Applied Linguistics
Applied Linguistics 2014: 35/2: 184–207 � Oxford University Press 2013 doi:10.1093/applin/amt013 Advance Access published on 13 July 2013
Dynamics of Complexity and Accuracy: A Longitudinal Case Study of Advanced Untutored Development
*BRITTANY POLAT and YOUJIN KIM
Georgia State University
*E-mail: bpolat@student.gsu.edu or brittanypolat@gmail.com
This longitudinal case study follows a dynamic systems approach to investigate
an under-studied research area in second language acquisition, the development
of complexity and accuracy for an advanced untutored learner of English. Using
the analytical tools of dynamic systems theory (Verspoor et al. 2011) within the
framework of complexity, accuracy, and fluency (Skehan 1998; Norris and
Ortega 2009), the study tracks accuracy, syntactic complexity, and lexical
diversity in the speech of a Turkish immigrant over one year. Results from
these oral interviews show that most development occurred in the participant’s
lexical diversity, syntactic complexity showed potential but unverifiable gains,
and accuracy showed no development. These findings suggest that an untutored
language learner may develop advanced lexical and syntactic skills, but achiev-
ing grammatical accuracy without instruction may be more difficult. Overall,
dynamic systems theory seems to provide a suitable framework for examining
the linguistic development of advanced naturalistic learners, with important
implications for future research involving untutored immigrant and refugee
populations of English language learners.
INTRODUCTION
In recent decades, second language acquisition (SLA) research has
predominantly focused on issues in instructed language learning rather than
naturalistic language learning. Despite promising early research and several
seminal studies of untutored adult learners—Schmidt’s (1984) Wes study,
Schumann’s (1978) Alberto study, Huebner’s (1983) Ge study—which have
made significant contributions to the field, the vast majority of publications
today concentrate on instructed language learning. Although many of these
have certainly increased our understanding of how language learning works in
the classroom, there are compelling reasons to pay more attention to language
acquisition outside the classroom.
Whereas many of the students who participate in SLA studies have the
luxury of formal language instruction, the majority of the world’s language
learners acquire second and additional languages in naturalistic contexts
(Klein and Perdue 1993). Without knowing how this type of learning takes
place, SLA researchers are missing a crucial part of the language acquisition
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
process: the basic mechanisms that have allowed humans to create and pass
on languages for millennia (Klein and Dimroth 2009). As Klein and Dimroth
put it:
Untutored second language acquisition is not something exotic, it is the normal case, and if we want to understand the very principles according to which the human mind constructs, copies, and uses linguistic systems, then we must study how human beings cope with this task when not under the influence of teaching. (p. 519)
As the field discovers how important social and situational factors are in lan-
guage learning, it becomes increasingly apparent that tutored and untutored
acquisition may have very different driving factors. So far, however, only a few
studies in the past two decades have explored naturalistic language learning
(Klein and Perdue 1993; Ioup et al. 1994; Dimroth and Starren 2003), and of
these, only Ioup et al. (1994) have specifically investigated advanced natural-
istic learning, with an English-speaking learner of Arabic.
Perhaps the most extensive study of untutored language learning to date is
that conducted by the European Science Foundation from 1981 to 1988
(Perdue 1993). This study tracked the development of 40 language learners
from a variety of first language backgrounds in five host countries (Britain,
France, Sweden, Germany, and The Netherlands) over 30 months and found
that immigrants at first developed a basic variety (BV) of the target language.
The BVs were all very similar, lacking morphological inflection and consisting
of a rudimentary lexicon, and they mainly seemed independent of the lear-
ner’s first language and target language. Whereas about one-third of the im-
migrants remained at this basic level throughout the study, the others
continued to develop beyond the BV (Klein and Perdue 1993). The study
has yielded important insights into universals of basic language varieties, as
well as developmental stages that most naturalistic learners appear to pass
through. However, because the study did not report on learners beyond
basic development, we know neither how proficient these learners ultimately
became nor what their most advanced forms of learner language looked like.
