e.g., coffee cup) consist of a modiﬁer noun (i.e.,
coﬁ’ee) and a head noun (i.e., cup). A head inherits the basic attributes of a complex
nominal such as syntactic and semantic information, whereas a modiﬁer functions as a
quality adjective (Dressler, 2006). For example, a coﬂee cup is a type of cup and inherits
its syntactic and semantic attributes Of being an inanimate concrete noun from the head
(i.e., cup); the modiﬁer coffee offers a more detailed meaning such as “a cup used for
coffee”.

Interestingly, languages differ with respect to the headedness of complex
nominals (Dressler, 2006; Jarema, 2006). In many languages such as Chinese, English,
and Korean, the head is positioned on the right (e.g., apple-juice) and thus complex
nominals in these languages are in the order of a modiﬁer and a head. Some languages
such as Hebrew, Spanish, Thai, and Vietnamese are left-headed (e.g., juice-apple),

although Romance languages tend to insert a preposition between two nouns (e.g., sugo
di mele ‘juice of orange’ in Italian)4. In these left-headed languages, complexnominals

have the order of a head and a modiﬁer.

 

4 See 2.4.1.1. for a more detailed description of compound structures of Romance languages.

10

Semantic transparency of two constituents. In English, most complex nominals
are endocentric (Bauer, 2001; Plag, 2003). In an endocentric complex nominal in English,
the meanings of each constituent are semantically transparent and contribute to the
interpretation of the complex nominal as a whole. Put differently, the meaning of the
whole compounds can be clearly guessed from its constituents. In a word such as coﬂee

cup, for example, both the meaning of coﬂee and the meaning of cup are semantically

transparent and contribute to the interpretation of coffee cup as a cup used for coffees. In
this dissertation, I focus on endocentric complex nominals in which both constituents in
the complex nominals are semantically transparent (e.g., coffee cup) because it is more
interesting to see how one interprets the whole meanings based on each constituent,
which he or she knows the meanings of both.

Diverse relations between two nouns. Random selections of two nouns cannot
lead to the creation of a complex nominal. Complex nominals can be interpreted only
when the relation of a modiﬁer and a head in a complex nominal is established in a
meaningful way. There exist diverse semantic relations between two nouns embedded in
complex nominals. For example, coﬂee cup is a cup used for drinking coffee, whereas
paper cup refers to a cup made of paper. Indeed, the analysis of diverse semantic
relations of complex nominals has been one of the most demanding tasks for linguists
(Downing, 1977; Jespersen, 1942; Levi, 1978; Marchand, 1960; Shoben, 1991; Warren,

1978)

 

5 In English there are also exocentric compounds such as skinhead and hatchback do not have a

head, and the meaning of the whole compounds cannot be clearly guessed from its constituents (Bauer,

1983, p. 203). In many cases, exocentric compounds do not have a space between two nouns.

ll

7.

Table I. Levi (l978)’s categorizations of semantic relations in complex nominals6

 

 

Predicates Examples Subject/Object
CAUSE] tear gas Object
CAUSE2 drug deaths Subject
HAVE] apple cake Object
HAVE2 lemon peel Subject
MAKE] silkworm Object
MAKEZ snowball Subject
USE steam iron Object
BE soldier ant Object
IN ﬁeld mouse Object
FOR horse doctor Object
FROM olive oil Object
ABOUT price war Object

 

Several earlier linguistic approaches have been contradictory regarding how relations are
used to combine individual constituents in compounds (e.g., Downing, 1977; Gleitman &
Gleitrnan, 1970; Kay & Zimmer, 1976; Levi, 1978). Some linguists (Gleitman &
Gleitrnan, 1970; Levi, 1978) claimed that there is a ﬁnite set of relations when a modiﬁer
and head are combined. For example, Levi (1978) proposed that a set of 12 relations can
explain the majority of the interpretation of complex nominals, as illustrated in Table ].
In contrast, other linguists (Downing, 1977; Kay & Zimmer, 1976) argued that complex
nominals could have an inﬁnite set of interpretations. In sum, although there was

divergence on the number and characteristics of the relations embedded in complex

 

6 The distinction between Subject and Object is based on the role of a modiﬁer (i.e., the ﬁrst
constituent). Levi argued that the relationship between a modiﬁer and a head can be characterized in terms
of a two-place predicate (subject-predicate-object). For example, tear gas means gas CAUSES tear, in

which the modiﬁer tear is in the position of object.

12

 

nominals, various linguistic approaches have agreed that the relation between modiﬁers
and heads plays an important role in interpreting complex nominals.

Lexicalized versus novel complex nominals. As mentioned earlier, complex
nominals are highly productive in English. The high productivity entails the new addition
of complex nominals to the mental lexicon. A useful way to conceptualize the mental
representation of complex nominals is a distinction between lexicalized and novel
complex nominals (Fiorentino & Poeppel, 2007; Gagne', 2002; Gagne’ & Spalding, 2006;
Jaarsveld, Coolen, & Schreuder, 1994; Jaarsveld & Rattink, 1988; Libben, 2006;
Nicoladis, 2003; Plag, 2003). A lexicalized complex nominal (e.g., orange juice) refers to
a complex nominal that language users are familiar with because it is frequently used or
encountered. In contrast, a novel complex nominal (e.g., sand pie) refers to a complex
nominal that language users rarely or never encounter, but still understand the meaning of.

However, the distinction is not always clear-cut. Frequency is widely used as an
indicator of whether a complex nominal is lexicalized; the more frequently a word is used,
the more likely the word is lexicalized. The frequency-based assumption is a probabilistic
statement but not an absolute criterion. Gagné (2002: 724) notes that a continuum
between lexicalized and novel complex nominals could be a better way to characterize

the distinction between them.

2.2. L1 psycholinguistic models of compounds
Complex nominals, the target of this study, a type of multimorphemic word, have

been a linguistic focus in trying to broaden the understanding of the structure of the

13

mental lexicon. As described in 2.1., the recognition7 of a complex nominal can be
achieved (1) not only when language users know the meanings of each constituent in the
complex nominal (2) but also when they are able to establish the meaningful relationship
between the two constituents.

These two characteristics of complex nominals provide us a good rationale to
examine the dual functions in the lexicon. Psycholinguistically speaking, the
comprehension of complex nominals involves an interpretive process where (1) the single
meaning of the whole complex nominal exists in the mind or (2) the concepts of each
constituent are combined to express a single notion of a complex nominal. The two
interpretive processes of complex nominals allow us to examine an important insight not
only as to how word meanings are assessed, but also as to how meanings are created by
virtue of known familiar meanings (see Libben, 2006 for more theoretical importance of
complex nominals). In this regard, complex nominals provide a good venue to examine
the dual functions in the lexicon.

A theoretical question regarding the processing and representation of a complex
nominal such as orange juice is whether it is (1) stored as a whole, (2) stored discretely,
or (3) stored as three lexical entries (i.e., orange, juice, and orange-juice). There is a
theory in favor of whole-word representations (e.g., Bybee, 1995; McClelland, Rumelhart,
& the PDP research group, 1986; Rumelhart, McCelland, & the PDP Group, 1986); there

is also a theory in support of computations (e.g., Clahsen, 1999; Marslen—Wilson, Tyler,

 

I use “recognition”, “Interpretation”, and “comprehensron” Interchangeably. In this study, I do

not make any theoretical distinctions among them.

14

Waksler, & Older, 1994; Rastle & Davis, 2008; Rastle, Davis, & New, 2004; Taft, 1979 ,
2004); there is a theory that bolsters both whole-word representations and computations
(e.g., Baayen, Dijkstra, & Schreuder, 1997; Caramazza, Laudanna, & Romani, 1988;
Pinker & Ullman, 2002). Interestingly, the different models of complex nominals seem to
be related to the theoretical backgrounds of researchers in such a way that, in general,
psychologists prefer retrieval mechanisms associated with memory structures whereas
linguists favor computations in conjunction with the internal structures of complex
nominals. Although both the whole-word accounts and the computation-based accounts
appear plausible, insights from both linguistics (in favor of computation) and psychology
(in support of retrieval) seem necessary in order to examine the dual functions associated
with complex nominals.

In what follows, I sketch the theoretical background for the representation and
processing of complex nominals. The ﬁrst three accounts, as brieﬂy mentioned above, are
concerned with how the meaning of each constituent in a complex nominal is activated or
accessed: (1) a full-listing account, (2) a full parsing account, and (3) a dual-routes
account. The last two psycholinguistic accounts (the Competition-Among-Relations-In-
Nominals model and the schema-based model) are concerned with how each constituent

is integrated to obtain a meaningful interpretation.

2.2. 1. Activation of the whole word versus activation of each constituent versus both

2.2.1.1. F ulI-listing accounts versus F ull-parsing accounts
In the L] literature, there are two positions for the representations of multi-

morphemic words. The ﬁrst position supports the idea that the words are fully listed

15

whereas Opponents of this position claim that compound words are interpreted by a
compositional process of separate lexical entries.

According to the full-listing or full-storage accounts (e.g., Bybee, 1995;
Butterworth, I983; Manelis & Tharp, 1977; Monsell, 1985; Rumelhart, McCelland, &
the PDP Group, 1986), multimorphemic words such as complex nominals do not involve
compositional processes of morphemes, but rather they are stored as a “chunk” and
therefore directly retrieved from long-term memory. That is, these accounts see complex
nominals in terms of whole-word representations without resorting to complex internal
structures of the compound. Figure 1 illustrates how a complex nominal may be
represented in a full-listing model, in which the whole meaning of a complex nominal is

present in the long-term memory.

Figure 1. An illustration of a full-listing model

 

coffee cup

 

coffee cup

 

 

 

 

 

There are two main pieces of evidence in support of the full-listing account. First,
the formation of complex nominals is not ﬁilly rule-driven. Butterworth (1983), for
example, argued that compounds are stored and processed as whole words because
morphological rules cannot provide a full account for the productivity of compounds.

Secondly, the structures of complex nominals are characterizable in terms of frequency.

16

Bybee (1995), for instance, claimed that compounds are fully stored without abstract
representations of their complex structure. Her model argues that a constituent structure
is the effect of frequency-mediated associative activations of related forms. According to
her model, each individual word in a low-frequency compound will be recognized more
strongly as an independent constituent because they have stronger lexical activations.
Overall, the full-listing account claims that the meaning of a complex nominal is
represented as a whole word.

On the other extreme, compounds are assumed to be fully and invariably parsed.
This full-parsing approach claims that compounds are automatically and obligatorily
parsed into constituents via a decompositiona] route (e.g., Clahsen, I999; Giraudo &
Grainger, 2000; Marslen-Wilson, Tyler, Waksler, & Older, 1994; Rastle, Davis, & New,
2004; Taft, 1979, 2004). Figure 2 illustrates how a complex nominal may be represented

and processed in a full-parsing model.

Figure 2. An illustration of a full-parsing model

Computation

coffee

 

 

 

 

 

 

 

 

 

The main argument in support of the full-parsing models comes from the

constituent activation. That is, the meanings of each constituent in a complex nominal are

17

accessed during the processing of complex nominals. An early study by Taft and Forster
(1976) using a lexical decision task found that compounds undergo online decomposition,
arguing for the existence of morphological constituents in compounds. Furthermore, a
dense body of recent Ll psycholinguistic research has indicated that the constituent
activation is related to semantic transparency in recognizing complex nominals

(McQueen & Cutler, 1998). That is, the constituent activation in compound processing

out.

emerges at least when a constituent is semantically transparent. Priming for compounds,
for example, has identiﬁed that the interpretation of compounds is related to the
recognition of constituents in compounds, particularly with compounds having
semantically transparent constituents (Andrews, 1986; Inhoff, Briihl, & Schwartz, 1996;
Libben, Gibson, Yoon, and Sandra, 2003; Lima & Pollatsek, 1983; Sandra, 1990; Taft,
1994; Taft & Forster, 1976; Zwitserlood, 1994). Sandra (1990) found the priming effect
only with semantically transparent constituents but not with opaque constituents.
Zwitserlood (1994) reported that total and partial transparent compounds in Dutch are
associated with a priming effect. Libben, Gibson, Yoon, and Sandra (2003) examined
four types of compounds classiﬁed by semantic transparency: TT (transparent-transparent,
e.g., car-wash), OT (opaque-transparent, e.g., strawberry), TO (transparent-opaque, e.g.,
jailbird) and 00 (opaque-opaque, e.g., hogwash). They found morphological
constituency in all four types of compounds, regardless of semantic transparency.

A strong piece of evidence in favor of the full-parsing account is that people can
understand a novel compound without difﬁculty. Libben, Derwing, and de Almeida
(1999) examined the prelexical parsing of ambiguous novel compounds such as clamprod,

which could be interpreted either as clam + prod or as clamp + rod. Speciﬁcally, they

18

tested the two possibilities: (I) a ﬁrst possible parse where the ﬁrst possible constituent is
always parsed (i.e., clam + prod) and (2) a last possible parse where the constituent is
parsed until the parser reached the maximally possible constituent (i.e., clamp + rod).
They found that both interpretations emerged in a morpheme recall task and in a semantic
priming task, suggesting that morphological parsing yields multiple representations.
Recent studies suggest that even lexicalized complex nominals may undergo an
obligatory parsing process (e.g., F iorentino & Poeppel, 2007; Gagné and Spalding, 2004).
Fiorentino and Poeppel (2007) examined the morphological processing of lexicalized
compounds (e.g., airplane), using visual lexical decision with MEG
(Magnetoencephalography). They found the effect of early decomposition Of lexicalized
compounds, suggesting that morphologically complex words have internally structured
representations. Gagné and Spalding (2004) claimed that familiar compounds such as
snowball, which are likely to be lexicalized, seem to behave like novel compounds in the
sense that the familiar compounds get involved in computational processes. They found
that the interpretation of familiar compounds (e.g., snowball), patterned with that of novel
compounds when relation availability was manipulated. Speciﬁcally, participants
responded faster to target combinations (e.g., snowball, ‘a ball made of snow’) after
seeing a prime combination with the same relation (e.g., snowfort, ‘a fort made of snow’)
than that with the different relation (e.g., snowshovel ‘a shovel for snow’). The ﬁndings
do not necessarily contradict the presence of lexicalized complex nominals, while
indicating that even a lexicalized complex nominal is not irrelevant to a computational

process.

19

f‘

Overall, the full-parsing account claims that the meanings of each constituent are

accessed and combined via a computation.

2. 2. 1.2. Dual-routes accounts

Instead of taking such an extreme position as full-listing models and full-parsing
models, dual-route models use both a compositional route and a whole-word access route,
although there is divergence on how the two routes are utilized (e.g., Baayen, Dijkstra, &
Schreuder, 1997; Bertram, Schreuder, & Baayen, 2000; Caramazza, Laudanna, &
Romani, 1988; F rauenfelder & Shreuder, 1992; Laudanna & Burani, 1985; Pinker &
Ullman, 2002). That is, when a complex nominal is encountered, the language parser tries

to ﬁnd a uniﬁed representation of the complex nominal, as well as to construct the

meaning of the complex nominal using each of the constituentss. For example, the
Morphological Race (MR) Model, a parallel dual-route model, suggests that both
compositional and whole-word mechanisms are available and the two routes race in
processing complex nominals (Baayen, Dijkstra, & Schreuder, 1997). In general, the MR
model claims that familiar compounds are accessed as a whole-word because whole-word
access routes tend to win the race in processing familiar compounds. Figure 3 illustrates
how a dual-route account represents a complex nominal in its model. As in Figure 3, the
dual-route models not only search for the stored complex nominals (e.g., coﬂbe-cup) but

also attempt to combine two constituents (coﬂee and cup).

 

8 It should be also noted that the idea of dual function models are closely related to the
characterization of past tense in English. For example, Pinker (1999) claims that irregular forms are stored

as wholes while regular forms are computed/generated by rules (see Pinker and Ullman, 2002 for a review).

20

T

Much empirical evidence seems to support dual-route models in the L] literature
(e.g., Andrew, Miller, & Rayner, 2004; Inhoff, Radach, & Heller, 2000; Juhasz, Inhoff, &
Rayner 2005; Pollatsek, HyOna, & Bertram 2000). Inhoff, Radach, and Heller (2000)
found that German compounds are associated with a facilitative effect when there are
spaces between the constituents of compounds. However, the spaces in compounds had
interfering effects on the creation of an overall meaning of compounds. They interpreted
the ﬁndings as the involvement of both access to constituents and combining the
constituents together. Using an eye-tracking technique, Andrew, Miller, and Rayner
(2004) found that the frequency of the ﬁrst-constituent inﬂuenced ﬁrst ﬁxation, the
frequency of the second-constituent affected gaze duration, and the frequency of the

whole-word has a signiﬁcant effect on gaze duration and total looking time. They

 

proposed a process of g ‘ " through "' , in which compounds are

U

accessed via both constituent and whole-word representation.

Figure 3. An illustration of a dual-route model

Computation

 

21

Of relevance is the earlier discussion on characterizing complex nominals as a
continuum between lexicalized and novel complex nominals. Gagné and Spalding (2006b)
proposed that the presence of two independent parallel processes such as direct retrieval
from memory and an algorithmic route should be the most direct explicit methods of
viewing compound processing as a continuum from lexicalized to novel compound. For
example, Gagne & Spalding (2006c) claimed that while admitting the possibility of
storing familiar compounds, the even familiar compounds are fully parsed. They explain
the dual functions related to compound processing in terms of “availability” and
“familiarity”. A general assumption of compounds in the L1 literature is that a high
frequent compound (estimated by a corpus) is likely to be stored as a single lexical item
in memory. Gagne and Spadling named this possibility the “availability” in the sense that
the item is available in the memory and can be retrieved. However, they argue that high
frequency compounds may not simply mean holistic representations; rather, the high
frequency could mean “familiarity” in the sense that compounds that are encountered
very frequently could lead to easier computations.

In summary, the dual-route accounts incorporate both full-listing and full-parsing

models into their models.

2.2.2. The activation of relation: The locus of computation

The L1 psycholinguistic models in 2.2.1 questioned whether L1 speakers activate
a uniﬁed representation or separate representations of each constituent or both, when
processing complex nominals. Regarding the full-paring and the dual-function models,

however, the meaning activations of each constituent does not necessarily imply the

22

computations of the activated meanings of each constituent. As seen in 2.1., an
appropriate interpretation of a complex nominal requires the establishment of
combinations of two nouns in a meaningful relation. Thus, the integration of each
constituent in a meaningful way is as important as the activation of each constituent
and/or the whole word representation, in order to explain the recognition of complex
nominals in real-time. It is the obligatory combinatorial processes of each constituent in
complex nominals that provide use a rationale for the presence of computation in
processing complex nominal.

L] cognitive psychology research has paid attention to the psycholinguistic
mechanisms of establishing the relations of two nouns in novel complex nominals (Estes
& Glucksberg, 2000; Gagné, 2002; Gagné & Shoben, 1997, 2002; Murphy, 1988, 1990;
Wisniewski, 1996), suggesting that there is an obligatory process of selecting the
interpretable relation between two nouns to establish a uniﬁed representation. Overall,
the interpretation of complex nominals can be characterized either in terms of thematic
relations, for example ‘IN’ for the complex nominal mountain cloud (Gagne & Shoben,
1997) or with respect to property attribution, for example, ‘mushroom-shaped’ for the
complex nominal mushroom cloud (Estes & Glucksberg, 2000).

In the L] psycholinguistic literature, the activation of a relationship between two
nouns has been tested by examining whether previous exposure to a complex nominal
(e.g., orange juice) facilitated the interpretation of a subsequent complex nominal (e. g.,
strawberry juice) which has the same semantic relation as the previous one. Two major
lines of psycholinguistic research have developed which depend on whether either

modiﬁers or heads in complex nominals play a key role in establishing a relation between

23

two nouns in complex nominal processing: A schema-based theory (Murphy, 1988, 1990;
Wisniewski, 1996) versus CARIN (competition among relations in nominals) theory
(Gagné & Shoben, 1997, 2002). In general, the schema-based theory proposes that head
nouns in complex nominals are responsible for activating the relational structures while
the CARIN theory claims that modiﬁer nouns in complex nominals activate the relational
structures of the complex nominals.

According to the schema-based theory (a slot-ﬁlling approach), the head noun in a
complex nominal is represented by a structural set of slots and ﬁllers. The schema-based
theory claims that, in the course of concept combination (i.e., the interpretations of
complex nominals), the modiﬁer (the ﬁrst noun in English complex nominals) serves as a
ﬁller for a slot in the head noun (the second noun). For example, paper cup (a cup made
of paper) is interpreted by taking the modiﬁer paper into the MATERIAL slot within the
schema for cup. The head noun (i.e., cup) provides a slot where a modiﬁer (i.e., paper)
can be inserted. Therefore, the head noun plays a key role in interpreting a complex
nominal. In this theory, more availability of the slot in a head will lead to easier
interpretations of complex nominals. Conversely, no appropriate slot in the head noun
will result in nonsensical interpretations of complex nominals. For example, the reason
that a complex nominal air cup is nonsensical or difﬁcult to interpret without a context is
that the head noun cup does not have an appropriate slot for the modiﬁer noun air.

In contrast, the CARIN theory does not assume the replacement of a modiﬁer
inside the head noun’s schema. Instead, the CARIN theory emphasizes the role of the
modiﬁer (i.e., the ﬁrst noun) in complex nominal processing. It argues that the modiﬁers

trigger/activate the competition of possible semantic relations with the heads (the second

24

T

noun). Through the competition, the most appropriate semantic relation is selected and
used to connect the modiﬁer and head noun concepts. In the CARIN theory, the more
available the required relation, the easier it is to interpret complex nominals For example,
when mountain is a modiﬁer, the relation used to interpret a complex nominal is likely a
LOCATION relation (e.g., mountain lake) but less likely an ABOUT relation (e.g.,
mountain magazine).

When connecting the relation activation models to the previously discussed
constituent activation models, the relation activation models are compatible with the full-
parsing account and the dual-routes account because relation activation assumes
compositional processes. For example, Gagné (2002: 733) claimed that two routes might
be available for selecting a relation during conceptual combination: (1) an algorithmic
route that combines the two constituents based on information in the modiﬁer’s relational
distribution and (2) the retrieval route.

This dissertation will not test whether the schema-based theory or the CARIN
theory is more plausible. Rather, the reason that I introduce the two models is that, when
computational routes are assumed to be used, the recognition of complex nominals
requires the inhibitions of irrelevant relational structures among many possible relations.
Put differently, the interpretations of complex nominals are not a simple union of two
constituents in a complex nominal. What both the schema-based and the CARIN theories
have in common is that there are not only the activations of relational structures that
semantically connect each constituent in a complex nominal but also an obligatory
selection process of relationships between two nouns. Therefore, I assume that, during

the selection processes, computations should be a part of complex nominal processing.

25

Figure 4 illustrates how computational processes may occur during the recognitions of
complex nominals by taking the processes of relation selection into account. Figure 5 is a
modiﬁed version of a dual-route model where computational processes are a search for
an appropriate relation between each constituent.

Figure 4. A revised illustration of a full-parsing model

 

 

 

 

 

 

 

coffee cup

 

 

 

 

 

 

 

 

 

 

 

 

 

26

2.3. The theoretical importance of examining L2 complex nominal data in

understanding the structure of the mental lexicon

2.3. 1. Within SLA research

The psycholinguistic attributes of complex nominals (i.e., storage and
computation) are closely analogous to the architecture of the mental lexicon proposed in
the L] literature, where it has been suggested that both computation and storage are
necessary for an appropriate characterization of the lexicon (e.g., Clashen, 1999;
Jackendoff, 1997; Pinker, 1999; Ullman, Pancheva, Love, Yee, Swinney, & Hickok,
2005). For example, Libben (2006) claimed that the lexicon may be functionally
structured to maximize both computation and storage in the mental representation.
Therefore, the notions of computation and storage in the lexicon may not be mutually
exclusive; but complementary, reﬂecting behavioral patterns of language users, and
therefore, the individualized architecture of lexicons.

However, the role of computation in the lexicon has not been highlighted in L2
vocabulary learning (see Abel, 2003 for a notable application of dual functions in L2
idiom research). Much L2 vocabulary research tends to be storage-based (e.g., de Bot,
1992; de Bot, Paribakht, & Wesche, 1997). For example, one of the most common
research areas in L2 vocabulary learning is the study of formulaic sequences such as as
soon as and on the contrary. The formulaic sequences refer to “a sequence, continuous or
discontinuous, of words or other meaning elements, which is, or appears to be,
prefabricated” (Wray, 2000: 465). In general, the formulaic-sequence studies support the
holistic representations of formulaic sequences; that is, formulaic sequences are stored as

whole words and retrieved from long-term memory (Altenberg, 1998; Jiang & Nekrasova,

27

2007; Raupach, 1984; Schmitt & Carter, 2004; Weinert, 1995; Wray, 2002). The holistic
representations of formulaic sequences seem plausible considering that the whole
meaning of a formulaic sequence such as as soon as is more important in production and
comprehension than the meanings of individual words in a formulaic sequence.

There are two important differences between the current study and the previous
studies of formulaic sequences: (1) The semantic transparency of constituents in multi-
words and (2) the productivity. First, a study of formulaic sequences such as as soon as is
concerned not with how each constituent in the formulaic sequence is combined. That is,
each constituent in a formulaic sequence such as as soon as is not necessarily
semantically transparent in the sense that the meaning of the whole formulaic sequence
may not be guessed from the constituents made up of the formulaic sequence. In contrast,
the current study is concerned with a case where two words are semantically transparent
(e. g., apple juice) while the formulaic-sequence literature is mostly based on a case where
the meaning of the whole phrase (e.g., as soon as) is useful. Note that, as mentioned in
2.2.1., the meaning activation of each constituent is a key idea of investigating the
validity of either full-parsing or dual-route models of complex nominals. A clear point in
Section 2.2. is that the interpretation of complex nominals involves lexical access to both
the modiﬁer and head noun concepts and/or to the whole-phrase concepts as well as
computational or retrieval processes. The understanding of complex nominals must
include not only how word meanings are accessed, but also how meanings are retrieved
or created by the concepts of known words. Contrary to the formulaic-sequence studies,
therefore, this study is in a better position to examine the role of both computations and

retrievals in L2 lexical, morphological processing.

28

The second difference between the current study and the previous studies of
formulaic sequences is that the current study is concerned with a highly productive
morphological process while the studies of formulaic sequences deal with prefabricated
or ready-made lexical items.

