Professional Documents
Culture Documents
net/publication/324808335
CITATIONS READS
2 1,886
1 author:
Timothy Hall
University of Pennsylvania
5 PUBLICATIONS 2 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Timothy Hall on 27 April 2018.
by
Dissertation Committee:
2017
ABSTRACT
The field of Second Language Acquisition has sought to explain how learners
Ellis (2002a) proposed that rule-like constructions emerge when domain-general learning
This exploratory study took N. Ellis’s proposal as a starting point to investigate how three
Naturalistic oral data was elicited over ten months by means of three language tasks. The
oral output was recorded, transcribed into corpora, and coded for chunks. Patterns of
chunk use were analyzed to indicate the following: (a) All learners used chunks to
communicate a variety of meanings; (b) There were inter-learner differences with regard
to chunk use; (c) There were inter- and intra-learner differences in how chunk
ii
ACKNOWLEDGMENTS
This dissertation was possible thanks to the consideration and support of many
individuals. I would like first to thank Dr. ZhaoHong Han, my dissertation advisor, for
what has been the most profound learning experience of my life. I would also like to
thank Drs. Vivian Lindhardsen, Hoa Nguyen, and Hervé Varenne for their insightful
review of my work. Past and present members of my doctoral seminar have motivated me
decade of work feasible and collegiate. Thanks to them all. In particular, I wish to thank
Adrienne Lew, Sunhee Song, Sarah Sok, Phillip Choong, Alice Chen, Farah Akbar, and
Rosette Finneran. I extend thanks to Marta Lynch, Ann Holby, Debra Billman, and Reyes
Llopis Garcia for coding, proofreading, and moral support, and to Carol Friend for access
and Jun. I also thank Nick Ellis for his willingness to engage in a thoughtful conversation
with me on the current topic, and of course, to pose such an interesting question in the
first place.
My academic work was completed with the support of friends, colleagues, and
family. Thank you Lonnie Lippert, Lea Rumbolo, Janet Immatteo, Pat Brokaw, and Andy
Black for your unwavering reliability. Thank you to Yiqiang Wu and Jean Wong, Nicole
Maldonado, Megan Gordon, Alan Amtzis, Amy Dell, Diane Gibson, and Mary Ann
Peterson for your encouragement. And heartfelt thanks to Kelly Blair and Michael
Heather, Liz, and Aaron for being the rock upon which I stand.
T.M.H.
iii
TABLE OF CONTENTS
I -INTRODUCTION………………………………………………………………..…….. 1
1.1 Background…………………………………………………..……….……. 2
1.2 Focus of the Study……………………………………………………… 8
1.3 Key Terms………….……………………………………………….…… 9
Acquisition……..………..…………………………………….……. 9
Interlanguage……..………..………………………………………… 9
Constructions …………………………………….…………………. 10
Learner Chunk………………………………………………………. 11
N-gram………………………………………………………………. 11
Language Processing………….…………………………………….. 12
Holistic Processing and Storage…………………………………….. 12
Processes and Mechanisms………………………………………….. 12
Associative Learning………………………………………………… 13
Type and Token Frequency…………………………………………. 13
Productivity …………………..…………………………………….. 14
Grammaticalization…………..……………………………………… 15
Comparison, Analysis, Abstraction, Schematization……………….