The present study addresses the research lacuna of advanced naturalistic
learning by examining the language development of an untutored adult lan-
guage learner beginning two and a half years after his arrival in the target
language environment. The methodological approach taken is that of dynamic
systems theory (DST) (Verspoor et al. 2011), and the variables investigated are
two widely used constructs of language performance, complexity and accuracy
(Ellis and Barkhuizen 2005). In the following text, we will address the theor-
etical considerations behind researching complexity and accuracy from a
dynamic systems perspective, and then we will present the 12-month case
study. We conclude with a discussion of implications for untutored language
acquisition and the importance of integrating research on untutored learning
over time into mainstream SLA research.
B. POLAT AND Y. KIM 185
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
CAF AND DST
Theoretical platforms
Researchers of second language (L2) development are increasingly relying on
measures of complexity, accuracy, and fluency (known together as CAF) to
assess learners’ written and oral proficiency and to probe more deeply into
the cognitive processes of language learning (Ellis and Barkhuizen 2005).
Originally conceived as a way to distinguish aspects of task performance,
these three components are oriented toward either form (complexity and
accuracy) or meaning (fluency; Skehan1998). One of the major advantages of
CAF-based research is that it provides a sophisticated framework for investigat-
ing the multicomponential nature of language use and development. As form
has consistently been shown to be challenging for naturalistic learners (Schmidt
1984; Dimroth and Starren 2003), the present study focuses on the two form-
oriented components of L2 oral performance, complexity and accuracy.
The CAF constructs have frequently been examined by task-based SLA
researchers who are interested in the role of task design and implementation
in L2 performance (Ellis 2003; Samuda and Bygate 2008). Some researchers
assert that the CAF measures have been inconsistently defined and operatio-
nalized (Housen and Kuiken 2009), leading to calls for consistent, specific, and
validated CAF measures to be used across studies (Norris and Ortega 2009;
Pallotti 2009). Norris and Ortega (2009), for example, argue that different
operationalizations of complexity capture different facets of language develop-
ment, and that researchers should use multiple construct measurements to
provide a more complete picture. They also call for ‘more organic practice’
(p. 574) and a deeper consideration of context as an influence on CAF. In
other words, according to Norris and Ortega (2009), ‘our measurements
must provide multivariate, longitudinal, and descriptive accounts of constructs
in L2 performance in order to capture the complex, dynamic, and develop-
mental nature of CAF phenomena’ (p. 574).
At the same time, many SLA researchers are embracing the perspective that
SLA is an individualized nonlinear endeavor, and that research should con-
sider the variability and interaction of its components (Larsen-Freeman and
Cameron 2008a, 2008b). Although the idea of nonlinearity in language devel-
opment is not new, investigations undertaken within this dynamic systems
framework have applied new conceptual tools and analyses to the study of
developmental variability, showing the complex interrelations of CAF vari-
ables within language acquisition (Verspoor et al. 2011). DST researchers
advocate longitudinal, fine-grained, and microgenetic studies to discover indi-
vidual learning trajectories and the interrelationships of parts within the whole
(van Geert and van Dijk 2002; de Bot 2008; Larsen-Freeman and Cameron
2008a). Because DST is centered around time and variability, Ortega and
Byrnes (2008) propose that this theoretical approach is very well-suited to
the longitudinal study of advanced language capacities.
186 COMPLEXITY AND ACCURACY
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
Empirical research on CAF through DST
Because of the relative newness of the dynamic systems paradigm—and the
shift in perspective and analytical tools that it requires—only a few studies
have as yet connected this framework with CAF. Larsen-Freeman (2006)
was the first applied linguist to examine CAF through a DST lens, with a
focus on the variability between learners. Her investigation of five instructed
Chinese learners of English measured their written development of grammat-
ical complexity (clauses per t-unit), lexical complexity (a kind of type-token
ratio), accuracy (ratio of correct t-units to all t-units), and fluency (words per t-
unit) over four months, in addition to analyzing oral narrative idea units for
qualitative differences. The study revealed that although averaged group data
showed steady improvement in all three CAF components for the learners,
patterns of development for each individual were far removed from the aver-
aged trajectory. Learners exhibited unique trajectories, with different rates of
improvement and even decreases in some areas, an important fact that had
been obscured by the group averages.
Given this important individual variability in language acquisition, other
CAF studies have examined single language learners over a period of several
years. Verspoor et al. (2008) analyzed an advanced English learner’s academic
writing for development of vocabulary (measured by average word length,
type-token ratio, use of words from the Academic Word List) and complexity
(measured by length of noun phrase and number of words per finite verb).