Novel complex nominals are less likely to be stored as a whole in the lexicon because
the creation of a new complex nominal implies the absence of the complex nominal in the
lexicon. An intriguing aspect of complex nominals is that English native speakers
interpret the meaning of novel complex nominals easily and rapidly (Gerrig & Murphy,
1992; Murphy, 1990; Potter & Falconer, 1979; Springer & Murphy, 1992). Thus, if non-
native speakers know the meanings of each word, they may not show difﬁculty in
understanding the nominal. This, in turn, suggests that they can produce meaningful
computations using the known meanings of each word, which is not the part of research
questions of formulaic-sequence study.

While highlighting the importance of computations in the L2 lexical processing,
this study would provide empirical support for researching stored chucks because it
handles both computations and retrievals. A general assumption of the usage-and-storage
based account of formulaic sequence is that the more proﬁcient an L2 learner, the more
words stored in their mental lexicon (e.g., Ellis, 2005; Schmitt, 2004). In the current
study, L2 learners may be more likely to rely more on computational routes in processing
even familiar complex nominals because familiar complex nominals may function as
novel complex nominals. The more reliance on computational routes by L2 learners

would indicate that L2 learners are less efﬁcient in using retrieval routes or have less

29

number of stored lexical information, providing a good rationale to examine the role of
stored lexical information.

In summary, complex nominals allow us to examine the role of computation and
storage in the lexicon. I do not argue against the usage-and-storage based L2 vocabulary
accounts, dominant in second language acquisition research. However, the accounts
should accommodate another key aspect of the lexicon, that is, the mental computation of
words. L2 vocabulary research becomes more balanced by integrating two key aspects

(i.e., storage and computation) of L2 vocabulary acquisition.

2.3.2. Beyond SLA research

Complex nominals have been used as a key linguistic target in the L1 research
literature to reveal the structure of the mental lexicon in relation to dual functions,
including Ll acquisition (e.g., Clark, 1993; Nicoladis, 2003), L1 psycholinguistics (e.g.,
Gagné, 2001; Libben, 1998), L] neuro-imaging studies (e.g., Fiorentino & Poeppel, 2007)
and L1 aphasia studies (e.g., Badecker, 200]). However, second language research has
not been reﬂected in the mainstream of complex nominal studies.

Empirical research, as Juffs (200] , p. 31]) noted, enables SLA researchers to
contribute to L1 and L2 theoretical literature through the testing and validating of
theoretical constructs. Researching L2 complex nominal processing will enrich the
research basis Of the mental lexicon. Methodologically, L2 research is able to use
“authentic” or “real” complex nominals that can be found in a large corpus because the
real complex nominals may not be stored in L2 learners’ minds. L1 research has heavily
relied on novel compounds to examine how conceptual information is combined because

familiar compounds are likely to be stored in memory, yet novel compounds must

30

7..

involve combinatorial processes to be interpreted (e.g., Gagné & Shoben, 1997; Gagné,
200]). A methodological issue of using novel compounds is related to the familiarity and
plausibility of the novel compounds, as pointed out by Wisniewski and Murphy (2005).
That is, the use of novel compounds leads to difﬁculty in controlling the familiarity and
plausibility of the stimuli, which is well-known for signiﬁcant variables for online studies.
In this regard, the L2 data based on the responses to “real” complex nominals are
appealing and provide an important clue of the continuum between lexicalized and
unlexicalized complex nominals.

Theoretically, the less efﬁcient processing system of L2 learners could provide a
better understanding of the continuum between lexicalized and novel complex nominals,
enriching the research base on the mental lexicon. Morphological processing by native
speakers (NS) tends to be too efﬁcient and quick to observe the mechanism, and therefore
the processing is typically invisible. For example, Gagne and Spalding (2006c, p.13) state
that the relation selection process in complex nominal processing cannot be observed in a
normal circumstance with native Speakers. Rather, the processing of complex nominals is
observable when the meaning is difﬁcult to access, as in aphasia (e.g., Badecker, 200];
Blanken, 2000; Delazer & Semenza, 1998; Dressler & Denes, 1989; Libben, 1998;
Rochford & Williams, 1965) or when the relation selection process is accelerated via
priming paradigms (e.g., Estes & Jones, 2006; Gagné, 200] , 2002; Gagné & Shoben,
2002; Raffray, Pickering, & Branigan, 2007). In this regard, the less efﬁcient processing
ability of non-native speakers as compared to native speakers could provide an important

clue to complex nominal processing.

3]

 

2.4. Compounding Factors that affect L2 complex nominal processing

In this section, I outline the variables that may inﬂuence L2 learners’ performance
when they process complex nominals. Speciﬁcally, I put forward two learner variables
(i.e., the headedness of L] compounds and cognitive ability) and one item variable (i.e.,

frequency).

2.4. l. Learner variables

2. 4. I . 1. L1 inﬂuence: Headedness

L1 inﬂuence is a perennial, interesting topic in second language research. As
described in Section 2.1., languages differ with respect to the headedness of compounds.
In Chinese and Korean, compounds are always right-headed and thus have an identical
compound structure in terms of the position of the head. In Korean, for example, the
English translation of math teacher is swuhak ‘math’ + senseeyngnim ‘teacher’, where
the head ‘teacher’ is in the same right-position as in English (see Sohn 1999 for more
details about complex nominals in Korean). Thus, the compound structure of English and
Korean are identical at least on the surface, in terms of the position of the head and

modiﬁer, although detailed linguistic analyses will differ.

Table 2. Examples of complex nominals in Thai

 

 

Complex nominals Thai English translations

/aahaan dék/ ‘baby food’ > food + child

/caan khaaw/ ‘rice plate’ > plate + rice

loan dﬁan/ ‘monthly money’ > money + month
/thurj thaaw/ ‘socks’ > bag + foot

 

32

 

In contrast to English, Chinese, and Korean, compounds in Thai and Spanish are
left-headed, although there are some right-headed compounds in Thai and Spanish. For

example, a corresponding expression of an English word ‘house key’ in Thai is /kuncea/

‘key’ + /baan/ ‘house’, in which the head of the compound is the ﬁrst constituent9. Other
left-headed compounds in Thai can be seen in Table 2 (Iwasaki, Ingkaphirom, and Horie,
2005, p. 37).

Spanish is somewhat different from Thai in terms of the productive forms of
complex nominals, although both the languages share the left-headed structure of
complex nominals. In Spanish, a simple combination of noun and noun is less productive.
Table 3 shows a handful of examples for noun-noun compounds without a preposition

(Bradley & Mackenzie, 2004, p 304).

Table 3. Less productive forms of complex nominals in Spanish

 

 

Complex nominals in Spanish English translations
Pagina web Web page

Homo microondas Microwave

Video juego Computer game
Satellite espia Spy satellite

Placa madre Motherboard

 

 

9 According to Iwaski, Ingkaphirom, and Horie (2005), Thai has a limited number of right-headed
complex nominals. An example of right-head compounds is /muu thuu/ ‘hand-held telephone’, in which

/muu/ means ‘hand’ and /thuu’ refers to ‘carry’.

33

A more common case of compounds in Spanish is that two constituents in a compound

are linked by a preposition (Bradley & Mackenzie, 2004, p 304), as in Table 4.

Table 4. Productive forms of complex nominals in Spanish

 

 

Complex nominals in Spanish English translations
Jugo de piﬂa Pineapple juice
Ropa de playa Beachwear

Hoja de calculo Spreadsheet

 

A handful of studies have examined L1 inﬂuence on the production and
comprehension of L2 complex nominals with respect to the headedness of L1 complex
nominals. Slabakova (2002) tested whether the differences in L1 morphological structure
inﬂuence L2 performance, by comparing French learners of Spanish and English learners
of Spanish. In Spanish, a language with left-headed compounds, as described above, a
productive morphological process of compounding is to insert an obligatory preposition,
typically de, into two nouns in a compound, as in zapatos de tango (tango shoes); thus,
ungrammatical is a compound without the preposition such as zapatos tango. The
patterns are also true in French. Importantly, in English, both noun-noun compounds (e.g.,
tango shoes) and their phrasal counterparts (e.g., shoes for tango) are allowed. In forced
multiple choice questions, Slabakova (2002) found that French learners of Spanish
outperformed signiﬁcantly English learners of Spanish in rejecting ungrammatical noun-
noun compounds and complex predicates, when both NNS groups have comparable low
levels of Spanish proﬁciency. She claimed that the ﬁndings indicate the L1 inﬂuence of

morphological structures on comprehending the L2 constructions because the

34

constructions are not used or ungrammatical in French (as in Spanish) whereas they are
available in English.

Also, Liceras and Diaz (2000) have examined whether the headedness of L1
complex nominals affects the production of L2 complex nominals. In picture-naming
tasks, they examined learners of Spanish with different levels of L2 proﬁciency, in terms
of (l) noun-noun compounding strategies and (2) the trigger of the L2 acquisition of
noun-noun compounding. They argued that noun-noun compounding is not a marked
option because both intermediate and advanced learners of Spanish produced noun-noun
compounds frequently, which are not a productive process in Spanish. Also, they claimed
that directionality (changing from a right-headed to a left-headed word order) triggers the
L2 acquisition of Spanish noun-noun compounds because all learners, regardless of their
Ll backgrounds and L2 proﬁciency, did make mistakes in producing gender-marking
patterns, in which only the head is supposed to bear gender-marking morphemes in
Spanish.

Some bilingual child studies also have identiﬁed compound reversals, in which
the order of a head and a modiﬁer is reversed. Nicoladis (1999) reported that, in
spontaneous speech, a three-year old French-English bilingual child reversed the order of
French compound approximately 60 % and that of English compounds about 20%. Also,
Nicoladis (2002) found that in a production task, French-English bilingual children
reversed compounds in English more often than monolingual English-speaking children.
However, she did not ﬁnd the performance difference between the bilingual children and

the monolingual children in a comprehension task.

35

The aforementioned L2 and bilingual studies suggest that the headedness of the
L1 may affect the processes of L2 complex nominals. Yet, the previous studies are based
on off-line measures of comprehending complex nominals or on less controlled
production tasks. Thus, this study examines the role of L1 inﬂuence on the processing of
L2 complex nominals via online elicitation tasks.

In general, the difference in the headedness among these languages offers an
interesting opportunity to examine the L1 effect on L2 complex nominal processing
during second language acquisition. The L1 inﬂuence is that there may be differential
interference effects, depending on the convergence and divergence between mother and
target languages. If the headedness of the L1 compounding structures plays a role, for
example, I expect that Korean learners of English may encounter less interference in
processing English compounds than Thai learners of English because English and Korean

are right-headed whereas Thai is left-headed.

2.4.1.2. Cognitive abilities

2.4.1.2.] . Working memory Capacity

Numerous empirical studies suggest that working memory (WM) capacity is the
locus of individual variations in various high-order cognitive tasks including language
processing. Individuals with high WM capacity are suggested to be more efﬁcient than
individuals with low WM capacity in highly-demanding cognitive tasks including L1
(e.g., Daneman & Carpenter, 1980; Waters & Caplan, 1996) and L2 language processing

(Harrington & Sawyer, 1992; Juffs, 2004). Assuming that complex nominal processing

36

involves computation, it is possible that high WM individuals may be faster at combining
the meanings of complex nominals than their low WM counterparts.

Working memory (WM) capacity involves the cognitive processes required to
perform mental operations on chunks of information and the temporary storage of those
chunks (Baddeley & Hitch, 1974; Cowan, 2005; Hasher, Lustig, & Zacks, 2007; Kane &
Engle, 2003). Thus, WM encompasses both processing and storage of information.
Importantly, a general consensus on WM is the complementary interaction between
processing and storage; automatic processing gives more rooms for storage while
controlled processing uses more spaces for storage (see Baddeley 2003 for an overview).

In cognitive science, there are many models of WM and there is considerable
controversy on the exact nature of WM. One model looks at WM as the locus of
controlled attention (Cowan, 2005 ; Cowan, Chen, & Route, 2004; Kane & Engle, 2003);
Inhibition interprets WM as the ability to suppress irrelevant information (e.g., Hasher,
Lustig, & Zachs, 2007 ; Gemsbacher & Faust, 1991, Gunter, Wagner, & Friederici, 2003);
Activation models claim that WM plays a key role in keeping relevant information active
(e.g., Miyake, Just, & Carpenter, 1994; Just & Carpenter, 1992); Another model
describes WM as a task-switching ability (Hegarty, Shah, & Miyake, 2000). Importantly,
the various models of WM are not mutually exclusive. Rather, they are complementary in
terms of the trade-off between processing and storage of information.

The processing of a complex nominal involves the semantic integration of two
nouns, in which the two nouns should be combined in a meaningful relationship. As
described earlier, the semantic integration of the complex nominal is likely to involve the

activations of various relational structures and an obligatory process of selecting a

37

 

relational structure that gives an appropriate interpretation of the complex nominal. In
this process, WM may play a role. Let us relate complex nominal processing to the
various models of WM; one who is good at inhibiting irrelevant relational structures
among many possible relations would be efficient in processing the complex nominal;
one who has a better controlled attention would be faster in selecting a best relational
structure and in suppressing irrelevant structures; one who are better at keeping relevant
relational structure would be effective in processing the complex nominal. Therefore, it is
worthwhile to examine whether high WM individuals may be faster at the semantic
integration of complex nominals than their low WM counterparts. Figure 6 illustrates this
process; the two solid lines represent the selection/maintenance of a relevant relation

between coﬂee and cup, while dotted lines shows the inhibition of irrelevant relations.

Figure 6. An illustration of the potential relationship between WM capacity and complex
nominal processing

 

 

 

 

 

coffee cup

 

 

 

 

An assumption I made is that the presence of computation in complex nominal
processing allows us to examine the potential role of WM capacity in the processing. As
discussed earlier, there are at least three models of complex nominal processing: (1) a

full-listing model, (2) a full-parsing model, and (3) a dual model. A ﬁJIl-listing model

38

does not predict any role of WM because the computation is not the part of the model. In
contrast, a full-parsing account and a dual model predict that WM capacity may play a
role in complex nominal processing because both models incorporate semantic
computations into their models.

The ﬁnal relevant point is that the effect of WM is generally observed with high-order
cognitive activities. Put differently, the complexity of linguistic targets interacts with
WM capacity and thus the effect of WM can be more prominent when individuals
encounter complex items (Conway, Kane, Bunting, Hambrick, Wilhelm, & Engle, 2005).
Taking an example from language processing, the WM effect has been found in ﬁller-gap
dependencies where long-distance integration is required (e.g., Fiebach, Schlesewsky,
and Friederici, 2001). King and Just (1991) found that WM capacity interacted with the
complexity of linguistic targets. They compared the processing of subject relative
clauses and object relative clauses; in general, object relatives are more complex than
subject relatives so that object relatives requires more processing costs (see Just &
Carpenter 1992 for a review). In a self-paced reading task, King and Just asked
participants to remember one or two unrelated words while reading sentence; then, the
participants were asked to judge the plausibility of the sentences. They found that the low
span readers are signiﬁcantly less accurate in comprehending object relatives than subject
relatives, suggesting that WM plays a greater role in processing a processing costlier
complex sentence such as object relatives than a sentence requiring less processing costs
such as subject relatives. When relating the complexity requirement for observing the
role WM capacity to complex nominal processing, there appears to be, to the best of my

knowledge, no Ll empirical research on the relationship between WM capacity and

39

complex nominal processing. The absence of L1 studies in this area is not surprising
because complex nominal processing by native speakers is a speedy process that may not
be observed in a normal environment.

However, L2 research could provide fruitful information in this area. A general
consensus of L2 processing is that L2 learners have less efﬁcient processing mechanisms
than L1 speakers. Therefore, the delayed processing of complex nominal (i.e., slow
semantic integrations in this study) would be related to L2 learners’ WM capacity. In
addition, the complexity requirement of WM allows us to test the effect of WM capacity
with L2 learners whose L1 morphological structures differ (right-headedness vs. leﬁ-
headedness in this study). For example, L2 learners whose Ll has left-headed structures
may reveal more interference in processing L2 complex nominals which are right-headed
than those who have right-headed compound structures in their Ll. In this context, the
effect of WM is likely to emerge in L2 learners whose Ll morphological structure differs

from L2 equivalents.

2.4.1.2.2. Inhibitory Control Ability

Inhibitory control refers to mental processes that allow one to ignore or suppress
irrelevant information when competition information is present. A representative measure
of inhibitory control ability is the Stroop task where, for example, when a color word
such as RED is visually presented in blue, one is asked to name the ink color but not the
word written in. It is important to note that individuals vary in inhibitory control ability

(Green, 1998). For example, when linguistic information competes, those who are good

40

at inhibiting irrelevant linguistic information may perform better than those who are not
(e.g., Kane & Engle, 2003; Kiefer, Ahlegian, & Spitzer, 2005).

Aphasia research provides a clue to the relevance of inhibitory control ability in
compound processing. Libben (1998) reanalyzed the impairment in the comprehension of
compounds by an aphasic patient, RH as a case of inhibition failure. When asked to
choose which one does not belong from the list of TULIP, CARNATION, ROSE, and
OAK, the patient RH chose CARNATION rather than the correct answer OAK. When
asked the reason, she indicated that she chose it because CARNATION is a brand name
for evaporated milk. Libben claimed that RH was unable to inhibit or ignore the
activation of the subordinate (evaporated milk) meaning, even when CARNATION was
presented with highly semantically related words such as TUILP and ROSE. That is,
according to Libben, all the meanings for CARNATION were activated, due to the
failure of RH’s inhibitory control ability. Another example reported in Libben (1998) is
that, when RS asked what dumbbell means, her response was “stupid weights. . .Amold”.
Libben (1998) interpreted her response as the meaning activations of both constituents
and the whole compound; the word stupid seems to be produced in conjunction with a
constituent dumb while such words as weights and Arnold (Arnold Schwarzenegger)
appear to be made through the meaning activations of the whole word (i.e., a type of
exercise weight). This, in turn, suggested that the interpretation and production of
complex nominals can be gained by inhibiting irrelevant competing information,
particularly during the process of semantic integration.

An expectation for inhibitory control ability in complex nominal processing is

that the inhibitory control ability may function similar to the WM capacity because both

41

cognitive abilities are a type of executive functioning that controls and manages cognitive

processes (e.g., Green, 1998; Norman & Shallice, 1986). As discussed earlier,

compositional processes are a part of processing complex nominals particularly in the

sense that two nouns in a complex nominal must ﬁnd a relational structure that leads to a

meaningful relationship between the two nouns. For example, the meaning relationship of

coﬂee cup is FOR (i.e., a cup for coffee) but not is IN, BE, CAUSE or MADE OF. All

relational structures such as FOR, IN, and BE could be activated in recognizing a

complex nominal because, in theory, each constituent can have innumerable

combinations. For example, coﬂ'ee could be combined with diverse nouns.

Table 5. The top 20 high-frequency complex nominals with coffee as a modiﬁer

 

 

Rank Modiﬁer Head Frequency Rank Modiﬁer Head Frequency
1 COFFEE TABLE 1631 l l COFFEE MACHINE l 13
2 COFFEE SHOP 1 I69 1 2 COFFEE BAR 99
3 COFFEE CUP 599 13 COFFEE GROUNDS 99
4 COFFEE BEANS 223 14 COFFEE MAKER 99
5 COFFEE POT 223 15 COFFEE GRINDER 65
6 COFFEE MUG 2 l 7 16 COFFEE PLANTATIONS 59
7 COFFEE CUPS 195 17 COFFEE FARMERS 53
8 COFFEE HOUSE 16] 18 COFFEE CANS 49
9 COFFEE BREAK l 23 19 COFFEE FILTER 49
10 COFFEE CAKE 1 1 7 20 COFFEE ROOM 49

 

42

Table 5 illustrates the diverse combinations when coffee is used as a modiﬁer in a
complex nominal; the items in Table 5 is the most frequent top 20 items in the corpus of
contemporary American English (Davies, 2009; see Task 1 for the detailed description of
the corpus).

Given that all the possible relational structures are activated, the recognition of
complex nominals requires the inhibitions of irrelevant relational structures among many
possible relations. In this regard, one’s cognitive ability to inhibit irrelevant information
is likely to be related to how one processes complex nominals, when the use of
computational routes is assumed to be used in complex nominal processing.

Let me describe the relevance of inhibition to the process of combing two
constituents into a single representation or the process of relational selection in complex
nominal processing, from the perspective of the CARIN (Gagné & Shoben, 1997, 2002)
theory. As mentioned earlier, the CARIN emphasized that the availability of relations
increases the interpretability of complex nominals. In both unambiguous and ambiguous
complex nominals, for example, the CARIN model suggests that complex nominal
processing involves competition among the possible relations. Yet, an alternative using
an inhibition mechanism should also work. It may not be sufﬁcient enough to explain the
use of relations in complex nominal processing in terms of the availability of as many as
possible relations and competitions among relations. It is highly plausible that higher
availability of a relation will increase the range of selection and thus result in reduced
processing time. However, in order to end up with a single interpretation, an individual
should ignore or inhibit the irrelevant relation between modiﬁers and heads. Then, it is

plausible that an individual who is good at inhibiting irrelevant relations associated with

43

complex nominals would show better performance than a counterpart who has low
inhibitory control ability. Figure 7 illustrates the relation selection process in terms of the
CARIN theory; the dotted lines represent the inhibition of irrelevant relational structures
in processing coffee cup while the two solid lines show the selections of an appropriate

relational structure.

Figure 7. The potential role of inhibitory control ability on complex nominal processing
with the framework of CARAN

 

 

 

coffee cup

 

 

 

 

 

 

As Figure 7 illustrates, the irrelevant activated relational information should be
eliminated during semantic composition. Therefore, examining the processing of
complex nominals may provide a good linguistic venue for determining the role of
inhibitory control in second language acquisition.

Importantly, given the “invisibility” of the process in a normal circumstance
(Gagne & Spalding, 2006b), I may not ﬁnd that native speakers show the effect of
inhibitory control ability on the recognition of complex nominals. In this regard,
interlanguage data could provide another piece of evidence of the relevance of inhibitory
control ability to the recognition of complex nominals, as suggested by the

aforementioned aphasia studies.

44

2.4.2. An item variable: Frequency

Examining the frequency effect of complex nominal processing allows us to
evaluate the three models of complex nominal processing discussed in 2.2.1.: (1) full-
listing accounts, (2) full-parsing accounts, and (3) dual-routes accounts. A direct and
widely accepted method to test the presence of retrieval routes is to examine the
frequency of the whole words because empirical supports for full-listing models are
grounded on frequency of the whole compounds (see Bybee, 1995). The basic idea of
examining the frequency effect is that the more frequently complex nominals are used,
the more likely they are to become lexicalized or to be stored as a whole. Therefore,
looking at the role of frequency could be a good testing basis for the use of retrieval
routes in the interpretation processes of complex nominals.

At the same time, in the L1 literature, the validity of computational routes in
complex nominal processing has often been tested by examining the manipulated
constituent ﬁequency while keeping invariant the frequency of whole complex nominals
(Andrews, Miller, & Rayner, 2004; Hyonéi & Pollatsek, 1998; Juhasz, Starr, Inhoff, &
Placke, 2003; Lima & Pollatesk, 1983; Shoolman & Andrews, 2003; Taft, 1979; Taft &
Foster, 1976). Overall, empirical evidence revealed a facilitative effect of high-frequency
constituents on the recognition of complex nominals, although there are mixed results for
whether either the ﬁrst or the second constituent or both has a greater effect. For example,
farmhouse is easier to recognize than graveyard because both farm and house are high-
frequency words whereas grave and yard are low-frequency words (Juhasz, Starr, Inhoff,

& Placke, 2003).

45

Note that L1 research in support of full-parsing models or dual-route models has
focused more on the frequency of each constituent (e.g., coﬂee and cup) in a familiar or
novel complex nominal not only because the frequency effect of the whole familiar
complex nominal (i.e., coffee cup) is always expected but also because, understandably,
the frequency of novel complex nominals cannot be obtained. Thus, it may be less
interesting or may not even be falsiﬁable that the frequency effect of the whole complex
nominal emerges in native speakers. In contrast, interlanguage data allow us to examine
the frequency effect of familiar complex nominals (i.e., coffee cup) because L2 learners
would be less likely to store a familiar complex as a whole. Thus, it is a falsiﬁable
hypothesis that there is a frequency effect of the whole complex nominal in non-native
speakers.

Another difference between NS and NNS is that, in the case of NS, the effect of
the whole-word frequency does not necessarily mean the presence of retrieval routes (see
Gagne and Spalding, 2006b). The effect of the whole-word frequency in native speakers
may be due to faster processing or efﬁcient computations of complex nominals,
indicating that the whole-word frequency effect could be inconclusive for the presence of
retrieval routes. In contrast, the whole-word frequency effect by NNS could be a piece of
supporting evidence for the retrieval routes, given that NNS may have less efﬁcient
processing systems in comparison to NS.

In general, with respect to the three models of compound processing, the
examination of the frequency of whole complex nominals leads to three expectations. A

full-listing account predicts a strong association of the frequencies of complex nominals

46

and participants’ performance. A full-parsing account expects no strong association while

a dual-account predicts a moderate correlation.

2.5. General objectives

This dissertation investigates the regularity and the systematic patterns of second
language acquisition of complex nominals, and explores the source of individual
variation in second language acquisition. There are three major objectives in this
dissertation as listed below. Speciﬁc hypotheses of each objective will be put forward in

task 1, 2, and 3.

Objective 1. Diﬂerence between NS and NNS in terms of dual functions
Do both NS and NNS groups use retrieval and computational routes in

recognizing complex nominals?

Objective 2. LI inﬂuence on L2 complex nominal processing
Does Ll morphological knowledge (i.e., headedness) affect the processing

of L2 complex nominals?

Objective 3. The eﬂect of cognitive abilities on the recognition of complex nominals
Are L1 speakers and L2 learners with higher cognitive abilities (such as
better inhibitory control ability and higher working memory capacity)
more efﬁcient in processing complex nominals that those with lower

capacity?