15
Exemplars ………………………………………………………….. 16
Exemplar-Based Learning…………………………………………... 16
1.4 Outline of the Dissertation……………………………………………… 17
II - REVIEW OF THE LITERATURE…………………………………..……………… 18
2.1 Emergentism in SLA…………………………………………………… 18
Frequency Effects…………………………………………………… 23
Counterpoints to Frequency Effects………………………………… 27
2.2 Learner Chunks…………………………………………………………. 31
2.3 Development of Chunks………………………………………..……….. 37
2.4 The Pilot Study………………………………………..………..……….. 67
III - METHODOLOGY………………………………………………….……………… 72
3.1 Methodological Issues………………………………………….……….. 72
3.2 Design of the Study…………………………………………….……….. 80
3.2.1 Context of Study…………………………………………………. 80
3.2.2 Participants……………………………………………………….. 82
iv
3.2.3 Data…………………………………………………..........……… 84
Instruments………………………………………………………….. 84
Narrative Task…………………………………………………… 86
Role Play Task……………………..……………………………. 87
Live Talk Task…………………………………………………… 87
3.2.4 Procedure………………………………………………..……….. 90
3.2.5 Coding and Analyses…………………………………………….. 93
Descriptive Statistics………………………………….…............. 95
Single Task Analysis……………………………………….......... 96
Chunk-Based Analysis ……………………..……………………. 96
IV - INTER-LEARNER ANALYSES AND RESULTS………….…………… 98
4.1 Learner Chunks………………………………………………………… 98
4.2 Descriptive Statistics………………………………………………….... 100
4.3 Inter-Learner Task Comparison………….…………………………….. 104
V- INTRA-LEARNER ANALYSES AND RESULTS……….….…………………..116
5.1 Weike…………………………………………………………………… 116
5.2 Eddy…………………………………………………………………….. 127
5.3 Jun………………………………………………………………………. 134
5.4 Summary………………………………………………………………... 140
VI - DISCUSSION AND CONCLUSION………………………………………. 144
6.1 To What Extent Do Adult Learners of English as a Second Language Use
Chunks as a Resource for Acquisition ……………………………… 144
Inter-learner analyses and results………………………............... 144
Intra-learner analyses and results………………………............... 148
6.2 Study Limitations, Implications, and Directions for Future Research.… 157
REFERENCES ………………………………………...............…………………. 165
Appendix A: Tasks……………………………..............................……………. 179
Appendix B: Weike…………………………….....................…………………… 184
Appendix C: Eddy………………………........................…………………… 200
Appendix D: Jun………………………...........................…………………… 213
v
LIST OF TABLES
vi
5.6 Weike: <what’s happened / what’s going on>…………………………… 124
vii
LIST OF FIGURES
viii
1
I – INTRODUCTION
communicate in a second language (L2), which in turn has motivated a keen interest in
be predicated on understanding of how languages are learned. Over the course of four
decades, the field of Second Language Acquisition (SLA) has attempted to inform such
Scarcella, 1978; Myles, Mitchell & Hooper, 1999); L2 learning can follow predictable
paths or stages (Pienemann & Lenzing, 2015; VanPatten & Williams, 2015); and L2
outcomes tend to be variable across linguistic subsystems and across learners (Han, 2004,
2009). These phenomena have been described from different perspectives, including
universal grammar, sociocultural theory, skill acquisition theory, monitor theory, and
input processing theory, to name a few. However, the field has yet to embrace a cohesive
More and more, SLA researchers have been working in a theoretical framework
called emergentism (MacWhinney, 2015) and its sub-variants — dynamic systems theory
(deBot, 2008; Lowie & Verspoor, 2015), complex adaptive systems (Beckner et al.,
2009), learner varieties (Dimroth, 2012), ACT-R (Andersen, 1996) and usage-based
linguistics (UBL) (Bybee 2008; Ellis & Wulff, 2015; Ellis, O’Donnell, & Römer, 2013;
Eskildsen, 2015; Tomassello, 2003). All of these approaches share a common belief that
2
language acquisition emerges from the dynamics of language use. The present study
considers a particular emergentist assertion that L2 learners can develop productive rule-
resource, and proceed according to a specific pattern of development. (N. Ellis, 2002a;
Eskildsen, 2015). The present study seeks evidence of the relationship between chunks
and rules in three adult L2 learners, and seeks a greater specification of the role that
understanding of how L2 learners use chunks of language, what roles chunks play in
for SLA.
1.1 Background
Since the late 1960s, researchers in the field of SLA have sought to understand why
the outcomes of L2 learning differ from those of first language (L1) learning. Variable L2
outcomes have been observed on an intra-learner basis because learners often fail to
outcomes have also been observed on an inter-learner basis because some learners are
less successful than others (Cook, 2010; Han, 2004; Kellerman, 1995; Selinker, 1972).
Such variable outcomes led to an assertion that there is a fundamental difference both in
the processes and in the products of first and second language acquisition, as articulated
order to account for the divergences in L1 and L2 learning, SLA researchers formed
3
many guiding questions, the most relevant to the present study being: (a) With what
resources do L2 learners build the mental structure of a new language? (b) By what path
does L2 acquisition proceed? (c) Why and how do we see inter- and intra-learner
research into these questions. Broadly speaking, this theory seeks to describe “what we
know about language and where this knowledge comes from” (Cook & Newson, 2007, p.