47

 

CHAPTER THREE - RESEARCH DESIGN AND METHODOLOGY

3.1. Overview of the study

There were three computer-based tasks in this study: (1) a lexical sense decision
task, (2) a working memory test, and (3) an inhibition test. Participants took part in all
three tasks and did the three tasks in the order of (1) a lexical sense decision task, (2) a
working memory test, and (3) an inhibition test. The ﬁrst task was designed to examine
whether native speakers and non-native speakers use either computational routes or
retrieval routes or both when they comprehend complex nominals. For this, I created an
interference design for complex nominal processing, which has not been tested in either
the L1 or the L2 literature. Speciﬁcally, four experimental conditions were created in
terms of the number of meanings of each constituent in a complex nominal (i.e., single-
single such as coﬁiee cup, single-multiple such as violin bow, multiple-single such as well
water, and multiple-multiple such as lead mine). With an assumption that semantic
integrations are involved in recognizing complex nominals, the second and the third task
examined the role of cognitive capacity, which may be related to the performance of L2
learners. The second task was intended to investigate whether working memory (WM)
capacity is related to the interpretations of complex nominals. The third task was aimed

to examine the effect of inhibitory control ability on the recognition of complex nominals.

3.2. Task 1: Complex nominal processing in an interference paradigm
The goal of Task 1 was two-fold. First, I investigated the dual functions of the

second language lexicon through the recognition of complex nominals. Speciﬁcally, Task

48

 

1 tested whether some complex nominals involve either semantic computations or
retrieval mechanisms from long-term memory or both. Secondly, I examined whether the
recognition of L2 complex nominals is inﬂuenced by the headedness of compound
structure in the ﬁrst language.

For this, as explained in Chapter 2, the linguistic target was familiar and
semantically transparent complex nominals such as coffee cup. The term “familiar” is
operationalized as the existence of compounds in the COCA (Corpus of Contemporary
American English, Davies, 2009) corpus, a large English corpus containing more than i
400 million spoken and written data. The reason I determined to use “familiar” complex
nominals stems from empirical ﬁndings of L1 dual-route models suggesting that familiar
compounds are accessed via whole-word access while novel compounds are interpreted
via computational routes (e.g., Baayen, Dijkstra, & Schreuder, 1997). The dual route
models suggest that L2 learners are highly likely to use computational routes in
processing “novel” compounds, given that L1 speakers rely on the computational
mechanism when they process novel compounds. A more empirically interesting question
is whether L2 learners are able to use retrieval routes in processing “familiar” complex
nominals because L2 learners may not store as many whole-word representations of
complex nominals as NSs do. Put differently, in order to compensate for the shortage, L2
learners may rely more on computational routes.

Using the familiar and semantically transparent complex nominals, I propose a
new interference paradigm for the recognition of complex nominals in conjunction with a
robust psycholinguistic ﬁnding of multiple meaning activations. Speciﬁcally, I integrate

the two processes of semantic integration of complex nominals (i.e., the meaning

49

activations of individual constituents and the combinatorial processes of establishing
meaningful relationships between the individual constituents) with meaning activations
of words. This experimental paradigm, to the best of my knowledge, has not been
explored in either the L1 or the L2 literature.

A word can have many meanings. For example, an English homonym or
homograph bank has multiple entries in a dictionary (e.g., bank of a river versus bank
where you deposit money). Importantly, much psycholinguistic research has illustrated

that all the meanings of a word seem to be initially activated when people encounter the

’ ‘I"aﬂn’.~—f§‘j

word containing multiple meanings, typically through priming studies both with
decontextualized items (e.g., de Groot & Keijzer, 2000; Jared & Kroll, 2001; Marian,
Spivey, & Hirsch, 2003) and with contextualized items (Schwartz & Kroll, 2006;
Swinney, 1979). Frenck-Mestre and Prince (1997) found that L2 learners also showed
multiple activations of meanings when processing L2 homonyms, with an assumption
that L2 learners know all the meanings.

We know that complex nominals consist of two constituents (i.e., a modiﬁer and a
head). Table 6 illustrates four logically possible combinations, depending on whether or
not either a modiﬁer (the ﬁrst constituent in English) or a head (the second constituent in
English) or both has multiple meanings: (1) SS: single meaning - single meaning (e.g.,
coffee cup), (2) SM: single meaning - multiple meanings (e.g., violin bow), (3) MS:
multiple meanings — single meaning (e.g., well water), and (4) MM: multiple meanings —
multiple meanings (e.g., lead mine). I operationalize a word having multiple meanings as

a word containing at least two entries in a dictionary.

50

Table 6. The four experimental conditions in Task 1

 

 

Condition Modiﬁer Head Example
SS single-meaning single—meaning Orange juice
SM single-meaning multiple-meanings Violin bow
MS multiple—meanings multiple-meanings Well water

MM multiple-meanings multiple-meanings Lead mine

 

nu
I

Assuming computational routes are utilized in interpreting a complex nominal,
the four combinations provide us a good venue to examine the computational routes,
which may be a function of the L2 lexicon. In particular, when assuming that the two
processes of semantic integration are obligatory for complex nominal processing, the four
conditions imply differences in processing costs during the processing of complex
nominals. First, the four conditions differ with respect to the meaning activations of
individual constituents. The SS (single-single) condition implies the activations of single
meanings of each constituent whereas the MM (multiple-multiple) condition entails the
activations of multiple meanings of each constituent. SM (single-multiple) involves the
activation of a single meaning of the ﬁrst constituent and the activation of multiple
meanings of the second constituent. In the case of the MS (multiple-single) condition,
there would be the activations of multiple meanings of a modiﬁer and the activation of a
single meaning of a head.

Secondly, we know that the interpretations of complex nominals are not a simple
union of two meanings of each constituent. Semantic integration processes of complex

nominals involve the combinations of the meanings of each constituent into the whole

51

meaning of a complex nominal. In other words, one should look for an interpretable
relationship between two nouns. Also, we know that a word could be used in various
semantic relationships. For example, cup is not just for coffee cup, but there are numerous
contexts in which cup is used (e.g., paper cup, gold cup, etc). Within the contexts of the
four experimental conditions, a constituent having multiple meanings provides more
possible relationships between two nouns than a constituent having a single meaning. The
activations of multiple meanings in the SM, MS, and MM conditions imply the increase
in possible relationships between two nouns during the recognition of a complex nominal.
With respect to processing cost, processing of the MM condition would be costlier than
processing of the SM and MS conditions, which involve more processing costs than
processing of the SS condition. Below are graphical representations of the four

experimental conditions regarding the activations of meanings and relational structures.

Figure 8. The processing of a complex nominal in the SS condition

® I IN o
G... ~‘I‘.
Q2... ........... .4 FOR ~unuoo¢u::;:£?
.'. IIO'..-' '.‘.‘
ﬂ ._ o o o 5'...

coffee cup

 

 

 

 

 

 

 

 

Figure 9. The processing of a complex nominal in the SM condition

 

 

 

 

 

 

violin

 

 

 

52

Figure 10. The processing of a complex nominal in the MS condition

 

 

 

. IN @
water

 

 

 

 

 

 

Figure 1 l. The processing of a complex nominal in the MM condition

 

 

 

 

 

 

 

 

 

Figure 8, 9, 10, and 11 represent not only the meaning activations of each constituent in a
complex nominal but also the processes of relation selection among the activated
meanings. The rectangles show each constituent in a complex nominal, the ellipticals
illustrate the activated meanings, and arrows demonstrate the meaning activations. The
solid and dotted lines show the relation selection process and the capitalized bold words
such as [N and MAKE mean the possible activations of relation structures among the
activated meanings. As seen in the four Figures, the more the meanings of each
constituent are activated, the more relation selections are available. To illustrate, Levi

(1979) proposed that modiﬁers and heads in English compounds can be characterized as

53

12 meaning relationships. Let us assume that all of the 12 relations are activated in
interpreting a complex nominal. Then, each constituent could have 12 relationships in
semantic integration processes of complex nominals. For example, the interpretations of
items in the SS condition could be selected via 12 x 12 relational possibilities, given that
the items in the SS condition have a single meaning for both a modiﬁer and a head. In the
case of the MM condition, the interpretations of items may be 12 x 12 x 12 x 12 when
assuming each activated meaning participates in establishing the relation between two
nouns. In the same vein, the items in the SM and MS conditions would involve 12 x 12 x
12 activations of relational structure.

In the above processes, an assumption that I made is that all the possible relations
are activated. However, it is an empirical question whether all the relations are activated
in processing complex nominals. Some nouns, for example, tend to be used in a “typical”
relation that might be indicated by its frequency. Particularly, the activations of relational
structures could be guided by frequency. Examining the COCA corpus, a noun juice
tends to be frequently used for MAKE 2 (e.g., orange juice ‘juice made of orange’)
according to Levi (l 978)’s taxonomy. Table 7 contains the top 10 uses of juice as a head
in a complex nominal.

Nonetheless, my rationale for the four experimental conditions is that multiple
meaning activations increase the numbers of relational structures. Put differently,
probability would be a good term to describe the possibility of activations of relational
structures. Complex nominals containing a multiple meaning word are more probable to
get involved in the more complex selection processes of relational structures. Indeed, it

may not be possible to deﬁne how many relational structures are activated in processing a

54

complex nominal, as indicated by linguists’ divergence on the number of relations
between two nouns in a complex nominal (e.g., Gleitman & Gleitman, 1970; Kay &

Zimmer, 1976; Levi, 1978).

Table 7. The most frequent uses of juice as a head noun

 

 

Items The whole-word Items The whole-word
frequency frequency
Lemon juice 3006 Grape juice 245
Orange juice 1970 Cranben'y juice 21 3
Lime juice 1202 Tomato juice 174
Apple juice 409 Pineapple juice 125
Fruit juice 265 Grapefruit juice 106

 

The other assumption 1 made is that the computation is a part of complex nominal
processing. The activations of relational structures are not expected if retrieval routes are
fully responsible for recognizing complex nominals. With the assumption of the presence
of computational routes in complex nominal processing, the differences in the meaning
activations of each constituent and the activations of relation structures entail differences
in processing costs across the four experimental conditions. Given the aforementioned
numerical approximations of each condition, the SS condition involves the lowest
processing costs, the MM condition would have the highest processing costs, and the SM
and MS conditions would be in the middle.

Furthermore, in Chapter 2, we saw that the headedness of complex nominals
differs across languages, which may affect L2 processing of complex nominals. For
example, if Ll morphological knowledge plays a role in recognizing L2 complex
nominals, the LH group may show more interference than the RH group in processing

English complex nominals because the complex nominal structure of the LH group is

55

different from that of English whereas the complex nominal structure of the RH group is
identical to that of English. With respect to the four conditions, I predict that the
behavioral differences between LH and RH groups will be greater as the processing costs
increases. Yet, L2 proﬁciency needs to be taken into consideration. Particularly, when
both the LH and the RH groups have the high level of L2 proﬁciency, the effect of L1
headedness on processing L2 complex nominals may not emerge in the sense that L2
learners become more independent of L1 as L2 proﬁciency increases (e.g., Frenck-Mestre
& Prince, 1997; Kroll & Curley, 1988).

Taking all the consideration together, the following four research questions

emerge.

The research questions of Task 1
RQ 1. Difference between NSs and NNSs
Do NSs show signiﬁcantly better performance in all four conditions than the two

NNS groups of high L2 proﬁciency?

Hypothesis 1:
The NS group performs better than the NNS groups, given that the NS
group has an efﬁcient processing ability as well as having more stored

information.

I expect that the greatest difference between the NS group and the two NNS groups will

be in the MM condition, which involves the greatest processing costs. However, in the SS

56

condition where processing costs are low, I expect that the NS and NNS groups will not

differ signiﬁcantly.

RQ 2. LI inﬂuence of morphological knowledge in processing L2 complex nominals
Does the headedness of the L1 compound structure affect the processing of L2

complex nominals?

Hypothesis 2:
The difference in the headedness of the L1 compound structure affects L2

processing of complex nominals.

Figure 12. A prediction for the difference between the RH and the LH group in terms of
reaction times

 

 

 

 

 

 

 

 

 

 

SS SM MS MM

 

 

 

A general expectation is that, if L1 headedness plays a role in recognizing L2 complex
nominals, the LH group may show more interference than the RH group in processing
English complex nominals. More speciﬁcally, I expect that the difference between the

two groups would be greatest in the MM condition which requires the greatest

57

computational costs whereas the difference would be smallest in the SS condition where
computational costs would be small. This prediction is illustrated in Figure 12.

Contrary to Hypothesis 2, Hypothesis 3 takes L2 proﬁciency into account. That is,
the effect of L1 headedness on processing L2 complex nominals is not salient when two
NNS groups having different headedness in their L1 morphology are highly proﬁcient

learners of the L2.

Hypothesis 3:
When comparing two NNS groups of high, comparable proﬁciency, the
L1 effect, operationalized by the headedness of the L1, will be not

signiﬁcant between the two groups.

RQ3. Uniform interference in the NS group versus graded interference in the NNS groups
Do the NS group and the NNS groups differ in recognizing complex nominals in

the four experimental conditions?

I predict that the NS and NNS groups will behave differently in the four
conditions, particularly when the test items in the four conditions are familiar complex
nominals. Familiarity is operationalized as existence in a large English corpus. This study
uses complex nominals which actually occur in English, which enables us to make an

interesting prediction as to different behavioral patterns between the NS and NNS groups.

Hypothesis 4 (NS):

58

Since the stimuli of this study are familiar complex nominals (which are
found in a large English corpus), the NS group may rely more on the
retrieval mechanism (or compute complex nominals efﬁciently). Thus, I
expect that the NS group may not show a substantial difference in reaction

times and accuracy across the four conditions (i.e., MM =MS = SM :88).

Figure 13. A prediction for the NS group: Uniform interferences

 

 

 

 

-0- Reaction time/ Accuracy

 

 

SS SM MS MM

 

 

 

Although it is a bit controversial, it is generally accepted in the L1 literature that familiar
complex nominals are recognized via retrieval routes whereas unfamiliar counterparts are
interpreted via computational routes (e.g., Baayen, Dijkstra, & Schreuder, 1997; Bertram,
Schreuder, & Baayen, 2000; Caramazza, Laudanna, & Romani, 1988; Pinker & Ullman,
2002). Given that familiar words are highly likely to be stored in the L1 lexicon and/or to
be computed efficiently, 3 NS group may not show processing differences across the four
conditions. In other words, a NS group would reveal a uniform interference across the

four conditions which contain familiar complex nominals. Figure 13 demonstrates a

59

hypothetical case of uniform processing costs by a NS group in the four experimental

conditions.

Hypothesis 5 (NNS):

Since the stimuli of this study may function as novel complex nominals,
the NNS groups may depend more on a computational route to interpret
complex nominals. Thus, I expect that the NNS group may show a
substantial behavioral difference among the four conditions. Particularly,
given the manipulated difference of the four conditions in terms of
meaning and relation activations, I expect that reliance on the
computational route will yield the graded interference in accordance with
the processing costs of each condition: the greatest interference in the MM
condition, intermediate interference in the MS and the SM conditions, and

the smallest inference in the SS condition (i.e., MM > MS, SM > SS).

With an assumption that the recognition of complex nominals involves computational
processes, a NNS group may show graded processing difﬁculties in accordance with the
number of meanings of an individual word embedded either in modiﬁers or in heads. In
contrast to a NS group, a NNS group would demonstrate a different amount of processing
costs in recognizing the familiar complex nominals of the four conditions. Speciﬁcally, I
expect that unambiguous-unambiguous items involve the smallest processing costs
among the four conditions while ambiguous-ambiguous items are associated with the

greatest processing costs. Also, complex nominals containing an ambiguous item (i.e.,

6O

MS or SM) would involve higher processing costs than unambiguous-unambiguous items,
but may be related to less processing costs than ambiguous-ambiguous items. Figure 14

shows this hypothetical case of differences in processing costs.

Figure 14. A prediction for a NNS group: Graded interfere

/ +Reaction time

SS SM MS MM

 

 

 

 

 

 

 

 

 

 

RQ 4. Comparing NSs and NNSs in terms of whole-word representations
Do the frequencies of complex nominals as a whole play a signiﬁcant role in

recognizing complex nominals for both NS and NNS groups?

A canonical way to address the whole-word representations of a phrase is to
examine the relationship between the frequencies of the whole phrases and the various
measures related to the phrases such as response times. In this regard, the frequency of
the whole complex nominals was examined to test the possibility of holistic

representations of the complex nominals.

Hypothesis 6:

61

Given the robust effect of frequency in many Ll psycholinguistic studies
as well as the familiarity of the stimuli in this study, a frequency effect
will emerge for the NS group, particularly when the NS group processes
complex nominals with high frequencies in a corpus. However, L2
learners are expected to show weaker or no frequency effects on
recognizing L2 complex nominals because NNS groups may have a

relatively smaller number of stored complex nominals.

3.3. Task 2: Complex nominal processing and working memory capacity

Task 2 examines the potential role of a cognitive variable, working memory (WM)
capacity, on L2 complex nominal processing. Speciﬁcally, Task 2 looks at the question of
whether L2 learners with higher WM capacity are more efﬁcient in processing a second
language than their counterparts with lower WM capacity, when comparing the
processing ability of NNS groups with comparable proﬁciency.

We have seen that the semantic integration of complex nominals involves many
competitions, which are caused by (l) the meaning activations of each constituent and (2)
a semantic integration process to combine two constituents in a meaningful relationship.
Working memory (WM) capacity may play a role in this semantic integration process. As
discussed in Chapter 2, there are many WM models including WM capacity as either
inhibitory control/controlled attention or WM capacity as activation, yet these models are
not mutually exclusive. Speaking of the meaning activations of each constituent, the

interpretation of lead mine requires the suppression of irrelevant meanings of each

62

constituent such as “guide” and “pronoun” or the controlled attention to relevant
meanings such as “metal” and “place.” Similarly, the integrations of two nouns require
the suppression of irrelevant relational structures such as IN and CAUSE or the
maintenance of a relevant relational structure such as HAVE. Considering the two
possible roles of WM on the recognition of complex nominals, therefore, an individual
having high WM capacity may be more efﬁcient in interpreting lead mine than an
individual with low WM capacity.

Figure 15 illustrates the functions of the WM capacity in the recognition of a MM
(multiple meanings — multiple meanings) complex nominal such as lead mine. In Figure
18, the two solid ellipticals illustrate the maintenance of relevant meanings whereas the
two dotted counterparts show the suppression of irrelevant meanings during the
interpretive process of lead mine. Also, the two solid lines delineate the selection process
of an appropriate relation of lead mine while the other dotted lines demonstrate the

inhibition of irrelevant relations.

Figure 15. An illustration of the potential functions of the WM capacity in the recognition
of a MM complex nominal

-
n
h...

#3:: pronoun _3

 

 

 

 

mine

 

 

 

 

 

 

63

However, I do not expect that WM capacity exerts a role in processing all the
complex nominals. Rather, the effect of WM would emerge with cognitively demanding
tasks. An important assumption is that semantic integrations are a part of processing
complex nominals; the effect of WM capacity is less likely to be found if the processing
of complex nominals relies on retrieval routes, which is cognitively less demanding. That
is, WM capacity, as described in Chapter 2, tends to be observed when an individual does
cognitively demanding tasks. Within the current context, processing of MM complex
nominals such as lead mine would be more cognitively demanding or involve greater
processing costs than that of SS complex nominals such as coﬂee cup. Thus, the effects
of WM are less likely to appear with SS complex nominals that require less processing
costs than with MM equivalents.

The WM requirements for cognitively demanding tasks are related to the potential
differences between NS and NNS groups in response times and accuracy. In the L1
literature, the role of WM capacity on complex nominal processing is rarely investigated.
There might be two possible reasons in relation to the dual functions of the lexicon. First,
if complex nominals are dependent upon a simple retrieval system, it is difﬁcult to ﬁnd a
WM effect, which is more likely to be found in higher cognitive processes. Alternatively,
although native speakers derive the meanings of complex nominals by computational
routes (but not by a retrieval mechanism), the highly efﬁcient morphological processing
of native speakers does not allow us to observe the WM effect in natural circumstances.
In other words, processing familiar complex nominals may not be cognitively demanding

for the NS group.

64

In contrast, NNS groups lack two advantages that native speakers may possess
because NNSs may have (1) a reduced number of stored complex nominals or (2) an
inefﬁcient or less automatized processing system. The disadvantage of non-native
speakers may make the processing of complex nominals a cognitively demanding task,
which, in turn, allows us to observe the effect of WM. An expectation is that cognitive
ability is likely to compensate for higher processing loads of NNSs.

Furthermore, a different effect of WM would be expected between the two NNS
groups, which have different headedness in their L1. One might expect that processing
English complex nominals (right-headed structure) would be more cognitively
demanding for the LH group than the RH group because the RH group has to switch the
order of modiﬁers and heads in processing English complex nominals if their L1
morphological structure plays a role. Indeed, one WM model sees WM as a task-
switching ability (Hegarty, Shah, & Miyake, 2000). Therefore, it may be the case that the
effect of WM capacity on the recognition of complex nominals may be greater in the LH
group than in RH group.

In sum, I expect that a WM effect on L2 complex nominal processing will be
compounded with L1 background and the complexity of the test items. The two NNS
groups in this dissertation have comparably high L2 proﬁciency; yet, the two groups
differed with respect to the headedness of the compound structure in their L 1 s. The WM
effect would be greater in the LH group whose L1 compound structure invokes more
interference in processing L2 complex nominals than the RH group. As discussed in
2.4.1.2.1., the greater effect of WM emerged with the more complex items. Thus, I expect

to ﬁnd the effect of WM only in the MM condition, but not in the SS condition. Taking

65

the above considerations into account, the following research question will be examined

in Task 2.

The research question of Task 2
RQ. The effect of WM capacity on complex nominal processing.
Are L1 speakers and L2 learners with higher WM capacity more efﬁcient in

processing complex nominals than those with lower capacity?

Hypothesis:

There might be no substantial effect of WM on complex nominal
processing, due to the NS group’s efﬁciency in morphological processing.
In contrast, I expect the effect of WM capacity on L2 complex nominal
processing. Particularly, given that the complexity of the target is related
to WM capacity, I expect that the LH group will show a signiﬁcant WM
effect on L2 complex nominal processing in the condition involving the

highest processing costs (i.e., the MM condition).

To recapitulate, WM capacity is likely to be observed in relation to the dual functions.
The reliance on the retrieval mechanism or the use of faster computations would result in
no effect of the WM capacity in recognizing complex nominals, as illustrated in Figure
16. Semantic computations may be related to the WM effect only when the computations
involve high processing costs. With respect to the reaction times of the participants, WM
capacity is expected to be inversely related to the semantic integration of the NNS groups

because those who have the higher WM capacity would be faster in responding to

66

complex nominals than those who have lower WM capacity. In contrast, the accuracy
scores may be positively related to the WM capacity, given that high WM individuals are
efﬁcient in processing information and have more storage of information. The idea is

shown in Figure 17.

Figure 16. A prediction of the WM effect in case of retrievals or efﬁcient computations

 

 

 

 

 

 

 

+RT/Accuracy

 

 

 

Low WM High WM

 

 

 

Figure 17. A prediction of the WM effect in case of the delayed computations

 

 

 

'0'.1 \ .*RT
...o"A \ 00... Accuracy

 

 

 

 

 

Low WM High WM

 

 

 

67

3.4. Task 3: Complex nominal processing and inhibitory control ability

Inhibitory control ability refers to resistance to interference or the ability to
suppress or inhibit irrelevant information. The Stroop task (Stroop, 1935) is known as the
representative measure of inhibitory control ability (MacLeod, 1991). In the Stroop test,
participants see color words (e.g., RED, BLUE, and GREEN), in which color could be
either matched (e.g., RED in red) or unmatched (e.g., RED in blue). The task is to name
the color of the font but not the name of the word; when seeing a word RED written in
BLUE, for example, participants have to name RED. In the Stroop task, participants are
in essence asked to inhibit irrelevant information (the color of the words) between two
pieces of competing information that are automatically accessed (i.e., the orthography of
the words and the color of the words). Importantly, individuals vary in this inhibitory
control ability.

Inhibitory control ability appears highly related to the performance in the
recognition of complex nominals. A piece of supporting evidence, as noted in 2.4. comes
from aphasia studies which show that aphasic patients failed to inhibit irrelevant
information in comprehending and producing complex nominals. Such examples as
carnation and dumbbell reported in (Libben, 1998) illustrate that an aphasic patient did
not inhibit irrelevant meanings of compounds while using both the meanings of the whole
word and the meaning of the constituents.

A general prediction of this study is that inhibitory control ability functions
similarly to working memory capacity, given that inhibitory control ability is a member
of executive functioning (Anderson & Green, 2001; Green, 1998; Norman & Shallice,

1986) and a type of WM is based on inhibition (e.g., Hasher, Lustig, & Zachs, 2007;

68

Gemsbacher & Faust, 1991, Gunter, Wagner, & F riederici, 2003). Also, previous
research found the interaction between working memory capacity and Stroop interference,
in terms of goal maintenance and inhibition of competing information (Chiappe &
Chiappe, 2007; Kane & Engle, 2003; Long & Prat, 2002). Thus, I expect that WM and
Stroop effects would provide a converging piece of evidence; those who may have better

inhibitory control ability would be more efﬁcient in recognizing complex nominals.

Figure 18. An illustration of the potential functions of inhibitory control ability in the
recognition of a MM complex nominal

 

 

 

 

 

 

 

 

 

Particularly, the semantic integration of complex nominals is likely to be related
to inhibitory control ability in the sense that the integration processes invoke
competitions. As described earlier, two competition resources are available in
recognizing complex nominals: (1) the meaning activations of each constituent and (2) a
semantic integration process to establish a meaningful relation between two nouns. The
interpretation of lead mine, for example, requires the suppression of irrelevant meanings
of each constituent such as “guide” and “pronoun” among the activated information, as
illustrated in Figure 18. Also, the integrations of two nouns require the suppression of

irrelevant relational structures such as IN and CAUSE. Therefore, an individual having

69

high inhibitory control ability may be more efﬁcient in interpreting lead mine than an
individual with low inhibitory control ability.