4). UG proposes that L1 acquisition is guided and constrained by universal principles and
containing predetermined linguistic categories such as nouns, verbs, and modifiers that
combinatorial rules (Hauser, Chomsky, & Fitch, 2002). The universal principles and
corresponding innate system are argued to constrain the range of possibilities that a
One of the central problems that UG attempts to address is what is called the logical
problem of language acquisition. It asserts that “learners of a language know more about
what can and cannot be done in a language than they could have possibly learned from
the input alone” (Gass, 2013, Kindle Locations 5295-5296). Proponents of UG assert that
implicit and abstract knowledge of language” (VanPatten & Benati, 2015, p. 104) and
4
performance, which is the language that speakers produce. This distinction is grounded in
The logical problem argument, the assertions of an innate language system, and the
distinction between competence and performance carry over into the field of SLA under
the moniker of generative SLA (GenSLA) (Hawkins, 2009; Slabakova, Leal, & Liskin-
Gasparro, 2014, 2015; White, 1992, 2003). Like UG, GenSLA focuses on resolving
inconsistencies between the exposure to language input and the resulting development
(Han, 2009) and like UG, GenSLA emphasizes the role of an innate language learning
faculty and de-emphasizes the role of language input, whose essential value is restricted
There are objections to UG’s relevance in SLA (deBot, 2015), notably because the
relationship between input and acquisition does not appear to be the same in the L2 as in
the L1. Unlike L1 learners, L2 learners do not consistently arrive at target-like intuitions
about their new language. L2 learner language abilities do not seem to be reliably
minds of learners possess some forms that remain L1-like, and yet others that are neither
L1 nor L2-like (Han, 2009; Selinker, 1972). These observations led researchers to an
understanding that UG shapes the problem space of L2 learning, but that it has “a highly
The GenSLA perspective has neglected other factors that have been shown to shape
L2 learning. These include cognitive faculties (Schmidt, 1990; Selinker, 1972; Skehan,
1998; VanPatten, 1996), factors related to language use such as input frequency and
recency (Hoey, 2005; Larsen-Freeman, 2002), interaction (Long, 1996), output (Swain,
2005), and how frequently and reliably language forms can be traced to their underlying
2000; Conway & Christiansen, 2002; Croft & Cruise, 2004; N. Ellis, 1996; Ellis &
Wulff, 2015; Eskildsen, 2015; MacWhinney, 2015; Tomassello, 2003) have taken a
different approach to describing the relationship between the exposure to language input
and L2 acquisition. Emergentists believe that in the L1 and the L2, language structures
emerge and change in the mind as a result of patterns of language use (Tomassello,
2003). They assert that there is no innate language endowment. Instead, there are domain-
general cognitive mechanisms that interact with language input. This interaction is
sufficient initially to produce and subsequently to change language structure in the mind,
organization of one’s experience with language” (Bybee, 2008 p. 216), and linguistic
knowledge is more gradient than categorical (Crocker & Keller, 2006; O’Grady, 2010a).
(Barlow & Kemmer, 2000; Bates & Goodman, 1999; Bybee, 2013; Tomassello, 2003)
as associative learning, attention, and memory (Baddeley, Gathercole, & Papagno, 1998;
N. Ellis, 2001; Ellis & Sinclair, 1996; Gathercole & Baddley, 1993; Engel de Abreu &
Gathercole, 2012; Robinson, 1995, 2015; Schmidt, 1990; Skehan, 1998), and processing
tendencies (Han & D’Angelo, 2009; O’Grady, 2003; VanPatten, 1996, 2002). These
input, specifically the frequency and variety of input features (N. Ellis, 2002a, 2002b,
2012; Ellis & Wulff, 2015; Larsen-Freeman, 2002; Year, 2009), the context and recency
of input features (N. Ellis, 2002a; Hoey, 2005), and the extent to which meanings and
language forms are mutually retrievable (Han, 2008, 2009, 2013; Miller & Schmitt,
2010). Language use includes the consideration of language output because the patterns
of neural activation that underlie encoding processes are deemed to affect the language
system (Swain, 2005). Finally, language use entails interaction (Long, 1996) because of a
example, Krashen (1985) broadly stipulated that input must be meaningful for acquisition
to take place. Han (2013) further stipulated that form-meaning relationships must be
robust in the input and transparent to the learner for structures to be more easily acquired.