Some related predictions are borne out regarding the effect of inhibitory control
ability on complex nominal processing. With respect to the dual functions of the lexicon,
the effect of inhibitory control ability on the recognition of complex nominals could be
observed when the processing is in a more controlled fashion but not in an automatic
manner. It is not surprising there is no L1 literature that looks at the role of inhibitory
control processes on complex nominal processing. Retrieval routes or efﬁcient
computations would lead to difﬁculty in observing the functions of inhibitory control
ability. Thus, I predict that inhibitory effects could be found not in a NS group but in a
NNS group, given that a NNS group is generally assumed to have less efﬁcient
processing systems in comparison to a NS group. Secondly, the effect of inhibitory
control ability would be more salient when processing loads are costly. In the current
study, for example, I expect to ﬁnd the effect of inhibitory control ability on the MM
condition, but not in the SS condition because the MM condition would trigger the
greatest interference. Thirdly, the effect of inhibitory control ability may be related to L1
morphological knowledge, given that L1 knowledge may interfere in L2 processing. In
this study, for example, the LH group that has different compound structures from
English would show more interference in processing complex nominals than the RH
group. Put differently, the processing loads of complex nominals would be more costly
for the LH group than the RH group. Finally, L2 inhibitory control ability is related to L2
proﬁciency. A general expectation is that lower proﬁciency L2 learners do not show L2

interference. That is, proﬁcient L2 learners are likely to have more interference with an

70

L2 task than L2 learners having lower L2 proﬁciency. For example, Preston and Lambert
(1969) found that level of proﬁciency in a second language inﬂuences color-naming
interference. Zied et a1. (2004) found that unbalanced bilinguals revealed declines in
inhibitory mechanisms in their non-dominant language whereas balanced bilinguals did
not show comparable declines, suggesting an important function of language proﬁciency
to inhibitory control. A recent study by Gass and Lee (in press) did not ﬁnd a signiﬁcant
L2 Stroop effect with low and intermediate levels of L2 Spanish, leading to a call to look
at higher proﬁciency learners. Therefore, this study used learners with higher L2
proﬁciency, expecting that the Stroop effect would emerge in complex nominal
processing (see Discussion section for the validity of the L2 inhibition in this study).
Overall, the expectation in this study is that high proﬁciency L2 learners who have high
L2 inhibitory ability process complex nominals more efﬁciently than those who have low

L2 inhibitory control.

The research question of Task 3
RQ. The eﬂect of inhibitory control ability on complex nominal processing.
Are L1 speakers and L2 learners with higher inhibitory control capacity more

efﬁcient in processing complex nominals than those with lower capacity?

Hypothesis:
As for NSs, the familiarity and high frequency of the stimuli would not
induce the effect of inhibitory control abilities on complex nominal

processing. Although the Stroop effect would be minimal for the NS group,

71

I expect to ﬁnd that inhibitory control ability plays a role in L2 complex
nominal processing. I expect that the effect will be greater in a condition
involving the highest processing costs (i.e., the MM condition) and in the

LH group whose L1 ’8 compound structures differ from the target language.

To recapitulate, inhibitory control ability is expected to function similar to WM capacity,
with reference to the dual functions. No effect of the inhibitory control ability is predicted
when either the retrieval mechanism is mostly used or computations are efﬁcient in
recognizing complex nominals, as illustrated in Figure 19. In contrast, the effect of
inhibitory control ability is expected when the computations involve high processing
loads or are less efﬁcient. Figure 20 shows that, with an assumption of the presence of
semantic computations in complex nominal processing, those who have greater
interference (i.e., lower inhibitory control ability) may be slower and less accurate in
recognizing complex nominals than those who have smaller interference (i.e. higher

inhibitory control ability).

Figure 19. A prediction of the inhibitory control ability in case of retrievals or efﬁcient
computations

_._Wm.“ “M..- ..._ ._ .... ..........x.. ..._-._.._-... -.. .. . . ”42.4,..L -3 ..--__._._.... v- -- -.. ... . - __ .

 

 

 

 

 

 

-0-RT/Accuracy

 

 

 

Low WM High WM

 

 

 

72

Figure 20. A prediction of the inhibitory control ability in case of the delayed
computations

 

 

/.

/

/ -0-RT/Accuracy

 

 

 

 

Smaller interference Greater interference

 

 

 

3.5. Participants

Participants in this study were 76 undergraduate and graduate students at
Michigan State University. There were 20 native speakers (N 55) of English and 56 non-
native speakers of English. The non-native speakers of English were divided into two
groups depending on the compound structure of their ﬁrst language. The ﬁrst non-native
speaker group (hereaﬁer RI-I group) consisted of 17 Korean learners of English and 10
Chinese learners of English. Both Korean and Chinese have right-headed compound
structure (i.e., the order of modiﬁer-head) just as English does. The second non-native

speaker group (hereaﬁer LH group) included 19 Thai speakers of English and 8 Spanish
speakers of English“). These languages have left—headed compound structure (i.e., the

order of head-modiﬁer)l 1. All students received 20 US dollars for their participation. The

 

10 The initial sampling of the LH group included a French speaker of English and a Vietnamese
speaker of English. Although French and Vietnamese also have left-headed compound structure, one
sample from each language led me to the uncertainty in generalizing their patterns of the performance. I

decided not to include the two participants in the data analysis.

73

number of men and women and the average age of all participants are provided in Table 8.

The age of the participants was not signiﬁcantly different, F (2, 71) = 2.09, p= .13.

Table 8. Gender and age information of participants

 

 

 

NS (n= 20) RH (n= 27) LH (n= 27)
Gender
Male 7 5 5
Female 13 22 22
Average Age
25.45 29.33 28.63

 

The target population of L2 learners in this study was high proﬁcient learners of
English due to a concern that lower proﬁciency L2 learners would be less likely to show
the dual functions in the lexicon or they may not know the meanings of individual words.
All the non-native speakers participated in a proﬁciency test in order to ensure the
comparability of English proﬁciency. The Michigan State University English Language
Test (MSUELT) was used for the independent measure of English proﬁciency. The

MSUELT is designed to measure the English proﬁciency of non-native speakers. The

 

H Spanish and Thai have left-headedness in common. However, there is a potential concern
because, strictly speaking, Spanish and Thai are somewhat different in terms of compound formation. As
described earlier, in Spanish, it is more productive to insert a preposition between two nouns in a
compound, although there are some Spanish compounds that do not have a preposition between two nouns.
In contrast, Thai does not require the insertion of a preposition between two nouns. In this study, I decided
to merge both Spanish and Thai speakers of English into the LH group because there were no statistical
differences in comparing a group that contains both Thai and Spanish with a group that includes Thai alone

(yet, see the conclusion section in this study for a future research agenda).

74

functions of the MSUELT test are to place students in a level of classes at the English
Language Center of Michigan State University and to determine students’ level of
English in order to gain admission to regular university classes. One version of the

MSUELT test consists of listening, reading, vocabulary, and grammar sections. In this

study, the L2 participants took the vocabulary section and the grammar section”. The
vocabulary section consisted of 25 test items and was presented in a multiple choice
format. The grammar section was made up of 40 multiple choice questions. In both
sections, there was a sentence with a blank and the participants were asked to choose the
word or the phrase that best ﬁts into the blank from four choices. The participants had 15
minutes to complete the vocabulary section and 20 minutes to answer the grammar

section. The results of the proﬁciency tests are provided in Table 9.

Table 9. The summary of the L2 proﬁciency tests

 

 

RH (n = 27) LH m: 27)
Vocabulary .87 (.09) .84 (1.00)
Grammar .86 (.47) .85 (.75)

 

Note. The numbers in the parentheses indicate standard deviations.

For the ease of comparison, the raw scores were converted into percentages by dividing
raw scores of the two tests with the maximum scores in corresponding each test. Overall,
the two groups scored above 80 percent on each test. The inferential statistics revealed

that the RH and LH groups did not differ signiﬁcantly in vocabulary scores, t(52) = .40, p

 

l . . . . .
2 Due to the copyright issue, the two tests used In thlS study could not be prowded.

75

= .53 or grammar scores, t(52) = 1.98, p = .17. Therefore, the proﬁciency levels of the

two groups were comparable.

3.6. Method — Task 1

3.6.1 . Materials

All the stimuli in Task 1 are based on naturally occurring complex nominals taken
from the COCA (Corpus of Contemporary American English) corpus, so as to minimize
the familiarity and plausibility issue of stimuli. Four types of complex nominals were
included in relation to the number of meanings of either the modiﬁer or head: MM, MS,
SM, and SS. The number of meanings of either the modiﬁer or head was determined with
reference to the on-line Wordsmyth dictionary. If a single word (e.g., bank) had two
entries in the dictionary, the word was categorized as multiple. If a single word (e.g.,
juice) had one entry in the dictionary, the word was counted as single.

With the meaning criteria of single words, the complex nominals were randomly
selected from a larger set of familiar compounds found in the COCA corpus (Corpus of
Contemporary American English, Davies, 2009). The corpus is the ﬁrst large and diverse
corpus of American English and consists of more than 400 million words including the
ﬁve balanced genres of spoken, ﬁction, magazines, newspapers, and academic journals.

Each of the four conditions has 24 complex nominals. Appendix A includes a full
set of experimental items. Also, 96 ﬁllers having insensible interpretations (Appendix B)
were also constructed to make the number of sensible and insensible items equal. Both
the frequency and the number of syllables of complex nominals were matched across the

four conditions. The frequency of the whole complex nominals were based on the COCA

76

corpus. In order to minimize the inﬂuence of outliers, the logarithmic transformations

were used for the frequency. Table 10 summarizes the statistics of each condition.

Table 10. The frequency and the number of syllables of complex nominals in Task 1

 

 

Condition Frequency (The whole words) Syllables (The whole words)
SS (n= 24) 3.08 (1.32) 2.78 (.65)

SM (n= 24) 2.92 (1.41) 2.79 (.66)

MS (n= 24) 2.86 (1.72) 2.71 (.81)

Wﬁ= 24) 2.59 (1.71) 2.33 (.56)

 

The test items of the four conditions did not signiﬁcantly differ in terms of number of
syllables, F (3, 92) = 2.51, p = .064, or frequency, F (3, 92) = .41, p = .75. Four native
speakers of English judged the acceptability of the test items on a six-point scale (1: Not
acceptable, 6: Acceptable). In the ofﬂine test, complex nominals were judged highly

acceptable, M = 5.4, SD = .7, and the ﬁllers were unacceptable, M = 1.3, SD = .5.

3.6.2. Procedure

Participants took part in a computer-based session that lasted approximately 30
minutes. Before beginning the actual computer session, participants ﬁlled out a short
biographical questionnaire about age, gender, native language, English language
background, and self-evaluations of English proﬁciency in listening, reading, grammar,
and vocabulary. Appendix C illustrates the instruction of Task 1 that was given to the
participants.

The task in Task 1 was a lexical sense decision task (Gagné, 2001, 2002; Gagné

& Spalding, 2004; Gagné, Spalding, & Ji, 2005). Participants were asked to judge

77

whether a complex nominal had a sensible interpretation by pressing a button labeled yes
or a button labeled no.

The test items were presented in isolation, although it is equally interesting to
examine how complex nominals are interpreted in contexts (see Conclusion for
suggestions for future research). Both semantic representations of each constituent and
contextual information contribute to interpretations of complex nominals. Discourse
context can inﬂuence the interpretation of a combined concept (Gagné & Murphy, 1996;
Gerrig & Bortfeld, 1999; Gerrig & Murphy, 1992). At the same time, context is not a
necessary condition for interpreting novel compounds (Coolen, van Jaarsveld, and
Schreuder, 1991). That is, a novel compound can be interpreted on the basis of the
meanings or semantic representations of each constituent. This dissertation focuses on the
processing of complex nominals in isolation. As noted in Coolen, van Jaarsveld, and
Schreuder (1991), the examination of compounds in isolation allows us to understand
how the representations of each constituent contribute to the interpretations of complex
nominals.

The task was constructed using E-Prime software (Psychological Software Tools,
Pittsburgh, PA). The task consisted of a practice session and an actual experimental
session. The practice trial included 5 sensible and 5 non-sensible complex nominals. To
begin, participants saw a “Ready” and “Please press the spacebar when you are ready.”
Immediately after participants pressed the spacebar, a ﬁxation point “+” appeared on the
computer screen and stayed on the screen for 1000 milliseconds. Next, a complex
nominal appeared on the computer screen and remained on the screen until participants

pressed a button labeled yes or a button labeled no on the keyboard for a judgment of the

78

sensibility of complex nominals. The complex nominal remained for a maximum of
5000 milliseconds, which was determined by pilot tests with ten L2 learners who had
comparable levels of English proﬁciency. The cycle started again with a ﬁxation point.
The actual experimental session had the same procedure as the practice trials. The
experimental session contained 96 sensible (as in Appendix A) and 96 non-sensible (as in
Appendix B) complex nominals. In order to lessen participants’ fatigue, participants saw
the pause sign “Now you can take a break. Please press the spacebar when you are
ready.” after 24 items were presented. Thus, the participants saw 8 blocks of complex
nominals (made up of 96 plausible and 96 implausible items). The 8 blocks were
randomized across the participants and the items in each of 8 blocks were presented in a

random fashion. Figure 21 demonstrates the procedure of the lexical sense decision task.

Figure 21. The procedure of the lexical sense decision task

 

 

 

 

Ready?

Please press the spacebar [:> + :;f orange juice

when you are ready.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

<1] J boat jam C l +

 

 

 

 

 

 

 

 

i1

 

 

 

 

Now you can take a break.

 

> Please press the spacebar ‘ NL

11

coffee cup

when you are ready.

 

 

 

 

 

 

 

 

79

3.7. Method —Task 2

3.7.]. Materials and procedures

Students participated in a computer-based WM task that lasted about 15 minutes.
A version of the operation span test (Hitch, Towse, & Hutton, 2001; Turner & Engles,
1989; Unsworth, Heitz, Schrock, & Engle, 2005) was designed to measure Ll working
memory capacity, regardless of the participants’ L1 backgrounds. The operation span
task was intended to increase the practicality and comparability of measuring the
construct of WM. In second language research, it is more common to use WM tasks
based on language such as reading or listening span tasks. Recent SLA studies indicated
the difﬁculties of using an L2 working memory task (e.g., Gass & Lee, in press; Sagarra,
2008). This caution of using an L2 WM task is related to practicality issues in a second
language leaming setting (but not in a foreign language learning setting). This study, for
example, included a NS group of English speakers and L2 learners of four different L1
backgrounds, which would require 5 different language-based WM tasks. Although it is
not impossible to make 5 different language-based tasks, another issue is whether the
language-based WM tasks using 5 different languages are comparable in measuring WM
capacity because it is not simply assumed that the same formats using different languages
can measure the same underlying construct of WM. In order to overcome the issues of
practicality and comparability, this study used an Operation span, which is not only
common in cognitive psychology (e.g., Hitch, et. al., 2001; Kane & Engle, 2003; Kane, et.
al., 2001; Turner & Engles, 1989; Unsworth et. al., 2005) but also is reported to measure
the same underlying construct of WM when testing latent variable modeling (e.g.,

Conway etal., 2005).

80

In operation span tasks numbers are used for recalls, instead of words. It is
common to use words as recalls in the WM tests (Conway et al., 2005). L2 words could
be adopted as recalls. However, the measure of WM memory might be compounded with
L2 proﬁciency when recalls are L2 words because L2 vocabulary knowledge is an aspect
of L2 proﬁciency. Alternatively, L1 words could be used as recalls, instead of numbers.
Again, it is not practical to use Ll words in the context of a second language learning
environment. In this study where the participants were ﬁom ﬁve different Lls, for
example, it is possible to have ﬁve different language versions of the recalls such as
English, Chinese, Korean, Spanish, and Thai. Furthermore, the translation equivalents of
a language do not necessarily result in a measure of the same underlying constructs in the
various languages. For example, the Korean translation of the English word snow is nwun,
which has two meanings in Korean (“snow” and “eye”). Therefore, it was more feasible
to use numbers as recalls.

In general, the WM task of this study is similar to a version of reading span tasks
(e.g., Daneman & Carpenter, 1980), in which sentences are presented as interfering
stimuli and the recall items are either the last word of each sentence or a series of words
presented independently after each sentence. In this study, however, sentences in a
reading span task were replaced by mathematical equations in the operation tasks.
Another notable characteristic of this version of operation span is that it uses numbers
(but not words) as the recall stimuli. In order to increase the costs of processing and
storage, this study used mathematical equations using four digits (e.g., (2X3) + 2 = 8) as

interference and two digit numbers as recalls.

81

A series of pilot tests was administered to determine the methodological details of

the operation span task that was used in the main study”. I decided to present the WM
test via computer so that 1 could control the processing time of interference and the
storage time of recall numbers. The ﬁrst pilot test was intended to measure the timeline of
presenting interferences (i.e., mathematical equations) and presenting recalls. Four
participants (two N85 and two NNSs) saw four blocks of WM tests consisting of four
equations and four recalls. The three blocks differed in terms of how long each equation
stayed, ranging from 4000 ms to 5000 ms, to 6000ms. According to the four participants,
the block presented with 4000 ms-interval was highly challenging while the blocks
presented with 5000 ms or with 6000 ms intervals were manageable. In order to increase
individuals’ variability, I decided to use 5000 ms as the maximum time of presenting
each mathematical equation. Secondly, the presentation of recalls was also tested.
Another group of ﬁve students (two NS5 and three NNSs) saw three blocks of WM tests
made up of four mathematical operations and four recalls. The operations were presented
in 5000 ms and the presentation of recalls varied ranging from 1000 ms to 2000 ms to
3000 ms. In the case of the 1000 ms-intervals, the participants could recall approximately
40% of the recalls, suggesting that the task was too challenging. The participants could
remember above 90% of the recalls when the recalls stayed for 3000 ms while about 75%
of the recalls were remembered with 2000 ms-interval. In order to decrease the rehearsal

time of the recalls, I decided to present the recalls for 2000 ms. Using the above timelines,

 

13 Since pilot studies are based on a small number of participants, I did not run inferential

statistics. The rigor of this test remains as future research.

82

the second part of the pilot test was administered. The purpose of the pilot test was to
determine the maximum number of recalls. The results of a group of 10 NNS students
revealed that the main variations of the recall scores were found in 4 and 5 mathematical
equations, about 78% and 58%, respectively. In a block made up of 6 operations, the
chance of the recalls dropped to less than 30%. Therefore, 1 determined to use blocks

containing 5 recalls as the maximum recalls.

Table l 1. A sample of the working memory test

 

 

Equations Recalls
(2 x 2) + 3 = 6 ' 34
(7 + 5) - 6 = 6 39

 

Based on the pilot tests, the following method was used in the main test. The WM
task consisted of two parts. The ﬁrst part of the task involved a judgment of correctness
of a series of mathematical equations in order to ensure that the equations were being
processed and not just stored. Participants read a series of 2, 3, 4, and 5 simple
mathematical equations (using addition, subtraction, multiplication, and division). The
participants were asked to answer whether each solution was correct by clicking
corresponding buttons. The second part was to remember numbers. Immediately after
solving each equation, participants saw two digit numbers, which they were asked to
remember. At the end of a block of equations, they were asked to recall the two-digit
numbers they saw after each equation. Identical two-digit numbers (e.g., 22 or 33) were

not used for recalls. A sample of a block containing two mathematical equations and

83

recalls are shown in Table 11. The full list of the operation span can be found in
Appendix D.

The WM task began with a ﬁxation cross (+), which remained on the screen for
1000 milliseconds. Next, an equation of the set appeared on the computer screen and
remained on the screen until participants pressed a button labeled yes or a button labeled
no to make a judgment on the correctness of the equation. The operation remained for a
maximum of 5000 milliseconds. Immediately after the participants pushed the button
indicating the correctness or incorrectness of the operation, a two-digit number appeared
and remained on the screen for 2000 milliseconds. The cycle started again with a ﬁxation
point. After all the operations and letters 'of a set had been presented, a screen appeared
which instructed participants to write the letters that they had seen on a piece of paper.
The procedure of the WM test is illustrated in Figure 22.

The task was created using E-Prime software (Psychological Soﬁware Tools,
Pittsburgh, PA). Each participant saw two practice blocks, each of which consisted of two
levels. The actual test consisted of 6 blocks, 2 at each level: 2 through 5. The maximum
number of recall numbers was 5 as determined by a pilot test with 6 students. The six
blocks of operations were totally randomized, so that the blocks of operations were

presented differently across the participants.

84

Figure 22. A sample of a block containing two equations and two recalls in the WM test

 

 

 

 

Ready?

Please press the spacebar k + (2 X 2) + 3 = 6

 

 

 

 

when you are ready.

 

 

 

 

 

 

 

 

ﬂ

 

 

 

 

(7+5)-6=6 C: + (:3 s4

 

 

 

 

 

 

 

 

ﬂ

 

 

 

 

Great! Now you have seen a series of

equations and two-digit numbers.

 

 

34

 

<_Ar

Write as many of the numbers as you

<_Ar

 

can in your response booklet. Please

press the spacebar when you are ready.

 

 

 

 

 

 

 

 

3.7.2. Scoring Procedures

A version of partial-credit load scoring (Conway et al., 2005) was used.
Regardless of the order of the recalled numbers, the scoring consisted of giving credit for
each correct number in a set. In scoring, a single digit of each of the two-digit numbers
was credited. For example, when a recall was 23 and a participant recalled it as 24, one
point was credited. If students recalled the two-digit number correctly regardless of the
order of the recalled numbers, they received two points for the recall. For instance, those
who recalled 39 as either 39 or 93 got two points for the recall.

The total number of points for participants was 56 (2 blocks of 2 operations with

2 digit recall number = 8; 2 blocks of 3 operations with a 2 digit recall number = 12; 2

85

blocks of 4 operations with a 2 digit recall number = 16; 2 blocks of 5 operations with a 2
digit recall number = 20). 1 also gave a zero for each item for which a participant judged

an incorrect equation (e.g., l + 2 = 4) as correct or vice versa.

3.8. Method - Task 3

3.8.1. Materials

 

The Stroop test is intended to measure how successfully a participant inhibits g’
certain irrelevant information when conﬂicting information is presented. In cognitive
psychology, Stroop tests (Stroop, 1935) have been reported as the “golden” measure of
inhibitory control (MacLeod, 1991, 1992). In the Stroop test, participants see words and
determine in what color the word is written. Some of the words will be congruent with
color (the word "red" written in red); some will be neutral (e.g., a symbol @ written in
red); and some will be incongruent (the word "blue" written in red). This task usually
yields longer reaction times for the incongruent condition (Stroop, 193 5). These longer
reaction times are assumed to be measuring linguistic interference.

A modiﬁed English version of the Stroop test, used in Gass and Lee (in press),
was used to measure Ll/L2 inhibitory control. The test measures the L2 inhibitory
control of L2 learners of English and the L1 inhibitory control of native speakers of
English.

In the Stroop test, participants see a series of color words (e.g., RED) and
determine in what color the word is written, by pressing keys on the keyboard which

were taped with red, green, or blue. Some of the words are congruent with color (the

86

word "red" written in red); some will be neutral (a symbol @ written in red); and some
will be incongruent (the word "blue" written in red).

The test items consist of 60 congruent items (i.e., colors and words are matched)
and 60 incongruent item (i.e., colors and words are not matched). There were 20

congruent and incongruent trials for each color.

3.8.2. Procedures and scoring procedures

The task began with a ﬁxation point on the screen, which stayed on the screen for

 

1000 milliseconds. Then, a color word followed and the participants were to push a
button to indicate the color that the word was written in. The word remained on the
screen until the button was pushed. When there was no response, the word stayed on the
screen for 4000 milliseconds. A ﬁxation point appeared again, immediately after the
button was pushed.

There were three phases of the Stroop test: (1) a practice session using actual
words, (2) a Stroop test using actual words, and (3) a baseline session using @ symbols.
Participants saw the baseline session after the Stroop with the actual words. Across each
participant, the practice session was presented in a ﬁxed format, whereas the Stroop test
and the baseline session were randomly presented.

The practice session consisted of six examples using actual color words, which
contained three congruent and three incongruent trials. In the actual Stroop test,
participants saw both 60 congruent items (i.e., colors and words are matched) and 60
incongruent items (i.e., colors and words are not matched). There were 20 congruent and
incongruent trials for each color. Finally, as a baseline, participants saw 30 @ symbols

(10 in red, 10 in blue, and 10 in green).

87

The Stroop test was scored by subtracting the mean reaction time of the neutral
trials from the mean reaction time of the incongruent trials (Kane & Engle, 2003;
MacLeod, 1991). Individual data points exceeding two standard deviations above/below

the means were eliminated from the data analysis.

88

CHAPTER FOUR —RESULTS

There were three computer-based tasks in this study: (1) a lexical sense decision task, (2)
a working memory test, and (3) an inhibition test. The ﬁrst task was designed to examine
whether native speakers and non-native speakers use either computational routes or
retrieval routes or both when they comprehend complex nominals. The second task was
intended to investigate whether working memory (WM) capacity is related to the
interpretations of complex nominals. The third task was aimed to examine the effect of

inhibitory control ability on the recognition of complex nominals.

4.1. Task 1
Both the mean response times and the accuracy scores of the three groups in the

four experimental conditions are used to address research questions. The response times

and the accuracy scores were analyzed separately”. Both by-subject (F 1) and by-item (F2)

 

14 There were no violations of statistical assumption in both the mean response times and the
accuracy scores of the three groups. First, the univariate normality of each of four dependent variables
among the three groups was examined via skewness, kurtois, Shapiro-Wilk’s test, normal Q-Q plots, and
detrended Q-Q plots. The multiple indications of the univariate normality suggest that every dependent
variable meets the normality assumption. The bivariate normality, one of the assumptions of multivariate
normality was checked through a scatter plot, in which every plot appears close to elliptical shapes. Thus,
in general, the dependent variables seem not to violate the univariate and multivariate assumption. Second,
the equality of univariate variance and multivariate covariance was examined. The univariate homogeneity
assumption was checked through Levene’s test, in which there was no violation of equal variance in each

of four dependent variables (p > .05). MANOVA assume that for each group the covariance matrix is

89

analyses were conducted”. Individual data points exceeding above/below two standard

deviations were eliminated from data analysis”. The descriptive statistics of the mean
response times the four conditions is given in Table 12 and Figure 23. Table 13 and
Figure 24 show the accuracy scores of the three groups in the four experimental

conditions.

Table 12. The response times in the four experimental conditions

 

 

Conditions
SS SM MS MM
NS (11 = 20) 948.01 937.20 946.68 1002.38
(115.34) (132.16) (156.34) (167.39)
RH (n = 27) 1572.71 1754.66 1666.27 1876.87
(351.14) (366.63) (319.39) (431.48)
LH (n= 27) 1566.24 1771.39 1735.51 1994.87
(342.29) (604.01) (522.94) (694.84)

 

Note. The numbers in the parentheses are standard deviation.