Equally, Swain specified in her distinction between output and pushed output that the
Similarly, in her evaluation of interaction, Gass (2013) observed that feedback and the
7
thereby catalyzing language acquisition. Finally, MacWhinney (2015) echoed this idea,
communicate efficiently in ways that allow the listener to efficiently and accurately
decipher the message […] the forms of natural languages are created, governed,
constrained, acquired, and used in the service of communicative functions” (p. 2).
Following these emergentist assertions, it would seem that our entire language apparatus
learning mechanisms and patterns of language use in order to fulfill our drive to make
The arguments and data produced by researchers operating within the emergentist
paradigm have been robust enough that even GenSLA recently articulated a theoretical
interest. For example, recent reviews (Slabakova, Leal, & Liskin-Gasparro, 2014, 2015)
in the properties of linguistic input including frequency, and in the relationships between
form and meaning as important factors in SLA. However, there is a need to investigate
variable paths by which L2 learners use their new language and how they build their own
mental L2 structures. These last points constitute the scope of interest for the present
study.
8
which three adult L2 learners of English used holistically stored strings of language
(2002a) is that the emergence of language rules can start with chunks and proceed along a
a whole (e.g., howyadoin). Then, learning mechanisms and cognitive processes determine
which features are fixed and which are variable, and induce the gradual
L1 and child L2 learners (e.g., Hakuta, 1974; Myles, Hooper, & Mitchell, 1998; Myles,
Mitchell, & Hooper, 1999; Peters, 1983, 2009; Wong Fillmore, 1976). However, N. Ellis
(2002a) questions the extent to which this path is relevant in adult L2 language learning
given some additional complexities that are not present during L1 learning: a mind
optimally tuned for a preexisting L1, mature cognitive development, a history of formal
language instruction in some cases, and of course, fossilization. The literature presents
conflicting arguments about the role of chunks in adult L2 learning, thereby motivating
the present study. A number of key terms are used in this study, and are described in the
next section.
9
Acquisition
(VanPatten & Benati, 2015). Within the current framework, acquisition specifically
cognitive associations between language forms, meanings, and contexts of use such that
the learner can communicate in a second language. The present study uses acquisition,
Interlanguage
linguistic knowledge that underlie L2 language use, including “numerous elements, not
the least of which are elements from the NL [native language] and the TL [target
language]. There are also elements in the IL that do not have their origin in either the NL
or the TL […] the learners themselves impose structure on the available linguistic data
Processes that structure this data in the mind, according to Selinker (1972), include
Constructions
Following Langacker (1987, 2009) and Goldberg (2003, 2006), constructions are “stored
pairings of form and function, and include morphemes, words, idioms, partially lexically
filled and fully general linguistic patterns” (Goldberg, 2003, p. 219). A construction can
grammatical morpheme like the plural -s. It can be a full idiom with specified
represent larger semantic propositions (e.g., relative clauses, dependent clauses, and
discourse-level units) (Goldberg, 2003, 2006; Langacker, 1987, 2009). Because we are
learner output, the term construction will denote variable (i.e., syntactically or
Learner Chunk
A learner chunk is a type of construction created by the language learner. For the
processed in a holistic manner. Much of the recent SLA literature on the topic refers to
units (Peters, 1983), prefabricated patterns and routines (Hakuta, 1974, 1976), formulas
However, these terms, particularly the prevalent term, formulaic expression, can
N-gram
N-gram is a term used in corpus linguistics and denotes any sequence of N (3, 4, 5,
etc.) words. Software can be configured to scan a corpus of learner utterances and count
to the park, the software will segment the 5-word sequence into the following N-grams,
2-word grams (bi-grams): I walked, walked to, to the, the park (4 total)
3-word grams (tri-grams): I walked to, walked to the, to the park (3 total)
Language Processing
Language processing refers to computations that the mind performs during language
use. Processing, for example, occurs during comprehension (i.e., input processing),
during the accommodation of new features into the interlanguage (intake), and during the
2015).
learners may not necessarily have full knowledge of individual constituent units during
holistic processing (N. Ellis, 1996; Sinclair, 1991; Wray, 2000). For example, the phrase
whachadoin, in a holistic processing view, would presuppose that all constituents (e.g.,
what + are + you + doing) are accessed as a single item from long-term memory as if
they constituted one large word. Holistic storage denotes the intake of a multi-morphemic
through processes and mechanisms. The present study relies on Gernsbacher’s (1991)
usage, whereby processes are complex cognitive operations requiring multiple steps or
neural level that have catalytic or causative qualities (Bechtel, 2008; Gernsberger, 1991),
and can induce processes. Mechanisms include sensitivity to frequency and recency,
whereby neural activation is conditioned by how often and how recently a stimulus has
Associative Learning
fundamental, cognitive “sensitivity to the contingency between cues and outcomes” (N.
Ellis, 2008, p. 375). In language acquisition, associative learning mediates the acquisition
of form-meaning pairs through the co-perception of form and meaning. The more
frequently a form and a meaning can be reliably perceived together, the stronger the
Type and token frequency are properties of the input that interact with cognitive
the difference between tokens and types, consider the following set of linguistic forms:
[car, house, shop, car, shop]. The entire set contains five tokens; there are two tokens of
car, two tokens of shop, and one token of house. Token frequency denotes the number of
strengthen associations between linguistic features in the language system, to help form
Type frequency, however, denotes the number of different patterns available in a set,
or the number of contexts in which an item occurs. In the above example, the set contains
three linguistic types: car, house, and shop. When describing morphological features of
English, for example, the grammatical morpheme plural –s has a high type frequency
because of the great number of different nouns that it can attach to. The derivational
morpheme –ness has a comparatively low type frequency because of the fewer number of
nouns that it can attach to. In an emergentist view, encounters with high type frequency
items in the input ensure that a linguistic form is not associated with any particular
linguistic context. High type frequency facilitates the acquisition of abstract patterns
(Goldberg, 2006).
Productivity
denotes the number of linguistic contexts in which a construction can appear (Bauer,
2004; N. Ellis, 1998). High productivity is essentially synonymous with the concept of
high type frequency. For example, the plural -s is a highly productive construction in
English compared to the morpheme -ness. The passive construction is another productive
structure because it is used with many transitive verbs in a variety of discourse contexts.
15
Grammaticalization
within the chunk and changes outside the chunk. If a learner starts with knowledge of a
holistically stored chunk such as <there have> in the context of the utterance <there
have> three dogs, the learner gradually learns which of the features can vary (N. Ellis,
2002a). In this case, the learner breaks the exclusive association between there + have
and associates the constituents to other variations to produce utterances such as there had,
there is, there are, and there was. With regard to a learner chunk and its external
associative connections, a learner may possess a chunk <what are you> and produce an
utterance such as <what are you> do__ with that. Initially, the learner does not associate
the morphological slot for verb endings with relevant features inside the chunk, but over
the course of acquisition, the learner develops the appropriate association between the
stored, complex units such that their constituents are viewed individually by the language
system in their own right (Arnon & Christiansen, 2014). Abstraction creates a
16
paradigmatic “slot” in sequences based on regularities shared by two or more units (N.
Exemplars
Exemplars are memorized experiences of language (Gahl & Yu, 2006). When a
which contributes to the aggregate of his language experience and serves as a basis for
Exemplar-Based Learning
exemplar basis and form linguistic categories based on similarities across features
(Bybee, 2013; N. Ellis, 2006a, 2006b; Gahl & Yu, 2006). Learning mechanisms compare
new exemplars that are apperceived in the input to those that are already stored in the
the new memory. The new exemplar need not have the same core properties as all of the
exemplars in a comparison category (Rosch, 1973; Taylor, 2008) nor does it have to be
target-like. Instead, the new exemplar must have enough properties in common with some
of the existing exemplars that populate the category to become associated, which results
The present study is an exploratory attempt to investigate the role of learner chunks
in adult L2 acquisition. Chapter II reviews the relevant literature as it pertains to the role
of chunks in second language acquisition. Chapter III describes the methodology of the
present study. Chapter IV presents the findings from analyses that consider data from an
inter-learner perspective. Chapter V presents the findings from analyses that consider
data from an intra-learner perspective. Chapter VI synthesizes and discusses the findings,
and concludes with theoretical implications, limitations of the study, and directions for
future research.