 

similar, and Box’ M test suggests the equality of covariance matrix is met (p= .23). Finally, since each task

was answered by a different participant, their data are considered to be independent.

15 In psycholinguistics literature, by-subject (F1) and by-item (F2) analyses in psycholinguistics
had been used as standard quantitative methods since an inﬂuential work by Clark (1973). The rationale for
the use of F1 and F2 analyses was that both human participants and linguistic materials should be modeled
as random variables. Yet, the use of F1 and F2 analyses has not remained unchallenged (see Baayen,
Davidson, & Bates, 2008; Raaijmakers, Schrijnemakers, & Gremmen, 1999 for the recent developments of

quantitative analyses of psycholinguistic data).

‘6 Researchers vary in handling outliers. One uses 3sd as the cutting point of outliers and the
other uses 2.5 sd as the outer bounds of outliers. In this study, I followed a common practice of handling
outliers (2 sd) in second language psycholinguistic research such as Beck (1998), Gass (2001), and Bley-

Vroman and Masterson (1989).

90

Figure 23. The mean response times of the three groups in the four conditions

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2500
2000 a m"...-
1500 “"' “mug —o—Ns
1000 sw— "C -." RH
500 ...‘.. LH
0 r
SS SM MS MM
Table 13. The accuracy scores in the four conditions
Conditions
SS SM MS MM
NS (n = 20) .95 (.06) .90 (.09) .90 (.09) .83 (.12)
RH (n = 27) .85 (.12) .79 (.17) .77 (.17) .70 (.17)
LH (n= 27) .80 (.17) .74 (.18) .76 (.17) .65 (.23)
Note. The numbers in the parentheses are standard deviation.
Figure 24. The accuracy scores in the four conditions
1
0.9
0.8
0.7
0.6 .
0.5 NS
0.4 "" RH
0.3 ...‘.. LH

0.2
0.1
0

 

 

 

4.1.1. Research question 1: The difference between the NS and the NNS groups

The ﬁrst research question addressed the difference between the NS group and the

two NNS groups in interpreting complex nominals. Speciﬁcally, I questioned whether the

NS group is signiﬁcantly faster and is more accurate in comprehending items in all the

91

 

 

four conditions than the two NNS groups. A multivariate analysis of variance
(MANOVA) was performed with a Group variable (three-levels: NS, RH, and LH) as an
independent variable as well as the response times and the accuracy scores in the four
conditions as dependent variables. Reaction times and accuracy scores were analyzed
separately.

Evidence of Diﬂerence between NS and NNS. The MANOVA analysis of the

response times revealed that the main effect across the four conditions was signiﬁcant,
Wilks’ Lambda = .44, F (8, 136) = 8.51, p = .0001 , 772: .33. The four separate univariate

analyses of variance were conducted to locate the source of the signiﬁcant difference in a

multivariate space. The results showed that there were signiﬁcant differences across all

the four conditions in both by-subject and by-item analyses, F ,_MM (2, 71) = 25.56, p

= .0001, ,7? = .42; F2_MM (2, 46) = 78.70, p = .0001, 1,2 = .77; F,_MS (2, 71) = 29.09, p

.0001, ’72: .45; F2_M5 (2, 46) = 63.49, p = .0001, 3,2 = .73; F,_SM (2, 71) = 26.55, p

.0001, 772: .43; F2_SM (2, 46) = 106.30, p = .0001, 372 = .82; F)_SS (2, 71) = 30.68, p

.0001, n2= .46; F 2_55 (2, 46) = 101 .05, p = .0001, n2: .82. Post-hoc comparisons of

response times were carried out using Bonferrnoni adjustment. As seen in Table 14, the
NS group was signiﬁcantly faster in judging the meanings of complex nominals than the
two NNS groups across all the four conditions.

The MANOVA analysis of the accuracy scores also revealed that the main effect

across the four conditions was signiﬁcant, Wilks’ Lambda = .74, F (8, 136) = 2.77, p
= .007, n2: .14. The univariate results of accuracy scores showed that there were

signiﬁcant effects across the four conditions in both by-subject and by-item analysis,

92

FI_MM(Zs 71) = 6.43, p = .003, '72: .15; F2_MM(21 46) = 8.64,p = .001, I72: .27; F1_MS(2,

71) = 5.56, p = .006, 42= .14; F2_M5 (2, 46) = 8.95, p = .001, 37’: .28; F,_SM (2, 71) = 5.97,

p = .004, 42: .14; F2_SM (2, 46) = 14.16, p = .0001, ’72: .38; F,_SS (2, 71) = 6.43, p = .003,

n2: .15; F 2_SS (2, 46) = 7.39, p = .24. Post-hoe pairwise comparisons using Bonferrnoi

correction were performed, as seen in Table 15.

Table 14. Bonferroni Post-hoe comparisons of response times

 

 

 

 

 

Condition Group
NS RH

MM RH 874.49"

LH 992.49“ 1 17.99
MS RH 719.59“

LH 789.82" 70.23
SM RH 817.46"

LH 834.19" 16.72
SS RH 624.70M

LH 618.24" 6.46

 

**. p < .01. Note. The numbers indicate the mean differences between two groups

Table 15. Bonferroni Post-hoe comparisons of accuracy scores

 

 

 

 

 

Condition Group
NS RH

MM RH .13 (p = .055)

LH .19“ (p = .002) .06 (p = .66)
MS RH .126*(p=.017)

LH .14" (p = .009) .01 (p = 1.0)
SM RH .11*(p=.07)

LH .16* (p = .003) .05 (p = .71)
SS RH .09 (p = .054)

LH .13 (p = .002) .04 (p = .67)

 

*. p < .05. Note. The numbers indicate the mean differences between two groups

93

The post-hoe comparisons of accuracy scores indicated that the NS group interpreted
complex nominals signiﬁcantly more accurately across all conditions than the RH and
LH groups, although the differences between NS and RH were marginal in MM and SS

conditions.

4.1.2. Research question 2: The difference between the LH and the RH groups

The second research question asks whether there is the inﬂuence of L1
morphological knowledge on processing complex nominals. Of particular interest is
whether the headedness of the ﬁrst language affects processing L2 morphology, even for
advanced learners of a second language. Potentialdifferences between the LH and RH
groups were examined using two separate mixed ANOVAs, one for response times and
the other for accuracy scores. Within-subject factors were the measures of each group
(i.e., response times and accuracy in the four conditions) and a between-subject factor
was a group variable with two-levels (i.e., RH and LH).

No evidence of LI inﬂuence: RH versus LH. The results of response times showed

that the within-subject effect was signiﬁcant, F1(3, 156) = 11.58, p = .0001, n2 = .40; F 2
(3, 92) = 7.49, p = .007, r72: .08. However, there were no signiﬁcant between-subject
effect, F1 (1, 52) = .23, p = .63; F 2 (1, 92) = .61, p = .44. The interaction between within

and between-subject factors was not signiﬁcant, F 1 (3, 156) = .39, p = .76; F 2 (3, 92)

= .90, p = .45. Therefore, RH and LH groups did not differ signiﬁcantly in terms of
response times.

The results of accuracy scores were similar to those of response times. The

within-subject difference was signiﬁcant, BB, 156) = 11.74, p = .0001, n2 = .43; F 2 (3,

94

92) = 3.33, p = .023, n2 = .10, yet there was no between-subject difference, F 1 (1 , 52) =
l.49,p = .23; F2 (1, 92) = 1.01,p = .31 as well as no interaction, F1(3, 156) = .36,p = .79;
F2 (3, 92) = .27,p = .27.

In general, the results of both response times and accuracy scores demonstrated

that there was no signiﬁcant difference between the RH and LH groups, suggesting that,

at least with the participants of the current study, L1 morphological knowledge did not

play a role in recognizing L2 complex nominals”.

4.1.3. Research question 3: Within-group differences

The third research question concerns whether the NS group and the NNS groups
differ in recognizing complex nominals in the four experimental conditions. Two
different hypotheses have been proposed. The NS group is not expected to show a
substantial behavioral difference across the four conditions (i.e., MM =MS = SM =SS),
considering the high probability of reliance on retrieval routes or efﬁcient morphological
computations. In contrast, the NNS groups are expected to produce the graded
interference in accordance with the processing costs of each condition. Speciﬁcally, the

NNS groups are predicted to reveal the greatest interference in the MM condition, the

 

17 . . . . . . .
When the reaction times across the four condltrons collapse 1nto a measure (1.e., the combined

average of the four conditions), the RH group (M =1 717.63, SD = 302.87) was faster than the LH group
(M= 1767.25, SD = 438.77), descriptively. The accuracy scores were similar to the reaction times; the RH
group (M =.78, SD = .1 l) judged the test items a bit more correctly than the LH group (M = .74, SD =.14).
However, inferential statistics showed that the two groups did not differ signiﬁcantly in both the reaction

times (t(52) = .48, p = .63) and in the accuracy scores (t(52) = 1.22, p = .23).

95

intermediate interference in the MS and the SM conditions, and the smallest inference in
the SS condition (i.e., MM > MS, SM > SS).

Two separate within-subject design were used to test the two hypotheses in the
ﬁrst research question. Dependent variables were reaction times and accuracy scores of
the NS group and the NNS group (combining RH and LH groups) in the four conditions,
which were run separately.

Mixed evidence of uniform interference in NS group. Results of response times
revealed that the NS group did not differ signiﬁcantly in responding to all the four
conditions, F1_N5 (3, 57) = l.83,p = .15; F2_NS (3, 96) = 2.18,p = .10. In terms of
accuracy, the by-subject analysis of accuracy revealed N83 to show a signiﬁcant main

effect, F ,_NS (3, 57) = 8.53, p = .0001, 772: .31, and the by-item analysis indicated
marginal differences, F 2_NS (3, 92) = 2.51 , p = .06. A follow-up Bonferroni test was

performed to examine the locus of the signiﬁcant difference, as seen in Table 16.

Table 16. Post-hoe comparisons of accuracy scores of the NS group

 

 

MM MS SM
MS .067* (p = .03 8)
SM 072* (p = .022) .01
ss .113* (p = .005) .05 .04

 

*.p < .05.

Table 16 shows clearly that the NS group made more mistakes in interpreting the items in
the MM condition.
Evidence of the graded interference in the NNS group. The results of response

times indicate that both the RH and LH groups did differ signiﬁcantly in responding to all

96

the four conditions, F LR” (3, 78) = 7.60, p = .0001, n2 = .23; F 2_RH (3, 92) = 6.82, p

.0001, ’72 = .18; F1_LH(3, 78) = 5.37, p = .002, ’72: .17; F2_LH (3, 92) = 11.89, p

.0001, n2 = .28. The follow-up pairwise comparisons revealed that, in the MM and SS

conditions, the 304.17 mean difference of the RH group (p = .0001) and the 428.63 mean
difference of the LH group (p = .017) were signiﬁcant. Thus, the locus of signiﬁcant
differences in the two NNS groups was the MM and SS conditions, which corresponded
to the highest and the lowest processing costs, respectively.

Similarly, the accuracy analyses revealed that there were signiﬁcant main effects

on each ofthe two NNS groups, F/_RH (3, 78) = 5.43,p = .002, n2 = .17; F2_RH (3, 92) =
2.74, p = .048, 3,2 = .08; F,_LH (3, 78) = 6.62, p = .0001, ’72 = .20; F2_LH (3, 92) = 3.25, p

.025, n2 = .10. A follow-up Bonferroni test was reported in Table 17.

Table 17. Post-hoe comparisons of accuracy scores of RH and LH group

 

 

 

 

Group Condition
MM MS SM

RH MS .07

SM .02 .02

SS .15* (p = .035) .08 .06
LH - MS .12"I (p = .02)

SM .10 .02

SS .16" (p = .001) .05 .07
*. p < .05.

In terms of accuracy scores, the RH group interpreted the items in the MM condition less
accurately than the SS condition; the LH group made more mistakes in judging the MM

condition than the MS and SS conditions.

97

4.1.4. Research question 4: Comparing the NS and the NNS groups with respect to
holistic representations of complex nominals

The fourth research question is whether the NS group and the NNS groups store
some of the complex nominals as a whole. In order to address this question, I look at the
relationship between the frequency of the whole complex nomninals and the performance
of NS group and NNS groups. This method is common in inferring the holistic
representations of multi words such as formulaic sequences (e.g., J iang & Nekrasova,
2007).

A series of correlation analyses was carried out with (1) frequency measures and
(2) the measures of performances of the three groups in terms of reaction times. First, 1
create new measures of the whole word frequency based on the COCA corpus (Corpus
Of Contemporary American English). Speciﬁcally, regardless of the 4 experimental
conditions (SS, SM, MS, and MM), 96 test items were merged and divided into three
conditions (High, Mid, Low) according to the frequency of the whole complex nominals.
An assumption of full-listing models (in favor of the holistic representations of complex
nominals) is that the semantic integrations are not the part of the mental representation.
Note that I assume that semantic computation would lead to the graded interference
across the four experimental conditions. With the context of this study, the holistic
representations of complex nominals imply that the manipulated processing costs in the
four experimental conditions are not an important consideration. Rather, full-listing
models predict that the frequency of the complex nominals is the key to the performances

of the NS and the NNS groups. These three variables are used for the measures of the

98

frequency. Table 18 shows the descriptive statistics of the frequencies of the three

conditions.

Table 18. The descriptive statistics of the 96 test items according to the frequency of the
whole words

 

 

Frequency
Hi (n = 32) Mid (n = 32) Low (n =32)
4.51 (.10) 3.00 (.39) 1.10 (.15)

 

Note. The numbers in the parentheses are standard deviation.

Inferential statistics of the frequency measures revealed that there is a signiﬁcant
difference among the three conditions, F (2, 93) = 237.88, p = .0001. A post-hoe
comparison with Bonferroni correction shows that all the three conditions differ
signiﬁcantly; Hi vs. Mid, M= 1.51, SD = .16,p = .0001; Hi vs. Low, M: 3.43, SD = .16,
p = .0001; Mid vs. Low, M= 1.91, SD = .16,p = .0001.

Overall, I predict that the reaction times of the participants are inversely related to
the frequency of the whole complex nominals (see Monsell, 1991 for the overview of
frequency effect). That is, the higher frequency words are expected to lead to faster
reaction times. Yet, I predicted that the frequency effect would emerge only with the NS
group, but not with the two NNS groups, given that the NS group is likely to have more
stored complex nominals as a whole.

A series of separate correlation analyses were conducted for each group. The
reaction times of each group are examined in relation to the three frequency measures.

Evidence of the ﬁequency eﬂect of complex nominals for the NS group. The

results of the correlation analyses are presented in Table 19. In general, the NS group

99

behaved in a predicted fashion. With respect to the directionality of the correlations, the
frequencies of the whole complex nominals are inversely correlated with the reaction
times of the NS group. In terms of the magnitudes, the high frequency complex nominals
showed the strongest correlation with the reaction times of the NS group, r (30) = -.3 8, p
= .03., the mid-frequency words revealed the marginal correlation, r(30) = -.34, p = .059,
and there is a weak relationship between the low-frequency words and the reaction times
of the NS group, r (30) = -25, p = .16. In case of the NS group, therefore, some of the

high frequency complex nominals are likely to be stored as a whole.

Table 19. The correlations between the frequency and the reaction times in the NS group

 

 

The frequency of the whole complex nominals The reaction times of the NS ﬂup (n = 20)
Hi (n= 32) r = -.38, p = .03
Mid (n= 32) r = -.34, p = .06
Low (n= 32) r = -.25 p = .16

 

Some evidence of the ﬁequency ejfect of complex nominals for the RH group.
Table 20 shows the results of the correlation analyses of the RH group. The directionality
of the high- and the mid-frequency words are in a predicted way. However, the low-
frequency words are positively correlated with the reaction times of the RH group, r (30)
= .20, which is not predicted by full-listing accounts. The positive correlation between the
low-frequency words and the reaction times of the RH group suggest that the RH group
may not store the low-frequency words as a whole but rather is likely to interpret the low-
frequency complex nominals via computations. In terms of the magnitudes, the marginal
correlations were found in high-frequency words, r(30) = -.34, p = .055, and in the mid—
frequency words, r(30) = -.32, p = .072. Therefore, although it is not highly strong, the

RH group may store some of the complex nominals as a whole.

100

Table 20. The correlations between the frequency and the reaction times in the RH group

 

 

The frequency of the whole complex nominals The reaction times of the RH group (n = 27)
Hi (n= 32) r = -.34, p = .06

Mid (n= 32) r = -.32, p = .07

Low (n= 32) r = .20p = .27

 

Weak evidence of the ﬁ‘equency effect of complex nominals for the LH group. The
results of the LH group, in general, replicate the performances of the RH group; however,
the strength of associations in the LH group was much weaker than in the RH group.

Table 21 shows the results of the correlation analyses of the LH group.

Table 21. The correlations between the frequency and the reaction times in the LH group

 

 

The frequency of the whole complex nominals The reaction times of the LH group (n = 27)
Hi (n= 32) r=-.19,p=.3l

Mid (n= 32) r = -.16, p = .39

Low(n=32) r=.11p=.56

 

As in the case with the RH group, the LH group behaves in responding to the high- and
the mid- frequency words in a predicted way. That is, the LH group responded faster to
the complex nominals with higher frequency. However, the reaction times of the RH
group was positively proportional to the low-frequency words, r(30) = .11. Similar to the
RH group, the fact that low-frequency words and the reaction times of the LH group are
positively correlated suggest that the low-frequency words may be not stored as a whole
in the memory of the LH group. In terms of the magnitudes, there was no strong
correlations in high-frequency words, r(30) = -.19, p = .31, and in the mid-frequency

words, r(30) = -.16, p = .39. Given the weak correlations even for the high-frequency

101

complex nominals, the LH group is likely to recognize the complex nominals in this

study through computational routes but not retrieval routes.

To recapitulate, Table 22 summarizes the research questions that were addressed

in Task 1 and the corresponding ﬁndings for each research question.

Table 22. A summary of research questions and the corresponding ﬁndings in Task 1

 

 

 

 

 

 

Research questions Findings Summary

RQl. Diﬂerence between NSs and NNSs Yes The NS group performed better than the NNS
groups in all experimental conditions.

Do NSs show signiﬁcantly better performance L

in all four conditions than the two NNS groups

of high L2 proﬁciency?

RQ 2. LI inﬂuence of morphological Partially The LH group was descriptively slower than

knowledge in processing L2 complex nominals yes the RH group in responding to complex
nominals in English; however, the L1 effect,

Does the headedness of the L1 compound operationalized by the headedness of the L1,

structure affect the processing of L2 complex was not signiﬁcant between the two groups,

nominals? presumably due to high, comparable
proﬁciency of the two NNS groups.

RQ3. Uniform interference in the NS group Yes The NS group did not show a substantial

versus graded interference in the NNS groups difference in reaction times and accuracy
across the four conditions (i.e., MM =MS =

Do the NS group and the NNS groups differ in SM =SS). In contrast, the NNS groups did

recognizing complex nominals in the four show the graded interference in accordance

experimental conditions? with the processing costs of each condition
(i.e., MM > MS, SM > SS).

RQ 4. Comparing NSs and NNSs in terms of Yes for The NS group showed a strong frequency

whole-word representations the NS effect. In contrast, the NNS groups revealed a

group; weak frequency effect for the high-frequency

Do the frequencies of complex nominals as a partially complex nominals; however, the low-

whole play a signiﬁcant role in recognizing yes for frequency words are positively correlated with

complex nominals for both NS and NNS the NNS the reaction times of the two NNS groups,

groups? groups which was not predicted by full-listing models.

 

102

4.2. Task 2

Task 2 examines the potential role of working memory (WM) capacity on L2
complex nominal processing. Of particular interest in Task 2 is to examine the question
of whether L2 learners with higher WM capacity are more efﬁcient in processing a
second language than their counterparts with lower WM capacity, when comparing the
processing ability of NNS groups with comparable proﬁciency. l

The descriptive statistics for WM are provided in Table 23. The maximum score [I
of the WM test was 56. No signiﬁcant difference was found among the three groups, F (2,

71): .78,p= .46.

Table 23. A summary of working memory scores

 

 

Group
NS (n= 20) RH (n= 27) LH (n= 27)
43.10 (94:0 46.30 (6.44) 44.37 (10.39)

 

Note. The numbers in parentheses refer to the standard deviations.

A series of bivariate regression analyses were conducted with the WM scores of
each group as a predictor variable; criterion variables were the response times and the
accuracy scores of MM and SS conditions on the lexical sense decision task of each

group, which were measured in Task 1.

4.2.1. NS group
No evidence of the effect of WM capacity on response times in the MM condition.
The response times of the NS group in the MM condition was a criterion variable and the

WM scores of the NS groups was a predictor variable. Results showed that the WM

103

scores of the NS group cannot explain a signiﬁcant variance in their response times in the
MM condition, F(l, 18)= .03,p = .87.

No evidence of the eﬂect of WM capacity on accuracy in the MM condition. The
accuracy scores of the NS group in the MM condition was a criterion variable in this
regression model. Results revealed that the accuracy scores of the NS group in the MM
condition cannot be explained by the WM scores of the NS group, F (l , 18) = .13, p = .73.

No evidence of the effect of WM capacity on response times in the SS condition. A

 

dependent variable was the response times of the NS group in the SS condition. The
regression analysis indicated that the WM scores was not a good predictor of the response
times of the NS group, F(1, 18) = .08, p = .78.

No evidence of the eﬂect of WM capacity on accuracy in the SS condition. The
accuracy scores of the NS group in the SS condition was a criterion variable in this
analysis. Results demonstrated that the accuracy scores of the NS group in the SS
condition cannot be predicted by the inhibition scores of the NS group, F (l , 18) = .69, p
= .42.

Summary of NS group. As predicted, the performances of the NS group could not

be explained in terms of WM capacity.

4.2.2. RH group
No evidence of the eﬂect of WM capacity on response times in the MM condition.
The response times of the RH group in the MM condition was a criterion variable and the

WM scores of the RH group was a predictor variable. Results showed that the working

104

scores of RH group did not have an explanatory value for variance in the response times
of the RH group in the MM condition, F(1, 25) = .04, p = .85.

No evidence of the effect of WM capacity on accuracy in the MM condition. The
accuracy scores of the RH group in the MM condition was a criterion variable. Results
revealed that the accuracy scores of the RH group in the MM condition was not explained
signiﬁcantly by the inhibition scores of the RH group, F (1, 25) = .48, p = .49.

No evidence of the eﬂect of WM capacity on response times in the SS condition. A
dependent variable was the response times of the RH group in the SS condition. The
regression analysis showed that the WM scores of the RH group was not a good predictor
of the response times of the RH group, F(1, 25) = .01, p = .91.

No evidence of the eﬂect of WM capacity on accuracy in the SS condition. The
accuracy scores of the RH group in the SS condition was a criterion variable in this
analysis. Results revealed that the inhibition scores of the RH group did not explain a
signiﬁcant proportion of the accuracy scores of the RH group in the SS condition, F(l , 25)
= .01, p = .93.

Summary of RH group. The WM scores of the RH group was not a good predictor

of the performances of the RH group.

4.2.3. LH group

Evidence of the effect of WM capacity on response times in the MM condition. A
regression analysis was run using the response times of the LH group in the MM
condition. The response times of the LH group in MM condition were a criterion variable

and the WM scores of the LH group were a predictor variable. Results showed that the

105

WM scores of the LH group can explain a signiﬁcant proportion of variance in the
response times of the LH group in the MM condition, ,8 = -.41, t(25) = -2.27, p = .03, R2

= .17. There was an inverse relationship between the WM scores and the response times
of the LH group, as in ,B = -.41; those who have high scores of WM tend to respond faster
to the stimuli in the MM condition.

Evidence of the ejfect of WM capacity on accuracy scores in the MM condition. A ’

regression analysis was run using the accuracy scores of the LH group in the MM 5 :

 

condition. The accuracy scores of the LH group in the MM condition were a criterion E.
variable and the WM scores of the LH group were a predictor variable. Results showed

that the WM scores of the LH group can explain a signiﬁcant proportion of variance in
the accuracy scores of LH group in the MM condition, ,8 = -.41, t(25) = -2.21, p = .04, R2

= .16. The negative value of the beta indicates an inverse relationship between WM
scores and accuracy scores. This implies that the higher WM scores are, the more
accurately the LH group responded.

No evidence of the eﬂect of WM capacity on response times in the SS condition.
The dependent variable was the response times of the LH group in the SS condition. The
regression analysis showed that the WM scores of the LH group was not a good predictor
of the response times of the LH group, F (1, 25) = .24, p = .63.

No evidence of the effect of WM capacity on accuracy in the SS condition. The
accuracy scores of the LH group in the SS condition was a criterion variable in this
analysis. Results revealed that the WM scores of the LH group did not explain a

signiﬁcant proportion of their accuracy scores in the SS condition, F(l , 25) = .23, p = .64.

106

Summary of LH group. The WM scores of LH group can explain a signiﬁcant
proportion of the LH group in both their response times and accuracy scores in the MM
condition. In contrast, the WM scores of LH group were not a good indicator of their
response times and accuracy scores in the SS condition.

In summary, Table 24 shows the research question of Task 2 and its

corresponding ﬁndings.

Table 24. A summary of research questions and the corresponding ﬁndings in Task 2

 

Research questions Findings Summary

 

The eﬁiact of WM capacity on complex nominal Yes for The NS group and the RH group did not

processing. the LH show any substantial effect of WM capacity
group on complex nominal processing. In

Are Ll speakers and L2 learners with higher contrast, the LH group did reveal the effect

WM capacity more efficient in processing of WM capacity on L2 complex nominal

complex nominals than those with lower processing, particularly when they

capacity? processed the items in the MM condition

involving the highest processing costs.

 

4.3. Task 3
The descriptive statistics for inhibition scores are provided in Table 25.

Descriptively, the NS group showed greater interference than the RH group and the LH

18
group .

 

'8 An ANOVA revealed that the inhibition scores of the three groups did not differ signiﬁcantly,

F(2,71)=1.14,p=.33.

107

 

Table 25. A summary of inhibition scores

 

 

Group
NS (n= 20) RH (n= 27) LH (n= 27)
161.24 (.96) 113.55 (.96) 121.47 (.97)

 

Note. The numbers in the parentheses refer to the accuracy scores (Max = 1, Min = 0).

Note that the inhibition scores are based on the reaction times of the participants. Thus,
the higher inhibition scores correspond to the lower inhibitory control ability because the

higher scores mean the greater interferences. For example, those who have 100 inhibition

 

scores were better/faster in inhibition or underwent less interference than those who have
200 corresponding scores.

A series of a bivariate regression analysis was conducted to answer the research
question. The predictor variable was inhibition scores and the criterion variables were ‘
both response times and accuracy scores of MM and SS conditions from the lexical sense

decision task, which were measured in Task 1.

4.3.1. NS group

No evidence of the eﬂect of inhibitory control ability in response times in MM
condition. The response times of the NS group in the MM condition was a criterion
variable and the inhibition scores of the NS group was a predictor variable. Results
showed that the inhibition scores of the NS group cannot explain a signiﬁcant variance in
the response times of the NS group in the MM condition, F(l , 18) = .003, p = .96.

No evidence of the eﬂect of inhibitory control ability in accuracy in MM condition.
The accuracy scores of the NS group in the MM condition was a criterion variable in this

regression model. Results revealed that the accuracy scores of the NS group in the MM

108

condition cannot be explained by the inhibition scores of the NS group, F (1 , 18) = .10, p
= .75.

No evidence of the effect of inhibitory control ability in response times in SS
condition. A dependent variable was the response times of the NS group in the SS
condition. The regression analysis indicated that the inhibitory scores was not a good
predictor of the response times of NS group, F(l , 18) = 3.00, p = .10.

No evidence of the effect of inhibitory control ability in accuracy in the SS
condition. The accuracy scores of the NS group in the SS condition was a criterion
variable in this analysis. Results demonstrated that the accuracy scores of the NS group in
the SS condition cannot be predicted by the inhibition scores of the NS group, F ( 1 , 18)
= .90, p = .36.

Summary of NS group. As predicted, the performances of NS group could not be

explained in terms of inhibitory control ability.

4.3.2. RH group

Some evidence of the effect of inhibitory control ability in response times in MM
condition. In this regression model, the response times of the RH group in the MM
condition was a criterion variable and the inhibition scores of the RH group was a
predictor variable. Results showed that, although marginal, the inhibition scores of the

RH group can explain some proportion of variance in the response times of the NS group
in the MM condition, ,3 = .34, t(25) = 1.80, p = .084, R2 = .11. Also, the positive

directionality as in )8 = .34 is in a predicted direction because the inhibition measures of

109

this study correspond to the interference (i.e., high scores = higher interference in
inhibiting irrelevant information).

No evidence of the effect of inhibitory control ability in accuracy in the MM
condition. The accuracy scores of the RH group in the MM condition was a criterion
variable. Results revealed that the accuracy scores of the RH group in the MM condition
was not explained signiﬁcantly by the inhibition scores of the RH group, F (1 , 25) = 2.24,
p = .15.

No evidence of the eﬂect of inhibitory control ability in response times in the SS

 

condition. A dependent variable was the response times of the RH group in the SS
condition. The regression analysis showed that the inhibitory scores of the RH group was
not a good predictor of the response times of the RH group, F (1 , 25) = 1.34, p = .26.

No evidence of the effect of inhibitory control ability in accuracy in the SS
condition. The accuracy scores of the RH group in the SS condition was a criterion
variable in this analysis. Results revealed that the inhibition scores of the RH group did
not explain a signiﬁcant proportion of the accuracy scores of the RH group in the SS
condition, F(1, 25) = 1.39,p = .25.

Summary of the RH group. The response times of the RH group in the MM
condition can be explained to some extent in terms of inhibitory control ability of the RH
group. In other conditions, the inhibition scores were not a good indicator of predicting
the performances of the RH group.

4.3.3. LH group
Evidence of the effect of inhibitory control ability in response times in the MM

condition. A regression analysis was run using the response times of the LH group in the

110

MM condition as a criterion variable and the inhibition scores of LH groups as a
predictor variable. Results showed that the inhibition scores of the LH group can explain
a signiﬁcant proportion of variance in the response times of the LH group in the MM
condition, I9 = .47, t(25) = 2.57, p = .02, 122 = .21.

No evidence of the eﬂect of inhibitory control ability in accuracy in the MM
condition. The accuracy scores of the LH group in the MM condition was a criterion
variable. Results revealed that the accuracy scores of the LH group in the MM condition
was not explained signiﬁcantly by the inhibition scores of the LH group, F (l, 25) = 2.31,
p = .14.

No evidence of the effect of inhibitory control ability in response times in the SS
condition. A dependent variable was the response times of the LH group in the SS
condition. The regression analysis showed that the inhibitory scores of the LH group was
not a good predictor of the response times of the LH group, F (1, 25) = .01, p = .93.

No evidence of the eﬂect of inhibitory control ability in accuracy in the SS
condition. The accuracy scores of the LH group in the SS condition was a criterion
variable in this analysis. Results revealed that the inhibition scores of the LH group did
not explain a signiﬁcant proportion of the accuracy scores of the LH group in the SS
condition, F(1, 25) = 2.01,p = .17.

Summary of LH group. Inhibitory control ability can explain a signiﬁcant
proportion of the response times of the LH group in the MM condition. As in the case of
the RH group, the inhibition scores of the LH group were not a good indicator of the
response times of the LH group in the SS condition and their accuracy scores in the MM

and SS conditions.

111

 

In summary, Table 26 demonstrates the research question and the major ﬁndings

of Task 3.

Table 26. A summary of research questions and the corresponding ﬁndings in Task 3

 

 

Research questions Findings Summary

RQ. The eﬂect of inhibitory control ability on Yes for The NS group did not show any substantial

complex nominal processing. the NNS effect of inhibitory control ability on
groups complex nominal processing. In contrast,

Are L1 speakers and L2 learners with higher
inhibitory control capacity more efﬁcient in
processing complex nominals than those with

lower capacity?

the two NNS groups showed the effect of
inhibitory control ability on L2 complex
nominal processing, particularly when they
processed the items in the MM condition
involving the highest processing costs.
Furthermore, the effect of inhibitory control
ability was stronger in the LH group than in

the RH group.

 

112

CHAPTER FIVE - DISCUSSION

4.1. The second language mental lexicon

Task 1 was intended to see how L2 learners represent and retrieve complex
nominals, a type of morphologically complex word, with reference to native speakers. Of
focal interest is an examination of the dual functions of the language faculty, that is,
storage and computation, in the second language mental lexicon.

Task 1 set out with four research questions. In order to examine the dual functions
of the second language lexicon, I proposed a new interference paradigm; speciﬁcally, I
manipulated the lexical ambiguity of each constituent in complex nominals (i.e., single
meaning-single meaning (SS) such as orange juice, single meaning-multiple meanings
(SM) such as violin bow, multiple meanings-single meaning (MS) such as well water, and
multiple meanings-multiple meanings (MM) such as lead mine).

In essence, the four conditions would differ in terms of processing costs when
there are delayed computations; the items in the SS condition would have the smallest
processing costs while those in the MM condition would have the greatest cost in terms
of processing. The processing costs of the items in both SM and MS conditions would be
in the middle of the SS and the MM conditions. Ll dual-route models suggest that novel
compounds are interpreted via computational routes while familiar compounds are
recognized through retrieval routes (e.g., Baayen, Dijkstra, & Schreuder, 1997; Bertram,
Schreuder, & Baayen, 2000; Caramazza, Laudanna, & Romani, 1988; Pinker & Ullman,
2002). Looking at the four experimental conditions from the perspective of the distinction

between novel and familiar compounds, the four experimental conditions establish a

113

continuum from lexicalized to novel complex nominals. Words in the MM condition are
likely to function as novel compounds, the SS condition is comprised of lexicalized
compounds, and the SM and MS conditions are positioned along the continuum between
the SS and MM conditions. Therefore, I expected that, if complex nominals involve
delayed computations, the frequency-matched and syllable-matched stimuli in the four
conditions would result in a graded interference of their response times and accuracy
scores. Assuming that each constituent plays a role in the interpretation of complex
nominals, I expected that interference would be greatest when both constituents have
multiple meanings, whereas the processing costs would be minimal when each
constituent has a single meaning. Thus, I predicted that participants would respond
slowly and inaccurately in the MM (multiple meanings-multiple meanings) condition (i.e.,
the highest in processing costs), as compared to the SS (single meaning-single meaning)
condition (i.e., the lowest in processing costs).

With respect to potential behavioral differences between NS and NNS groups, I
expected that the predictions related to the four conditions would emerge only among
non-native speakers. Recall that the stimuli used in Task 1 are “familiar” complex
nominals, which can be found in a large English corpus. That is, native speakers could
either store the familiar nominals or compute them very efﬁciently or automatically,
given that, as Gagné and Spalding (2004) state, high-frequency words are more likely to
be stored in a long-term memory (availability) or are easy to process (familiarity). In
other words, high frequency complex nominals may not simply mean the holistic
representations (i.e., availability); rather, the high frequency could mean “familiarity” in

the sense that compounds that are encountered very frequently could enjoy easier

114

computations. In contrast, insofar as NNS groups are concerned, it is not obvious that the
stimuli of Task 1 are lexicalized even among the NNS groups with high L2 proﬁciency,
because high frequency in a large corpus does not necessarily guarantee the lexicalization
of complex nominals in the L2 lexicon. Assuming that both retrieval and computation
mechanisms are part of the L2 lexicon, a failure to retrieve the meaning of a whole
complex nominal necessitates the use of computational routes. Yet, given that L2 learners
are generally less efﬁcient in processing the L2, the NNS groups would yield delayed
computations, which would be reﬂected in the graded interference among the four
experimental conditions.

Diﬂerences between NS and NNS. The ﬁrst research question focused on the
difference between the NS and NNS groups in terms of their behavioral patterns. The
ﬁrst research question asked whether the NS group differs signiﬁcantly from the two
NNS groups, the RH (right-headed) and LH (left-headed) groups. Regarding response
times, as predicted, the result revealed that the NS group responded signiﬁcantly faster in
all four conditions than the two NNS groups. There are two possible interpretations for
this ﬁnding: (1) a storage difference and (2) a processing difference. First, the difference
between the NS group and the two NNS groups may be because the NS group stored or
lexicalized each of the stimuli in Task 1 as a whole unit. In contrast, the NNS groups may
have experienced more interference due to semantic computation in the same way that
NSs likely compute a novel complex nominal when encountering it. Alternatively, a
difference in processing efﬁciency could explain the performance discrepancy between
the NS and NNS groups. That is, the NS group was able to compute the stimuli very

efﬁciently, whereas L2 processing of complex nominals is relatively slower. The results

115

of accuracy scores, in general, replicated the ﬁndings of the differences in response times,
although there were some marginal differences between the NS and RH groups.

Overall, the signiﬁcant difference between the NS and NNS groups seems to
suggest that the two NNS groups relied more on computational processes in order to
recognize the meanings of complex nominals.

No diﬂerence between RH and LH groups. The second research question is
concerned with L1 inﬂuence on morphological processing. Particularly, I questioned
whether the compound structure of L1 5 (particularly, the headedness) affects the
processing of L2 complex nominals.

Complex nominals in English have the order of modiﬁer-head (i.e., orange juice).
L2 learners belonging to the RH group have the identical complex nominal order in their
native language as in English. Thus, 1 expected less interference would be involved in the
construction of meaning of complex nominals. In contrast, the LH group has to switch the
order of complex nominals from head-modiﬁer (L1 structure) to modiﬁer-head (English)
to derive the appropriate meaning for them, assuming that the LH group experience
interference due to the compound structure of their L1. Thus, I expected that, if the L1
plays a role in processing L2 complex nominals, the LH group, including Spanish and
Thai learners, would experience more interference than the RH group, including Chinese
and Korean learners, when processing English complex nominals, which are right-headed.

However, the ﬁndings revealed that there was no signiﬁcant difference between

RH and LH groups in terms of both reaction times and accuracy scores in all four

conditions”. This ﬁnding, in turn, suggests that the L1 compound structure did not play

 

116

an important role in recognizing L2 complex nominals. A possible interpretation for the
lack of difference between RH and LH groups is related to the two groups’ comparable
L2 proﬁciency. Note that the L2 sample populations both have high L2 proﬁciency.
Given that L2 learners become more independent of the L1 as L2 proﬁciency increases
(e.g., Frenck-Mestre & Prince, 1997; Kroll & Curley, 1988), the effect of L1 compound
headedness was minimal with the sample of this study. Alternatively, I would expect to
ﬁnd greater evidence of L1 inﬂuence with learners of lower L2 proﬁciency.

The mixed results of uniform and graded interferences in the NS group. The third
research question examined whether the NS group shows different amounts of
interference in processing the stimuli in the four conditions. In general, two competing
hypotheses have emerged in relation to the dual functions of the lexicon. First, if the NS
group relies on a simple retrieval or employs an efﬁcient computational mechanism, there
would not be graded interferences among the four conditions. A second possibility is
based on computational routes in recognizing complex nominal in which NSs would

show graded interference among the conditions.

 

'9 When Spanish learners are excluded from the LH group, the overall results were the same,
indicating that the Thai and Spanish speakers in this study behaved comparably. For example, when a
MANOVA was run with reaction times of the LH group and the Thai group only, there was no signiﬁcant
difference (W ilk’s Lamdba = .96, F (4, 41) = .38, p = .82). A corresponding analysis of accuracy scores
(i.e., the RH group vs. the Thai group) did not reveal any signiﬁcant difference (Wilk’s Lamdba = .93, F (4,
41) = .77, p = .55). However, since the number of the Spanish participants was small (i.e., less than 10), the
differences between Thai and Spanish learners in terms of processing L2 complex nominals are temporarily

tabled and invite a future investigation.

117

 

The response time data and the accuracy score data provided two conﬂicting
pieces of information. The reaction time data indicated that the NS group showed no
signiﬁcant differences in interference or lexical competition among the conditions
because the response times of the NS group are statistically comparable. A possible
interpretation is that the NS group simply retrieved the familiar complex nominals that
were used in Task 1, which would suggest holistic representations of complex nominals.

Yet, it should also be noted that, although statistically not signiﬁcant, the NS showed the

greatest interference in their response times in the MM condition (M _MM = 1002.3 8),

compared to the other three conditions (M _MS = 946.68; M _SM = 937.20; M _SS = 948.01).

When the differences between the response times in the MM condition and those in the
other conditions were estimated by effect sizes (Cohen’s (1) based the pooled standard
deviation corrected for the correlation, the practical signiﬁcance was large in the contrast
between MM and SM conditions (d= .61). Also the effect sizes was moderate in the
contrast between MM and SS (d = .47) and in the contrast between MM and MS
conditions (d = .48). The results of descriptive statistics and the effect sizes seem to
suggest that NS use computational routes, yet the efﬁcient computational parser of the
NS group reduced the competition between each constituent in the four conditions.

In the accuracy data, the NS group responded more inaccurately to the stimuli in
the MM condition than those in the remaining three conditions. It is plausible to say that
the stimuli in the SM, MS, and SS conditions may be stored as single lexical items. In the
MM condition, however, the accuracy results indicate lexical competitions between
modiﬁer nouns and head nouns. A simple retrieval mechanism failed to explain the

performance of the NS group in MM condition. That is, if the stimuli in the MM

118

condition such as lead mine are represented as individual items, it seems difﬁcult to
provide a rationale for the signiﬁcant inaccuracy found in the MM condition. Rather, NS
are highly likely to rely on semantic computations (i.e., integrating the meanings of
modiﬁers and heads) in interpreting the familiar complex nominals in the MM condition.

In order to explain the results of the NS group, it seems necessary to
accommodate the dual functions of the language faculty (i.e., computation and storage) or
full-parsing accounts. No difference in the response times among the four conditions
suggested holistic representations of familiar complex nominals; at the same time, the
difference in accuracy scores provided evidence of both holistic (SM, MS, and SS
conditions) and discrete representations (MM condition) of complex nominals.
Alternatively, the comparable performance of the NS group may stem from efﬁcient
morphological processing, which is in support of a full-parsing account.

The graded interference of the NNS groups. The fourth research question asks
whether NNS groups show competition effects in processing complex nominals. 1
expected that, since the familiar complex nominals in Task 1 could function as novel
items for non-native speakers, NNS groups would show competition effects across the
four conditions. More speciﬁcally, if computational processes are involved in the
recognition process, graded interferences are expected from the SS condition to the SM
and the MS conditions, and from there to the MM condition because the processing costs
will increase in relation to the number of ambiguous constituents.

The descriptive statistics for both reaction times and accuracy scores suggested
that graded interference emerged in both RH and LH groups, which is consistent with our

prediction based on the difference in processing cost. That is, both RH and LH groups

119

showed the greatest interference in the MM condition, the smallest interference in the SS
condition, and intermediate interference in the MS and SM conditions (i.e., MM > SM,
MS > SS). Inferential statistics revealed that the difference between the MM and SS
conditions is signiﬁcant, but other comparisons did not differ signiﬁcantly. This ﬁnding
does suggest a graded interference across the four conditions.

A full-listing account is unable to explain the graded interference because uniform
interference should be found with the established meanings of complex nominals. In
contrast, both full-parsing and dual-route accounts seem to explain the ﬁndings
appropriately. A full-parsing account could claim that the difference in the processing
loads of the four conditions led to the graded interference among the four conditions. The
NNS groups’ difﬁculty in judging the combinations in the MM condition imply that NNS
employed semantic computations to construct an interpretation and that the computation
loadings or interferences in computation are differentiable in terms of the ambiguity of
each constituent in the combination. The ﬁnding of differentiated competition suggests
that NNSs constructed the interpretation of a complex nominal rather than retrieved the
meaning of the familiar combination. From a dual-route account, there was a competition
between retrieval and computational routes. The retrieval route won for the SS condition
where the processing cost was lowest. The computational route played a major role for
the recognition of complex nominals, resulting in the slow-down in reaction times and the
inaccurate responses. The responses in the SM and MS conditions seem to show a middle

ground of the dual processes.

120

An assumption 1 made for the four experimental conditions is that L2 learners

actually know all the meanings of each multi-meaning word”. This is based on previous
research that has found that L2 learners do activate multiple meanings when processing
L2 homonyms (F renck-Mestre & Prince, 1997). A point can be used to verify whether the
L2 participants of this study who had high level of L2 proﬁciency were able to activate
the multiple meanings of a constituent. The idea of the graded interference is based on
competitions that may be caused by both the meaning activations of each constituent and
the activations of relational structure. If the semantic integration of complex nominals
does not involve competition, there should be similar uniform interference to the case of
the NS group. Given that a resource of competition is the meaning activations of each
constituent in a complex nominal, the graded interference displayed by the NNS groups
indicate multiple meaning activations of each constituent when a constituent has multiple

meanings.

 

20 An issue with regard to the stimuli is whether the homonyms in the complex nominals could
have diverse syntactic categories. In this study, I included items that contain the use of nouns either in the
ﬁrst meaning or in the second meaning or both. For example, the word bear could have a verb reading in
their ﬁrst meaning or have a noun reading in their second meaning. A related issue is that there are two
types of homonyms in terms of meaning dominance; balanced homonyms have two relatively equally
frequent meanings whereas polarized homonyms have a primary meaning which is much more frequent
than the next roost common meaning (e.g., Twilley, Dixon, Taylor, & Clark, 1994). The ﬁne-grained

classiﬁcations of homonyms with respect to meaning dominances remain as a future research.

121

 

Another implication of the graded interference in the NNS groups is that each
constituent is actively engaged in constructing the meaning of complex nominals. If one
of the constituents (i.e., either modiﬁer or head) plays a more dominant role in
interpreting complex nominals, I would expect not only a comparable interference either
between MM and SM or between MM and MS, but also different magnitudes for the MS
and SM condition. The greater interference in the MM condition than in both SM and MS
conditions found in this study favors an account of interactive roles of each constituent in

the interpretive processes of complex nominals. As discussed in Chapter 2, there are two

 

competing theories of establishing semantic relations between two constituents in a
complex nominal in L1 psycholinguistics literature: a schema-based theory (Murphy,
1988, 1990; Wisniewski, 1996) versus CARIN (competition among relations in nominals)
theory (Gagné & Shoben, 1997, 2002). In general, the schema-based theory claims that
the modiﬁer (the ﬁrst noun in English complex nominals) serves as a ﬁller for a slot in
the head noun (the second noun) and thus the head noun plays a key role in establishing
semantic relations of complex nominals; in contrast, the CRAIN theory emphasizes the
role of the modiﬁer (i.e., the ﬁrst noun) in complex nominal processing, arguing that the
modiﬁers trigger/activate the competition of possible semantic relations with the heads
(the second noun). Interestingly, the ﬁndings of this study based on an interference
paradigm provide counter-evidence for both the schema-based theory and the CRAIN
theory. In both theories, the responses to the MM conditions should be comparable to
those to the MS and the SM conditions because either the head or the modiﬁer plays a
key function in constructing a meaningful relationship between the two nouns in a

complex nominal. The greatest interference in the MM condition suggests that the

122

 

construction of a meaning relationship between the constituents should be viewed as an
interactive process, which is not entirely determined by either the head or the modiﬁer.

In general, the ﬁndings suggest that the two NNS groups did use compositional
routes. This, in turn, provides a challenge to storage-based accounts of L2 vocabulary
acquisition, which is dominant in second language lexicon research. Importantly, the
ﬁndings implicated the expansion of L2 vocabulary research to encompass both storage
and computational routes.

The Frequency effect. Frequency is often considered as evidence for the holistic
representations of multi-words such as formulaic sequences (e.g., J iang & Nekrasova,
2007). Thus, the examination of the frequency effect is a way to examine whether both
NSs and NNSs rely more on the retrieval route. The fourth research question addressed
this issue. A general expectation regarding the dual functions in the lexicon is that the
lexicalized items are directly retrieved from the lexicon whereas unlexicalized complex
nominals are derived by semantic computations of two constituents. The frequency effect
was expected for the NS group, given that the stimuli in this study are based on “familiar”
complex nominals that may be lexicalized in native speakers’ minds. At the same time, I
expected that the frequency effect may be moderate among the NS group because even
NSs might interpret some proportion of the familiar combinations through semantic
computation.

I examined the reaction times of the three groups to see if they are predictable
based on the frequency of the whole complex nominals. As predicted, the frequency
effect of complex nominals emerged only in the NS group, but not in the two NNS

groups. The response times of the NS group were predictable with the frequency of

123

 

complex nominals, which replicates the robust ﬁndings of the frequency effects, in
general. Not surprisingly, the NS group demonstrated a strong negative correlation
between their reaction times and the highly frequent complex nominals as a whole. The
ﬁnding suggests that the NS group is likely to represent the high-frequency complex
nominals as single lexical items. However, it should be noted that the effect size

(estimated by the r-squared) was moderate. even in the strongest correlation between the

high—frequency complex nominals and the reaction times of the NS group, r= -.38, r2

= .14, given that the r2 of 0.14 means that 14% of the variance of the frequency measures

is shared with the reaction times of the NS group. In other words, there are many
proportions that the holistic representations do not cover. Thus, the NS group’s behavior
can be characterized better in terms of the dual routes (i.e., storage and computation) of
complex nominal processing.

In the case of the NNS groups, there were mixed results in terms of the

associations between the frequency of complex nominals and the RTs of the NNS

groupsZI. Two noticeable patterns were identiﬁed in terms of the directionality of
correlations. First, both the RH and the LH groups showed negative correlations between
the frequency of the whole complex nominals when the complex nominals are either
high- or mid frequency. The inverse relationship corresponded to the patterns of the NS

group, which seems to suggest that the NNS groups stored small portions of complex

 

21 It may not fair to say that NNSs have the same degree of exposure to complex nominals. This
study follows a conventional method that the frequency is estimated through a corpus that is based on L1

data. A rigorous investigation of frequency effects for NNSs remains as a future research.

124

 

 

nominals holistically. Secondly, the two NNS groups revealed a positive correlation
between the low-frequency complex nominals and their reaction times. The positive
correlations suggest that, in contrast to the NS group, the NNS groups may not store the
low-frequency words as a whole but rather interpret the low-frequency complex nominals
via computations.

Overall, the convergence and divergence of the NS and the NNS groups in Task 1
suggests that the dual functions of language faculty (i.e., storage and computation) play a
key role in the comprehension of complex nominals. There was converging evidence that
both NS and NNS groups not only retrieve the established meanings of complex
nominals but also construct the meanings of complex nominals. At the same time, there
was diverging evidence between the NS group and NNS groups. NS seem to retrieve the
meaning of familiar items from Task 1 from their memory, evidenced by comparable RTs
in the four conditions; however, the NS group’s reliance on the retrieval mechanism was
a small proportion (14%), which was estimated by the relationship to frequency,
suggesting that the efﬁcient morphological parsing of the NS group minimized the graded
interference.

In contrast, the NNS groups tended to rely greatly on the computation route to
establish the meanings of complex nominals, as suggested by the highest interference in
the MM condition; yet, I also found minimal evidence that the NNS groups utilized the
retrieval route to interpret complex nominals, as indicated by the predicted inverse
relationship between reaction times and the frequency of the high- and mid-frequency

complex nominals. Therefore, both the NS and NNS data revealed a continuum of dual

125

mechanisms (i.e., retrieval route and computational route) in interpreting complex

nominals.

4.2. The role of cognitive ability on comprehending complex nominals

Tasks 2 and 3 are concerned with whether cognitive abilities play a role in the
recognition of complex nominals. The semantic integrations were observed to involve the
resolution of competing information, which is caused by the meaning activation of each
constituent in a complex nominal and the activation of relational structures. Given that
the semantic integrations are the part of the architecture of the L1 and L2 lexicon, I
questioned whether those with better cognitive ability may have an advantage in
resolving competitions associated with the recognition of complex nominals over those
with lower cognitive ability. The question was examined in relation to the processing
loads of the test items and the inﬂuence of L1 morphological knowledge.

Working memory capacity. Working memory (WM) capacity involves the
cognitive processes required to perform mental operations on chunks of information and
the temporary storage of those chunks (Baddeley & Hitch, 1974; Cowan, 2005; Hasher,
Lustig, & Zacks, 2007; Kane & Engle, 2003). I reasoned that, if the recognition of
complex nominals involves a computational process, then working memory capacity may
be related to the comprehension processes. That is, it is possible that high WM
individuals may be faster at combining complex nominals than their low WM
counterparts.

The effect of WM is salient when processing loads are costly or when tasks are

cognitively complex. Thus, I expected that the WM effect on the recognition of complex

126

 

nominals would emerge when not only processing cost is high, but when language
background also plays a role. My ﬁrst expectation was that computational costs would be
minimal in the NS group, intermediate in the RH group, and the greatest in the LH group.
The possible difference in processing cost allows us to predict different effects of WM
among the three groups. It is not likely to ﬁnd a WM effect in the NS group because their
interference would be minimal. However, I expected that the effect of WM would be
found in at least the LH group, given the possibility of L1 interference and great
processing costs. As for the RH group, it was predicted that WM capacity would play a
moderate role in processing complex nominals. Secondly, I did not expect that the effect
of WM would be observed in all the conditions. WM capacity generally states that those
who have higher WM capacity could be faster in a higher cognitive process than those
who have lower WM capacity. In our four conditions, the MM condition involves the
highest interference whereas the SS condition involves the lowest interference. Therefore,
I expect that the WM effect may be found only in the MM condition but not in the SS
condition.

The results conﬁrmed that the WM capacity of the LH group played a signiﬁcant
role in predicting both reaction times and accuracy scores only in the high interference
condition. That is, the participants in the LH group who have higher WM scores took less
time and were more accurate in interpreting complex nominals whose constituents are
homonyms than those who have lower WM scores. However, the effect was neither
found in the single meaning-single meaning condition, nor among the NS and RH groups,

as expected.

127

An implication for the ﬁndings of the WM effect on the LH group in the MM
condition is that the LH group made the stronger reliance on computational routes than
the RH group. The difference between the LH and the RH group was not substantial in
Task 1. Note that, as expected, the NS group did not show a signiﬁcant effect of WM
capacity. This, in turn, suggests that the recognition of complex nominals may not be
sufﬁciently difﬁcult for the NS presumably because the NS individuals could use more
simple retrieval mechanisms or compute complex nominals with high efﬁciency. As with
the NS group, the RH group did not reveal a substantial effect of WM capacity on
recognizing the complex nominals. The behavioral pattern of the RH group appears to
indicate that the recognition task invoked less processing costs for the RH group than for
the LH group. Therefore, the WM effect is compounded with L1 background in the sense
that the recognitions of complex nominals invoked greater computational loads to the LH
group than the RH group.

A notable ﬁnding of this study is that the role of WM capacity was observed even
with complex nominals. It is plausible that, at least in the domain of L1 processing,
morphological processing would involve less processing costs or efﬁcient computations.
Thus, the relationship between WM capacity and L1 morphological processing may be a
less interesting question, given that the WM effects were generally observed in complex
cognitive tasks such as sentence processing (e.g., Daneman & Carpenter, 1980;
Harrington & Sawyer, 1992; Juffs, 2004; Waters & Caplan, 1996). Thus, this L2 study
provides an important piece of information regarding the relationship between WM
capacity and morphological processing, information that is not available through an

examination of L1 data alone.

128

 

Inhibitory control ability. Complex nominal processing involves the inhibition of
irrelevant information, as supported by aphasia studies. Particularly, in Task 3, the role of
inhibitory control ability was examined in relation to (1) the effect of processing cost
manipulated by the number of meanings of each constituent and (2) the consequence of
language background. The results indicated that these two factors (i.e., processing cost
and language background) jointly affect the recognition of complex nominals.

In general, inhibitory control ability was expected to play comparable function to
working memory capacity, considering that the similarity between inhibitory control d
ability and the WM capacity such as goal maintenance and inhibition of competing
information (Chiappe & Chiappe, 2007; Kane & Engle, 2003; Long & Prat, 2002). Thus,
I expected that WM and inhibitory control ability effects would reveal a converging piece
of evidence; those who may have better inhibitory control ability would be more efﬁcient
in recognizing complex nominals.

The ﬁndings revealed that the LH group showed a signiﬁcant effect of inhibitory
control ability on comprehending complex nominals whereas the RH group was
associated with a marginal effect of inhibitory control ability. That is, in the LH group,
the individuals with higher inhibitory control ability signiﬁcantly outperformed those
with lower corresponding ability when inhibiting irrelevant information in
comprehending complex nominals. As expected, the NS group did not show the effect of
inhibitory control ability on the recognition of complex nominals. The ﬁndings suggest
that the cost of processing and L1 background jointly affect the recognition of complex

nominals.

129

Despite the different magnitudes of the effect, the LH group and RH groups
demonstrated that inhibitory control processes are involved in complex nominal
processing. This ﬁnding converges with those of aphasia studies (e.g., Badecker, 2001;
Libben, 1998). Therefore, interlanguage data add an important piece of information to
understand the mechanism of complex nominal processing in the same way as the
previous aphasia study did.

A methodological issue is that the measure of inhibitory control in this study is
based on a second language for the NNS groups. In case of L2 Stroop test, the level of L2
proﬁciency is at an issue because research suggests that L2 inhibitory control ability is
compounded with level of L2 proﬁciency (Brauer, 1998; Goldfarb & Tzelgov, 2007; La
Heij, de Bruyn, Elens, Hartsuiker, & Helaha, 1990; Preston & Lambert, 1969; Zied, et al.,
2004). Simply put, it is easy for L2 learners of low L2 proﬁciency to inhibit information
because they can ignore the language information in the Stroop task. For example, one
who does not know Korean shows no interference in doing a version of the Stroop test
where the stimuli are presented in Korean. Thus, an expectation for the L2 Stroop test is
that the effect of L2 inhibitory control ability would be observable with learners of high
L2 proﬁciency. In this regard, one may ask whether the L2 inhibition scores in this study
is a measure of L2 learners’ ability to suppress irrelevant information. Or one may
question whether the NNS participants in this study have “sufﬁcient” levels of L2
proﬁciency that lead to an appropriate measure of L2 inhibitory control ability. One way
to look at this issue is to examine the interaction between inhibitory control ability and
working memory capacity. The relationship between them is expected to be inversely

correlated; that is, those who have higher WM capacity would show less interference or

130

 

greater inhibition in processing information. A recent study by Gass and Lee (in press)
found the predicted inverse relationships between L1 working memory scores and L2
inhibition scores in both the ﬁrst year and the third year Spanish groups. In terms of the
magnitudes, the correlations were not signiﬁcant (r (41) = -.28) for the ﬁrst year Spanish
group but was signiﬁcant for the third year Spanish group (r (65) = -.24). Compared to
Gass and Lee (in press), the NNS participants were higher levels of L2 proﬁciency,
although the differences in the target languages do not allow us to directly compare Gass
and Lee (in press) and this study. In case of this study, when combining both the LH and
the RH group into a NNS group, there was a strong correlation between the L2 inhibition
scores and the WM scores measured by an operation span, r(52) = -.34, p = .01. Note that
not only the directionality is in the expected negative way, but also the magnitude of the
correlation coefﬁcient is stronger. Thus, it seems reasonable that the L2 inhibition scores

in this study did measure L2 learners’ inhibitory control ability.

131

 

CHAPTER SIX - CONCLUSION

The general goals of this dissertation were: ( 1) to investigate whether the dual functions
such as computational routes and retrieval routes can be demonstrated in both the L1 and
L2 lexicon structures, through the performance of NSs and NNSs in recognizing complex
nominals in English, (2) to examine whether the headedness of L1 complex nominal
structures inﬂuences the acquisition of L2 complex nominals, and (3) to explore whether
L2 learners with better cognitive abilities such as inhibitory control and working memory
may have an advantage in the processing of complex nominals. In order to address these
issues, three tasks were carried out with one NS group of English and two NNS groups
whose Ll compound structures differ in terms of headedness.

With respect to the dual functions in the lexicon, this study investigated an
interesting theoretical question of whether a complex nominal such as coﬂ'ee cup is
represented either holistically (i.e., coffee-cup) or discretely (i.e., coffee and cup) or in a
dual fashion (coffee-cup, coffee, and cup). In this regard, complex nominals could be a
window to the understanding of how lexical information is organized. Based on the
previous L1 research of complex nominals, this study examined both NS and NNS
groups, with respect to (l) full-listing models in favor of holistic representations, (2) full-
parsing models in support of discrete representations, and (3) dual-route models
accommodating both full-listing and full-parsing models. The ﬁndings of L1 and L2
speakers in this study support both full-parsing and dual-route models; the NS group

appeared to have a balanced use of retrieval and computational routes whereas, while

132

 

exhibiting moderate evidence for the use of retrieval routes, the NNS groups seemed to
rely more on computational routes.

A theoretical implication for second language lexicon research is that both
computation and retrieval are equally important to our understanding of the architecture
of the second language lexicon. The ﬁndings of the current study indicated that L2
learners may store individual words but have fewer complex nominals stored as a whole.
Despite this unbalanced display of dual functions in the second language lexicon,
importantly, L2 learners were able to recognize complex nominals correctly, as
evidenced by the accuracy scores of this study. The correct recognitions of L2 learners
indicate the use of computational routes or semantic integrations/computations in
interpreting complex nominals. However, L2 learners are less efﬁcient in integrating the
two constituents than NSs, showing greater processing loads in the multiple meanings —
multiple meanings condition of complex nominals. Therefore, this study provides a
strong piece of empirical evidence that computation is the structure of the second
language lexicon. At the same time, this study provides strong support for storage-based
L2 lexicon research. A clear disadvantage for L2 learners is that they possess a smaller
number of stored complex nominals. In order to compensate for this disadvantage, L2
learners have to rely more on computational routes. Thus, it is important to study how L2
learners can increase the amount of stored information in their second language lexicon
and to examine whether L2 learners beneﬁt from the use of retrieval routes.

Another implication for L2 lexicon research is that L1 morphological knowledge
affects the structure of the second language lexicon. At the level of descriptive statistics,

the left-headed NNS group showed slower processing of English complex nominals than

133

the right-headed NNS group, presumably due to the difference in the headedness of L1
and L2 complex nominals. Yet, it is also notable that, when comparing L2 learners of
comparably high L2 proﬁciency, the difference between the LH and the RH was not
signiﬁcant at the level of inferential statistics, suggesting that, as L2 proﬁciency increases,
the inﬂuence of L1 headedness may lessen.

Furthermore, the L1 morphological structures of L2 learners were associated with
the effect of cognitive variables on the recognition of complex nominals. The effects
were observed in conjunction with the processing loads of the test items and L1
backgrounds. Although the behavioral differences in the LH and the RH groups were not
signiﬁcant, individuals with high cognitive abilities in the LH group have greater beneﬁts
in processing English complex nominals than those in the RH group. An implication of
this ﬁnding is that the LH group yield greater interferences in processing L2 complex
nominals which are right-headed, indicating an L1 inﬂuence on L2 morphological
knowledge.

The current study leaves open several areas for future investigation. The ﬁrst
research area is related to L1 inﬂuence. Particularly, we should consider the more ﬁne-
grained classiﬁcations of L1 morphological structures in investigating the inﬂuence of L1
morphological structure on processing L2 complex nominals. For example, this study
collapses Spanish and Thai into one group because the two languages are left-headed in
terms of L1 morphological structures of complex nominals. Despite this commonality
between the two languages, as described earlier, the two languages somewhat differ in
terms of structures of complex nominals. Speciﬁcally, in Spanish it is more productive to

insert a preposition between two constituents in a complex nominal whereas in Thai,

134

 

complex nominals are made up without a preposition. More importantly, in the case of
Spanish, a preposition can have semantic functions to clarify the relationship between
two nouns. In Spanish, although the preposition de is widely used to link two constituents
in a complex nominal, a preposition such as para can also be used to express the intended
use of the head noun such as jarra para cerverza ‘beer mug’ (Bradley & Mackenzie,
2004). The examples from Spanish show clearly that a preposition is not a simple
syntactic link between two nouns, but has semantic functions to clarify the relation
between two constituents. In contrast, in such languages as Thai and Vietnamese, linking
prepositions are not used in the formation of complex nominals. Therefore, it is
worthwhile to examine whether the second language acquisition of complex nominals
differs depending on whether the L1 structure of complex nominal includes linking
prepositions.

A related research domain is that noun-noun compounding is a highly productive
system in some languages such as German whereas the compounding is relatively less
common in other languages such as Spanish. It may be possible that German speakers
have an advantage in learning compounds in other languages. In this regard, it would be
interesting to examine whether the morphological productivity of the L1 affects the
acquisition of L2 compounds.

The second future research area is to look at the role of L2 proﬁciency in
recognizing L2 complex nominals. This study only examined L2 learners who have high
L2 proﬁciency. The ﬁndings of this study imply that L2 learners with low L2 proﬁciency

may depend far more on computational mechanisms than those with high L2 proﬁciency.

135

 

Yet, this is an empirical question and it is also important to look at how exactly L2
learners who have low proﬁciency process L2 complex nominals.

The third domain of future research is to examine the difference in modality. This
study is based on the recognition of L2 complex nominals. Obviously, production data
broadens our understanding of how L2 learners acquire L2 complex nominals. For
example, the marginal difference between the LH and the RH groups in this study may be
magniﬁed in a study of complex nominal production.

The fourth area for a future investigation is to examine the effect of context on the
interpretations of complex nominals. As with the case of this study, it is important to see
how complex nominals can be interpreted on the base of the meanings or semantic
representations of each constituent. At the same time, it is also important to examine how
L2 learners are inﬂuenced by contexts or discourse in interpreting complex nominals, as
investigated in much L1 research (Gagne & Murphy, 1996; Gerrig & Bortfeld, 1999;
Gerrig & Murphy, 1992).

This dissertation has theoretical value for the broader domain of language science.
The contrasting ideas of retrieval and computation in the lexicon are related to the
theoretical division between symbolist/generative and connectionist accounts in language
science. Overall, a generative account supports the idea of computations whereas a
connectionist favors the idea of retrieval within the lexicon. However, this dissertation
illustrates that both retrieval and computation are not mutually exclusive; rather, both
notions are integral in providing an appropriate account of the architecture of both the L1
and L2 lexicon. The results of this study suggest that both L1 and L2 lexicons may be

structured to maximize both retrieval mechanisms and the computational system. Beyond

136

 

SLA research, this dissertation ﬁlled a gap in the complex nominal research base, in
which L2 ﬁndings were not reﬂected. The interlanguage data is particularly valuable
because the behavioral patterns of L2 learners reﬂect the continuum between lexicalized
and novel complex nominals. The dissertation demonstrated that L2 learners rely on the
dual functions of the lexicon in interpreting complex nominals, which corresponded to
the ﬁndings of much of previous L1 research, including L1 psycholinguistics (e.g., Gagné,
2001; Libben, 1998), L1 neuro-imaging studies (e.g., F iorentino & Poeppel, 2007) and
L1 aphasia studies (e.g., Badecker, 2001).

Furthermore, interlanguage data enable us to identify the role of cognitive
variables on the interpretation of familiar complex nominals, which is difﬁcult to observe
in L1 studies. L1 studies have rarely tested the effect of cognitive variables on the
recognition of complex nominals, partly because the efﬁcient processing of N85 fails to
demonstrate the effect under normal circumstances. Thus, most evidence for the role of
cognitive variables on complex nominal processing is based on aphasia studies. The
current study, based on L2 learners’ performance in a normal environment, identiﬁed the
effect of inhibitory control abilities and working memory capacity on recognizing
complex nominals, due to the delayed processing mechanism of L2 learners. Therefore,
this study provides an important piece of empirical evidence for understanding the
relationship between complex nominal processing and cognitive abilities.

Also, this dissertation has theoretical relevance for theories of L2 acquisition.
Within the domain of L2 vocabulary learning, this dissertation highlights the importance
of the dual functions in the L2 lexicon. Despite the fact that many SLA researchers have

pointed out that connectionist approaches (i.e., retrieval-preference) cannot explain the

137

entire process of second language acquisition, this research clarifyies the necessity of
both symbolist/ generative (computation-preference) and connectionist accounts in a
better understanding of SLA (Ellis, 2003; Gregg, 2003; Hulstijn, 2002; O’Grady, 2003;
Sharwood Smith, 1994). The ﬁndings of this study showed that L2 learners were more
dependent upon computational routes, suggesting a call for in-depth investigations of
computations in L2 lexicon research. At the same time, this study added the ﬁnding that
L2 learners are not efﬁcient in using retrieval routes. Assuming that both computation
and retrieval mechanisms are parts of the L2 lexicon, this dissertation also illustrates why
the study of the storage of L2 multi-morphemic words (e.g., complex nominals and
formulaic sequences) in the L2 lexicon matters.

This dissertation of the L2 lexicon provides an insight into the two key concepts --
storage and computation -- of understanding the language faculty (e.g., Nooteboom,
Weerman, & Wijnen, 2002), which are also fundamental considerations in explaining L2
acquisition. In this regard, this study illustrates how L2 empirical research can contribute
to both L1 and L2 theoretical literature through testing and validating theoretical

COIIStI'UCtS.

138

 

APPENDIX A

The sensible test items used in Task 1

 

 

SS SM MS MM

1 ankle bone apple peel arm chair arch bridge

2 battle scar army rank bear claws bank staff

3 beach towel art fair brush handle bat box

4 blood cell candy jar cloth napkin card pack

5 butter bread coffee pot corn soup clock tower

6 career plan diaper rash desert sand cow hide

7 church bells dog tag down coat cricket sound

8 city museum ﬁsh scales ear plugs egg noodles

9 clay brick ﬂower stem fast day ﬂy trap
10 engine noise glass pitcher job depression ground pipes
11 female teacher grass root key fact jet plane
12 ﬁnger nail hand drill liver damage lead mine
13 fog machine laugh lines mail route light pole
14 fox fur maximum ﬁne novel writer mint gum
15 goat milk moon rock pen name pet cat
16 mango juice morning prayers phone button pine bark
17 melon seed mountain soil piano pedals punch bowl
18 motor bike navy ﬂeet pie pieces race policy
19 mud bath right angle pool party rear post
20 night gown soccer fan saw blade rose bush
21 oil country steel case top quality rubber band
22 patient proﬁle travel log truck wheel spray wax
23 pocket knife violin bow well water tire size
24 tree branch zebra stripes wind gauge yard stick

 

139

 

The non-sensible test items used in Task 1

APPENDIX B

 

1 boom bound area dart bargain bud advance comer
2 cape strain ash comment beer sole antique hang
3 capital grave bang street board buffet approach cold
4 compound dove calf doubt care hatch beat poster
5 cube wound clip beneﬁt class bay bend door
6 felt boil count hood clinic sack carving fancy
7 ﬁt fuse cue trash comb pile century litter
8 fold palm dam printer contrast hail close news
9 guy foil lie missile council trace concern beam
10 hold chop meal police distance mill credit claim
11 host lap poker ivory engine stole depth soak
12 kid ﬂush pound anxiety event rest hammer drain
13 mark diet rent chew lecture low holiday ofﬁce
14 might ﬂat rung penalty marine bass kit bid
15 mole pick scrub Mounds medicine row mess shade
16 port drove seal surgery menu die opinion maple
17 shark pupil slip tide rabbit console painting blast
18 stake second sow layer river ﬁrm place smash
19 stunt junk spell criminal secret loaf price pass
20 swallow present tap visitor shell counter smell stain
21 tense pat tear infant student squash spin treat
22 till refuse temple scrap suite bit stamp wrong
23 vault blow toast ink tourist heel tiger practice
24 will stable toll beauty window riﬂe wheat star

 

140

 

APPENDIX C

The instruction in Task 1

 

 

INSTRUCTIONS
Welcome! ! !

In this experiment, you will be asked to judge whether a phrase made up of TWO words
is acceptable in English. For example, you will see a phrase like "computer printer",
which is OK in English. Also you will see a phrase like "water tape", which may not be
0k in English.

Here is the procedure that you will see in this experiment.

(1) You will see "+" sign, which shows that a two—word phrase will show up very soon.
You don't need to press any button at this moment, but please be prepared for responding
to a two-word phrase followed by "+".

(2) When you see the phrase, please press either "Yes" or "No" button.

(3) If you don't know either/both of the meaning(s) of the phrase, please press
SPACEBAR. For example, when seeing "computer printer", please press the
SPACEBAR only if you don't know the meaning(s) of either "computer" or "printer" or
both.

IMPORTANT: If you know the meanings of both words in a two-word phrase (i.e.,
"computer" and "printer" in "computer printer"), please press "Yes" or "NO", but do NOT
press the spacebar.

Ok, Let's begin with some practices!!!

Press the SPACEBAR when you are ready to begin.

 

141

 

The working memory test

APPENDIX D

 

 

 

 

 

Operations Correct responses Recalls
A block of2

(2X2)+3=6 No 84

(6-5)+1=2 Yes 49
A block of2

(4X3)-3=9 Yes 65

(7+5)-6=6 No 39
A block of3

(2X2)+5=8 No 46

(1X3)X2=6 Yes 73

(9-7)X2=4 Yes 23
Ablock of3

(8-5)X2=9 No 17

(4X1)X2=8 Yes 32

(3X5)-9=5 No 95
A block of4

(8-6)X3=4 No 78

(9-2)-5=3 No 64

(9-8)+1=2 Yes 37

(7-5)+1=3 Yes 19
Ablock of4

(3X2)+3=7 No 54

(9-4)-3=2 Yes 83

(2+1)+3=4 No 97

(7-4)+3=5 No 62
Ablock of5

(3+2)+5=9 No 68

(2+1)+2=5 Yes 25

(2+1)X2=3 No 74

(6+2)-5=2 No 13

(3X3)-4=5 Yes 92
A block of5

(9+1)—4=6 Yes 53

(2X2)X =8 No 89

(4X2)-1=7 Yes 47

(9-2)-5=2 Yes 35

(2+1)X3=4 No 71

 

142

 

 

BIBLIOGRAPHY

Abel, B. (2003). English idioms in the ﬁrst language and second language lexicon: A
dual representation approach. Second Language Research, 19, 329-358.

Adams, V. (1973). An Introduction to Modern English Word-Formation. London:
Longman.

Altenberg, B. (1998). On the phraseology of spoken English: The evidence of recurrent
word combinations. In A. P. Cowie (Ed.), Phraseology: Theory, analysis and
applications (pp.101—122). Oxford: Oxford University Press.

Anderson, M. C., & Green, C. (2001). Suppressing unwanted memories by executive
control. Nature, 410, 366—369.

Andrews, S. (1986). Morphological inﬂuences on lexical access: Lexical or nonlexical
effects? Journal of Memory and Language, 25, 726-740.

Andrews, 8., Miller, B., & Rayner, K. (2004). Eye movements and morphological
segmentation of compounds: There is a mouse in mousetrap. European Journal of
Cognitive Psychology, 16, 285-311.

Baayen, H., Dijkstra, T., & Schreuder, R. (1997). Singulars and plurals in Dutch:
Evidence for a parallel dual route model. Journal of Memory and Language, 3 7,
94-117.

Baayen, R. H., Davidson, D. J ., & Bates, D. M. (2008). Mixed effects modeling with
crossed random effects for subjects and items. Journal of Memory and Language
59, 390—412.

Baddeley, A. D. (2003). Working memory and language: An overview. Journal of
Communication Disorders, 36, 189-208.

Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G.A. Bower (Ed.), Recent
Advances in Learning and Motivation (Vol. 8., pp. 47—89). New York: Academic
Press.

Badecker, W. (2001). Lexical composition and the production of compounds: Evidence
from errors in naming. Language and Cognitive Processes, 16, 337-366.

Barcroft, J. (2004). Effects of sentence writing in L2 lexical acquisition. Second
Language Research, 20, 303—334.

Barcroft, J ., & Sommers, M. (2005). Effects of acoustic variability on second language
vocabulary Ieaming. Studies in Second Language Acquisition, 27, 387-414.

143

Bauer, L. (1978) The Grammar of Nominal Compounding, With Special Reference to
Danish, English, and French. Odense: Odense University Press.

Bauer, L. (2001). Morphological Productivity. Cambridge, UK: Cambridge University
Press.

Beck, M-L. (1998). L2 acquisition and obligatory head movement: English-speaking
learners of German and the local impairment hypothesis. Studies in Second
Language Acquisition, 20, 311-348.

Bertram, R., Schreuder, R., & Baayen, R. H. (2000). The balance of storage and
computation in morphological processing: The role of word formation type,
afﬁxal homonymy, and productivity. Journal of Experimental Psychology:
Learning, Memory, & Cognition, 26, 489-511.

Blanken, G. (2000). The production of nominal compounds in aphasia. Brain and
Language, 74, 84-102.

Bley-Vroman, R., & Masterson, D. (1989). Reaction time as a supplement to
grammaticality judgments in the investigation of second language learners’
competence. University of Hawai ’i Working Papers in ESL, 8, 207-245.

Bt‘irjars, K, & Burridge, K. (2001). Introducing English grammar. London: Arnold.

Bradley, P. T., & Mackenzie, I. (2004). Spanish: an Essential Grammar. London:
Routledge.

Brauer, M. (1998). Stroop interference in bilinguals: The role of similarity between the
two languages. In A. F. Healy & L. E. Baume, Jr. (Eds), Foreign language
learning (pp. 317—337). Mahwah, NJ: Erlbaum.

Butterworth, B. (1983). Lexical representation. In B. Butterworth (Ed.), Language
production, Vol. 2 (pp. 257-294). London: Academic Press.

Bybee, J. L. (1995). Regular morphology and the lexicon. Language and Cognitive
Processes, 10, 425-455.

Caramazza, A., Laudanna, A., & Romani, C. (1988). Lexical access and inﬂectional
morphology. Cognition, 28, 297-332.

Chiappe, D. L., & Chiappe, P. (2007). The role of working memory in metaphor
production and comprehension. Journal of Memory and Language, 5 6, 172—188.

Clark, E. V. (1993). The lexicon in acquisition. New York, NY: Cambridge University
Press.

Clark, E. V., Gelamn, S.A. & Lane, NM. (1985) Compound nouns and category
structure in young children. Child Development, 56, 84-94.

144

Clark, H. (1973). The language-as-ﬁxed effect fallacy: a critique of language statistics in
psychological research. Journal of Verbal Learning and Verbal Behavior, 12,
335-359.

Clashen, H. (1999) Lexical entries and rules of language: A multidisciplinary study of
German inﬂection. Behavioral and Brain Sciences, 22, 991-1060.

Conway, A. R. A., Kane, M. J ., Bunting, M. F ., Hambrick, D. Z., Wilhelm, O., & Engle,
R. W. (2005). Working memory span tasks: A methodological review and user’s
guide. Psychonomic Bulletin & Review, 12, 769-786.

Coolen, R., Van Jaarsveld, H. J ., & Schreuder, R. (1991). The interpretation of isolated
novel nominal compounds. Memory & Cognition, 19, 341-352.

Costello, F. J ., & Keane, M. T. (2001). Testing two theories of conceptual combination:
Alignment versus diagnosticity in the comprehension and production ofcombined
concepts. Journal of Experimental Psychology: Learning, Memory, and Cognition,
27, 255-271.

Cowan, N. (2005). Working memory capacity. New York, NY: Psychology Press.

Cowan, N., Chen, Z., & Rouder, J. N. (2004). Constant capacity in an immediate serial—
recall task: A logical sequel to Miller (1956). Psychological Science, 15, 634-640.

Daneman, M., & Carpenter, P. (1980). Individual differences in working memory and
reading. Journal of Verbal Learning and Verbal Behavior, 19, 450—466.

Davies, M. (2009). The 385+ Million Word Corpus of Contemporary American English
(1990-2008+): Design, architecture, and linguistic insights. International Journal
of Corpus Linguistics, I 4, 159-190.

de Bot, K. (1992) A bilingual processing model: Levelt’s speaking model adapted.
Applied Linguistics, 13, 1-24.

de Bot, K., Paribakht, T.S., & Wesche, M. (1997). Towards a lexical processing model
for the study of second language vocabulary acquisition: Evidence from ESL
reading. Studies in Second Language Acquisition, I 9, 309-3 29.

De Groot, A. M. B., & Keijzer, R. (2000). What is hard to learn is easy to forget: The
roles of word concreteness, cognate status, and word frequency in foreign-
language vocabulary learning and forgetting. Language Learning, 50, 1—56.

Delazer, M., & Semenza, C. (1998). The processing of compound words: a study
in aphasia. Brain and language, 61, 54-62.

Downing, P. (1977). On the creation and use of English compound nouns. Language, 53,
810-842.

145

Dressler, W. U. (2006). Compound types. In G. Libben & G. Jarema (Eds), The
representation and processing of compound nouns (pp. 23—44). Oxford: Oxford
University Press.

Dressler, W. U., & Denes, G. (1989). Word formation in Italian-speaking Wemicke’s and
Broca’s aphasics. In W.U. Dressler & J. Stark (Eds.), Linguistic Analyses of
Aphasic Language (pp. 69—88). New York: Springer.

Ellis, N. (2003). Constructions, chunking, and connectionism: the emergence of second
language structure. In C. Doughty and M. Long (Eds), Handbook of second
language acquisition (pp. 63-103). Oxford: Blackwell.

Ellis, N. (2005), At the interface: Dynamic interactions of explicit and implicit language
knowledge, Studies in Second Language Acquisition, 27, 305—3 52.

Estes, Z., & Glucksberg, S. (2000). Interactive property activation in conceptual
combination. Memory & Cognition, 28, 28-34.

Estes, Z., & Jones, L. L. (2006). Priming via relational similarity: A COPPER HORSE is
faster when seen through a GLASS EYE. Journal of Memory and Language,
55, 89-101.

Fiebach, C. J ., Schlesewsky, M., & Friederici, A. D. (2001). Syntactic working memory
and the establishment of ﬁller-gap dependencies: Insights from ERPs and fMRI.
Journal of Psycholinguistic Research, 30, 321—338.

F iorentino, R., & Poeppel, D. (2007). Compound words and structure in the lexicon.
Language and Cognitive Processes, 22, 1-48.

Frauenfelder, U.H., & Schreuder, R. (1992). Constraining psycholinguistic models of
morphological processing and representation: The role of productivity. In G.E.
Booij & J. van Marle (Eds), Yearbook of morphology 1991, (pp. 165-183).
Dordrecht: Foris.

Frenck-Mestre, C., & Prince, P. (1997). Second language autonomy. Journal ofMemory
and Language, 37, 481—501.

Gagné, C. L. & Shoben, E. J. (1997). The inﬂuence of thematic relations on the
comprehension of modiﬁer-noun combinations. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 23, 71-87

Gagné, C. L. & Shoben, E. J. (2002). Priming relation in ambiguous noun-noun
combinations. Memory & Cognition, 30, 637-643.

Gagné, C. L. (2001). Relation and lexical priming during the interpretation of noun-noun
combinations. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 27, 236-254.

146

Gagné, C. L. (2002). Lexical and relational inﬂuences on the processing of novel
compounds. Brain and Language, 81, 723-735.

Gagné, C. L., & Murphy, G. L. (1996). Inﬂuence of discourse context on feature
availability in conceptual combination. Discourse Processes, 22, 79-101.

Gagné, C. L., & Spalding, T. (2006a). Conceptual Combination: Implications for the
Mental Lexicon. In G. Libben, & G. Jarema (Eds). The representation and
processing of compound words (pp. 145-168). Oxford University Press.

Gagné, C. L., & Spalding, T. (2006b). Using conceptual combination research to better
understand novel compound words. SKASE Journal of Theoretical Linguistics, 3,
9-16.

Gagné, C. L., & Spalding, T. (2006c). Relation availability was not confounded with
familiarity or plausibility in Gagné and Shoben (1997): Comment on Wisniewski
and Murphy (2005). Journal of Experimental Psychology: Learning, Memory,
and Cognition, 32,1431-1437.

Gagné, C. L., Spalding, T. L., & Ji, H. (2005). Re-examining evidence for the use of
independent relational representations during conceptual combination. Journal of
Memory and Language, 53, 445—455.

Gagné, C., & Spalding, T. (2004). Effect of discourse context and modiﬁer relation
frequency on conceptual combination. Journal of Memory and Language,
50, 444-455.

Gass, S. & Lee, J (in press). Working memory capacity, Stroop interference, and
proﬁciency in a second language.

Gass, S. (1988). Second language vocabulary acquisition. Annual Review of Applied
Linguistics, 9, 92-106.

Gass, S. (2001). Sentence matching: a re-examination. Second Language Research, I 7,
421-441.

Gass, S., & Selinker, L. (2008). Second Language Acquisition: An Introductory Course
(3rd Edition). New York: Routledge/Taylor Francis.

Gemsbacher, M. A., & Faust, M. (1991). The mechanism of suppression: A component
of general comprehension skill. Journal of Experimental Psychology: Learning,
Memory, and Cognition, I 7, 245-262.

Gerrig, R. J ., & Bortfeld, H. (1999). Sense creations in and out of discourse contexts.
Journal of Memory and Language, 41 , 457-468.

Gerrig, R. J ., & Murphy, G. L. (1992). Contextual inﬂuences on the comprehension of
complex concepts. Language and Cognitive Processes, 7, 205—230.

147

Giraudo, H., & Grainger, J. (2000). Effects of prime word frequency and cumulative root
frequency in masked morphological priming. Language and Cognitive Processes,
15, 421-444.

Gleitman, L., & Gleitman, H. (1970). Phrase and paraphrase: Some innovative uses of
language. New York: Norton.

Goldfarb, L., & Tzelgov, J. (2007). The cause of the within-language Stroop superiority

effect and its implications. The Quarterly Journal of Experimental Psychology,
60, 179-185.

Green, D. W. (1998). Mental control of the bilingual lexico-semantic system,
Bilingualism: Language and Cognition, 7, 67—8 1.

Gregg, K. R. (2003). The state of emergentism in SLA. Second Language Research, 19,
95-128.

Gunter, T. C., Wagner, S. & F riederici, A. D. (2003). Working memory and lexical
ambiguity resolution as revealed by ERPs: A difﬁcult case for activation theories.
Journal of Cognitive Neuroscience, 15, 643-657.

Harrington, M., & Sawyer, M. (1992). L2 working memory capacity and L2 reading skill.
Studies in Second Language Acquisition, 14, 25-38.

Hasher, L., Lustig, C., & Zacks, R. T. (2007). Inhibitory mechanisms and the control of
attention. In A. R. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N. Towse
(Eds.), Variation in working memory (pp. 227-249). New York, NY: Oxford
University Press.

Hegarty, M., Shah, P., & Miyake, A. (2000). Constraints on using the dual-task
methodology to specify the degree of central executive involvement in cognitive
tasks. Memory and Cognition, 28, 376-385.

Hitch, G. J., Towse, J. N., & Hutton, U. (2001).What limits children’s working memory
span? Theoretical accounts and applications for scholastic development. Journal
of Experimental Psychology: General, 130, 184—198.

Hulstijn, J. (2002). Towards a uniﬁed account of the representation, processing and
acquisition of second language knowledge, Second Language Research, 18, 193—
223.

Hulstijn, J ., & Laufer, B. (2001). Some empirical evidence for the involvement load
hypothesis in vocabulary acquisition. Language Learning, 51, 539—558.

Inhoff, A. W., Briihl, D., & Schwartz, J. (1996). Compound word effects differ in reading,
on-line naming, and delayed naming tasks. Memory and Cognition, 24, 466-476.

148

 

Inhoff, A. W., Radach, R. & Heller, D. (2000). Complex Compounds in German:
Interword Spaces Facilitate Segmentation but Hinder Assignment of
Meaning. Journal of Memory and Language, 42, 23-50.

Iwasaki, S., Ingkaphirom, P., & Horie, I. P. (2005). A reference grammar of Thai. New
York: Cambridge University Press.

Jackendoff, R. (1997). The architecture of the language faculty. Cambridge, MA: MIT
Press.

Jared, D., & Kroll, J. F. (2001). Do bilinguals activate phonological representations in
one or both of their languages when naming words? Journal of Memory and
Language, 44, 2—3 1.

Jarema, G. (2006). Compound representation and processing: A cross-language
perspective, In G. Libben & G. Jarema (Eds.), The representation and processing
of compound words, (pp. 45-70). New York, NY: Oxford University Press.

J espersen, O. (1942). A Modern English Grammar on Historical Principles, IV.
Copenhagen, Denmark: Munksgaard.

Jespersen, O. (1968). The Philosophy of Grammar. London: Allen and Unwin.

J iang, N. (2000). Lexical representation and development in a second language. Applied
Linguistics, 21, 47-77.

J iang, N. (2002). Form-meaning mapping in vocabulary acquisition in a second language.
Studies in Second Language Acquisition, 24 , 6 1 7-63 7.

J iang, N. (2004). Semantic transfer and its implications for vocabulary teaching in a
second language. The Modern Language Journal, 88, 416-432.

J iang, N., & Nekrasova, T. (2007). The processing of formulaic sequences by second
language speakers. The Modern Language Journal, 91, 433-445.

9

Juffs, A. (2001). Discussion: Verb classes, event structure, and second language learners
knowledge of semantics-syntax correspondences. Studies in Second Language
Acquisition, 23, 305-313.

J uffs, A. (2004). Representation, processing and working memory in a second language.
Transactions of the Philological Society, 102, 199-226.

Juhasz, B. J., Inhoff, A., & Rayner, K. (2005). The role of interword spaces in the
processing of English compound words, Language and Cognitive
Processes, 20, 291—316.

149

Juhasz, B. J., Starr, M. S., Inhoff, A.W., & Placke, L. (2003). The effects of morphology
on the processing of compound words: Evidence from naming, lexical decisions

and eye ﬁxations. British Journal of Psychology, 94, 223-244.

Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual
differences in working memory. Psychological Review, 99, 122—149.

Kane, M. J ., & Engle, R. W. (2003). Working-memory capacity and the control of
attention: the contributions of goal neglect, response competition, and task set to
Stroop interference. Journal of Experimental Psychology: General, 132, 47-70.

Kane, M. J ., Bleckley, M. K., Conway, A. R. A., & Engle, R. W. (2001). A controlled-
attention view of working-memory capacity. Journal of Experimental Psychology:
General, 130, 169—183.

Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T.W., & Engle, R. W.
(2004). The generality of working memory capacity: A latent-variable approach

to verbal and visuo-spatial memory span and reasoning. Journal of Experimental
Psychology: General, 133, 289-217.

Katamba, F. (1993). Morphology. London: Macmillan.

Kay, P., & Zimmer, K. (1976), On the Semantics of Compounds and Genitives in English,
Sixth California Linguistics Association Conference Proceedings (pp. 29—35).

Kiefer, M., Ahlegian, M., & Spitzer, M. (2005). Working memory capacity, indirect
semantic priming and Stroop interference: Pattern of interindividual prefrontal
performance differences in healthy volunteers. Neuropsychology, 19, 332-344.

King, J ., & Just, M. A. (1991). Individual differences in syntactic processing: the role of
working memory. Journal of Memory and Language, 30, 580-602.

Kroll, J. F., & Curley, J. (1988). Lexical memory in novice bilinguals: The role of
concepts in retrieving second language words. In M. M. Gruneberg, P. E. Morris,
& R. N. Sykes (Eds). Practical Aspects of Memory: Current Research and Issues,
Vol. 2. (pp. 389-395). London: John Wiley & Sons.

La Heij, W., de Bruyn, E., Elens, E., Hartsuiker, R., & van Schelven, L. (1990).
Orthographic facilitation and categorical interference in a word translation variant
of the Stroop task. Canadian Journal of Psychology, 44, 76-83.

Laudanna, A., & Burani, C. (1985). Address mechanisms to decomposed lexical
entries. Linguistics, 23, 775—792.

Levi, J. (1978). The syntax and semantics of complex nominals. New York, NY:
Academic Press.

150

Libben, G. (1998). Semantic transparency in the processing of compounds:
Consequences for representation, processing, and impairment. Brain and
Language, 61 , 30-44.

Libben, G. (2006). Why study compounds: An overview of the issues. In G. Libben, & G.
Jarema (Eds.), The representation and processing of compound words (pp. 1—21).
Oxford, UK: Oxford University Press.

Libben, G., Gibson, M., Yoon, Y. B., & Sandra, D. (2003). Compound fracture: The role
of semantic transparency and morphological headedness. Brain and Language, 84,
50-64.

Libben, G., Derwing B., & de Almeida, R. G. (1999). Ambiguous novel compounds and
models of morphological parsing. Brain and Language, 68, 378-386.

Liceras, J. M. & Diaz, L. (2000). Triggers in L2 acquisition: The case of Spanish N-N
compounds. Studia Linguistica, 54, 197-211.

Lima, S. D., & Pollatsek, A. (1983). Lexical access via an orthographic code? The basic
orthographic syllabic structure (BOSS) reconsidered. Journal of Verbal Learning
and Verbal Behavior, 22, 310-332.

Long, D. L., & Prat, C. S. (2002). Working memory and Stroop interference: An
individual differences investigation. Memory and Cognition, 30, 294-301.

MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrated
review. Psychological Bulletin, 109, 163-203.

MacLeod, C. M. (1992). The Stroop task: The “golden standard” of attentional measures.
Journal of Experimental Psychology: General, 121, 12-14.

Manelis, L., & Tharp, D. A. (1977). The processing of afﬁxed words. Memory and
Cognition, 5, 690-695.

Marchand, H. (1960). The categories and types of present-day English word-formation.
Wiesbaden, Germany: Otto Harrassowitz.

Marian, V., Spivey, M., & Hirsch, J. (2003). Shared and separate systems in bilingual
language processing: Converging evidence from eyetracking and brain imaging.
Brain and Language, 86, 70—82.

Marslen-Wilson, W., Tyler, L. K., Waksler, R., & Older, L. (1994). Morphology and
meaning in the English mental lexicon. Psychological Review, 101, 3—33.

McClelland, J. L., Rumelhart, D. E., and the PDP research group. (1986). Parallel
distributed processing: Explorations in the microstructure of cognition. Volume
11. Cambridge, MA: MIT Press.

151

McQueen, J. M., & Cutler, A. (1998). Morphology in word recognition. In A. Zwicky &
A. Spencer (Eds.), The handbook of morphology (pp. 406-427). Oxford:
Blackwell.

Miyake, A., Just, A.M. & Carpenter, P. A. (1994) Working memory constrains on the
resolution of lexical ambiguity: Maintaining multiple representations in neutral
contexts. Journal of Memory and Language, 33, 175-202.

Monsell, S. (1985). Repetition and the lexicon. In A. W. Ellis (Ed.), Progress in the
psychology of language, Vol. 2 (pp. 147-195). Hillsdale, NJ: Lawrence Erlbaum.

Monsell, S. (1991). The nature and locus of word frequency effects in reading. In D.
Besner & G. W. Humphreys (Eds.), Basic processes in reading: Visual word
recognition. Hillsdale, NJ: Erlbaum.

Murphy, G. L. (1988). Comprehending complex concepts, Cognitive Science, 12, 529-
562.

Murphy, G. L. (1990). Noun phrase interpretation and conceptual combination. Journal
of Memory and Language, 29, 259—288.

Napoli, D. (1996). Linguistics. New York: Oxford University Press.

Nation, P. (2001). Learning vocabulary in another language. Cambridge: Cambridge
University Press.

Nicoladis, E. (1999) "Where is my brush-teeth?" Acquisition of compound nouns in a
F rench-English bilingual child. Bilingualism: Language and Cognition, 2, 245-
256.

Nicoladis, E. (2002). What’s the difference between ‘toilet paper’ and ‘paper toilet’?
French-English bilingual children’s crosslinguistic transfer in compound
nouns. Journal of Child Language, 29, 1-20.

Nicoladis, E. (2003). What compound nouns mean to preschool children. Brain and
Language, 84, 38-49.

Nicoladis, E. (2007). The effect of bilingualism on the use of manual gestures. Applied
Psycholinguistics, 28, 441—454.

Nooteboom, S., Weerman, F., & Wijnen, F. (Eds.). (2002). Storage and Computation in
the Language Faculty. Dordrecht, The Netherlands: Kluwers Academic Press.

Norman, D., & Shallice, T. (1986). Attention to action: Willed and automatic control of
behavior. In R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.), Consciousness
and self regulation: Advances in research and theory (vol. 4, pp. 1-18). New
York: Plenum Press.

152

O’Grady, W. (2003).The radical middle: nativism without Universal Grammar. In C.
Doughty and M. Long (Eds.), Handbook of second language acquisition (pp. 43-
62). Oxford: Blackwell.

Piera, C. (1995). On compounding in English and Spanish. In H. Campos and P.
Kempchinsky (Eds.), Evolution and revolution in linguistic theory (pp. 302-315).
Washington: Georgetown University Press.

Pinker, S. (1999). Words and rules: The ingredients of language. New York: Basic
Books.

Pinker, S., & Ullman, M. (2002). The past and future of the past tense. Trends in
Cognitive Science, 6, 456-463.

Plag, I. (2003). Word-formation in English. Cambridge: Cambridge University Press.

Pollatsek, A., Hyona, J ., & Bertram, R. (2000). The role of constituents in reading
Finnish compound words. Journal of Experimental Psychology: Human
Perception and Performance, 26, 820-833.

Potter, M. C., & Falconer, A. (1979). Understanding noun phrases. Journal of Verbal
Learning and Verbal Behaviour, 18, 509—521.

Preston, M. S. & Lambert, W. E. (1969). Inerlingual interference in a bilingual version
of the Stroop color-word task. Journal of Verbal Learning and Verbal Behavior,
8, 295-301.

Pulido, D. (2003). Modeling the role of second language proﬁciency and topic familiarity
in second language incidental vocabulary acquisition through reading. Language
Learning, 53, 233-284.

Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A comprehensive grammar of
the English language. London: Longman.

Raaijmakers, J. G. W., Schrijnemakers, J. M. C., & Gremmen, F. (1999). How to deal
with "The language-as-ﬁxed-effect fallacy": Common misconceptions and
alternative solutions. Journal of Memory and Language, 41, 416-426.

Raffray, C. N., Pickering, M. J ., & Branigan, H. P. (2007). Priming the interpretation of
noun-noun combinations. Journal of Memory and Language, 5 7, 380-395.

Rastle, K, & Davis, M. H. (2008). Morphological decomposition based on the analysis of
orthography, Language and Cognitive Processes, 23, 942-971.

Rastle, K., Davis, M. H., & New, B. (2004). The broth in my brother’s brothel: Morpho-
orthographic segmentation in visual word recognition. Psychonomic Bulletin &
Review, I, 1090—1098.

153

 

Raupach, M. (1984). Formulae in second language speech production. In H.
W. Dechert, D. Mole, and M. Raupach (Eds.), Second language
productions (pp. 114—137). Tiibingen, Germany: Gunter Narr Verlag.

Read, J. (2000). Assessing vocabulary. Cambridge: Cambridge University Press.

Rochford, G., & Williams, M. (1965). Studies in the development and breakdown of the

use of names: IV. The effects of word frequency. Journal of Neurology
Neurosurgery, and Psychiatry, 28, 407-413.

Rumelhart, D. E., McClelland, J. L., and the PDP research group. (1986). Parallel
distributed processing: Explorations in the microstructure of cognition. Volume
1. Cambridge, MA: MIT Press.

Sagarra, N. (2008). From CALL to face-to-face interaction: the effect of computer-
delivered recasts and working memory on L2 development. In A. Mackey (Ed.),
Conversational interaction in second language acquisition (pp. 229-248). Oxford:
Oxford University Press.

Sandra, D. (1990). On the representation and processing of compound words: Automatic
access to constituent morphemes does not occur. The Quarterly Journal of
Experimental Psychology, 42A, 529-567.

Schmitt, N. (Ed.). (2004). Formulaic Sequences: Acquisition, Processing and Use.
Amsterdam: John Benjamins.

Schmitt, N., & Carter, R. (2004). Formulaic sequences in action: An introduction.
In N. Schmitt (Ed.), F ormulaic sequences: Acquisition, processing, and
use (pp. 1—22). Philadelphia: John Benjamins.

Schwartz, A. 1., & Kroll, J. F. (2006). Bilingual lexical activation in sentence context.
Journal of Memory and Language, 55, 197—2 12.

Sharwood Smith, M. (1994). Second language learning: theoretical foundations. London:
Longman.

Shoben, E. J. (1991). Predicting and nonpredicting combinations. In P.J. Schwanenﬂugel
(Ed.), The psychology of word meanings (pp. 117—135). Hillsdale, NJ: Erlbaum.

Slabakova, R. (2002). The compounding parameter in second language acquisition.
Studies in Second Language Acquisition, 24, 507—540.

Sohn, H. (1999). The Korean Language. New York: The Cambridge University Press.

Springer, K., & Murphy, G. L. (1992). Feature availability in conceptual combination.
Psychological Science, 3, 111—1 17.

154

Stockwell, R., & Minkova, D. (2001). English words: History and structure. New York:
Cambridge University Press.

Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of
Experimental Psychology, 12, 643-662.

Swinney, D. A. (1979). Lexical access during sentence comprehension:
(Re)consideration of context effects. Journal of Verbal Learning and Verbal
Behavior, 18, 645—659.

Taft, M. (1979). Recognition of afﬁxed words and the word frequency effect. Memory
and Cognition, 7, 263-272.

Taft, M. (1994). Interactive activation as a framework for understanding morphological
processing. Language and Cognitive Processes, 9, 271—294.

Taft, M. (2004). Morphological decomposition and the reverse base frequency
effect. Quarterly Journal of Experimental Psychology, 5 7A, 745-765.

Taft, M., & Forster, K. I. (1976). Lexical storage and retrieval of polymorphemic and
polysyllabic words. Journal of Verbal Learning and Verbal Behavior, 15, 607-
620.

Turner, M.L., & Engle, R.W. (1989). Is working memory capacity task dependent?
Journal of Memory and Language, 28, 127-154.

Twilley, L. C., Dixon, P., Taylor, D., & Clark, K. (1994). University of Alberta norms of
relative meaning frequency for 566 homographs. Memory and Cognition, 22, 111-
126.

Ullman, M. T., Pancheva, R., Love, T., Yee, E., Swinney, D., & Hickok, G. (2005).
Neural correlates of lexicon and grammar: Evidence from the production, reading,
and judgment of inﬂection in aphasia. Brain and Language, 93, 185—238.

Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R.W. (2005). An automated version
of the operation span task, Behavior Research Methods, 3 7, 498—505.

Van Jaarsveld, H., Coolen, R., & Schreuder, R. (1994). The role of analogy in the
interpretation of novel compounds. Journal ofPsycholinguistic Research, 23,
111-137.

Van Jaarsveld, H., & Rattink, G. (1988). Frequency effects in the processing of
lexicalized and novel nominal compounds. Journal of Psycholinguistic Research,
I 7, 447-473.

Wardhaugh, R. (1995). Understanding English Grammar. Oxford: Blackwell.

155

Warren, B. (1978). Semantic Patterns of Noun-Noun Compounds. Gothenburg, Sweden:
Acts Universitatis Cothoburgensis.

Waters, G. S., & Caplan, D. (1996). The measurement of verbal working memory
capacity and its relation to reading comprehension, Quarterly Journal of
Experimental Psychology, 49A, 51—74.

Webb, S. (2005). Receptive and productive vocabulary learning: The effects of reading
and writing on word knowledge. Studies in Second Language Acquisition, 27, 33-
52.

Weinert, R. (1995). The role of formulaic language in second language acquisition: A
review. Applied Linguistics, 16, 180—205.

Wisniewski, E. J. (1996). Construal and similarity in conceptual combination. Journal of
Memory and Language, 35, 434—453.

Wisniewski, E. J ., & Murphy, G. L. (2005). Frequency of relation type as a determinant
of conceptual combination: A reanalysis. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 31, 169-174.

Wolter, B. (2001). Comparing the L1 and L2 mental lexicon. Studies in Second Language
Acquisition, 19, 41-69.

Wray, A. (2002). F ormulaic Language and the Lexicon. Cambridge: Cambridge
University Press.

Zied, K., Phillipe, A., Karine, P., Valerie, H., Ghislaine, A., Amaud, R., & Didier, L.
(2004). Bilingualism and adult differences in inhibitory mechanisms: Evidence
from a bilingual stroop task. Brain and Cognition, 54, 254-256.

Zimmer, K. (1971). Some general observations about nominal compounds. Working
Papers on Language Universals, 5, Stanford University.

Zimmer, K. (1972). Appropriateness conditions for nominal compounds. Working Papers
on. Language Universals, 8, 3-20, Stanford University.

Zwitserlood, P. (1994). The role of semantic transparency in the processing and
representation of Dutch compounds. Language and Cognitive Processes, 9, 341—
368.

156

 

     

AN A

”'11111113131111[111131711111111115’

3 63 2396