They found that although the learner showed development in almost all the
aspects investigated, progress was nonlinear and was different for each vari-
able. Several interesting patterns emerged, including a possible competitive
relationship between development of type-token ratio and sentence length,
and a supportive relationship between finite verb ratio and noun phrase
length. The authors conclude that in the dynamic system of language learning,
‘there can be no development without variability, and the amount and type of
variability can reveal the actual developmental process’ (p. 229).
In another DST/CAF study, Spoelman and Verspoor (2010) tracked a Dutch
learner’s acquisition of written Finnish over three years. They examined
accuracy (case usage) and several measures of complexity (morphemes per
word, words per noun phrase, and ‘difference between the average sentence
length in morphemes and the average sentence length in words’, p. 539). The
results once again showed the interaction of variables over time, with the
learner’s complexity variables sometimes competing and sometimes support-
ing each other. Interestingly, although accuracy and complexity seemed to be
in competition early in the study period, they later changed to a noncompe-
titive relationship as the learner became more proficient, suggesting that
proficiency level may have an impact on the interaction of variables. Similar
to Verspoor et al. (2008), Spoelman and Verspoor (2010) maintain that these
language learner systems demonstrate the ‘classic’ jumps, transitions, and
nonlinear development of dynamic systems.
B. POLAT AND Y. KIM 187
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
The findings of these DST-based CAF studies show the importance of seeking
emergent dynamic patterns within the components of language systems
(Larsen-Freeman 2009; de Bot and Larsen-Freeman 2011). So far, however,
research in this area has concentrated on instructed, mainly written language
learning in academic settings, and researchers have not applied the DST the-
oretical framework or CAF constructs to studying untutored language devel-
opment. In addition, although several of these studies involve high-
intermediate (Larsen-Freeman 2006) or advanced learners (Verspoor et al.
2008), by and large the CAF/DST paradigm has not explicitly engaged with
the special concerns of advanced language capacities. One of the main goals of
the current study, therefore, is to specifically tease out the important issues
related to L2 advancedness, particularly as they apply to nonacademic
contexts.
ADVANCED LANGUAGE CAPACITIES
Although studies of advanced language learners have often figured in import-
ant SLA research, a wake-up call was sounded by Ortega and Byrnes’ (2008)
collection of longitudinal research on advanced language capacities. There is
an acute need, they argue, to closely examine the question of how ‘learning
over time evolve[s] toward sophisticated second language capacities, indeed to
high-level multiple-language capacities’ (p. 282). Researchers may often call
for longitudinal studies of language development, but the field has yet to come
to a consensus on what advancedness means in terms of L2 capabilities, or on
how it should be measured. Researchers such as Harklau (2008), Myles (2008),
and Angelelli (2008) offer different definitions and methodological techniques
for capturing advanced L2 use, including various qualitative and quantitative
approaches that examine linguistic, sociolinguistic, and pragmatic aspects of
language acquisition.
By claiming to investigate an advanced language learner, therefore, the pre-
sent study grapples with the unresolved theoretical issue of what advancedness
actually is. Research set in any kind of instructional context can easily rely on
test scores, institutional status, or classroom performance to define advanced
language capacities (Ortega and Byrnes 2008), and laboratory-based research
can elicit advanced or late-acquired linguistic features to claim advancedness.
In the present type of research conducted with an untutored learner, none of
these options are available. We therefore prefer the criterion of ‘advanced
language use in context’ (Ortega and Byrnes 2008: 282), based on what the
focal participant uses language to do in everyday life.
This more naturalistic approach to advancedness allows us to take the study
of advanced language capacities outside of academically defined parameters
and into the context of untutored learning. Just as there have been few studies
of untutored language acquisition from a CAF or DST framework, so, too, are
studies of advanced naturalistic learners few and far between. Ortega and
Byrnes’ (2008) collection does not include any studies on untutored learners,
188 COMPLEXITY AND ACCURACY
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
mainly because most research on advanced language learning has overwhelm-
ingly privileged written and academic language. The authors conclude that
‘longitudinal research on advancedness would benefit from sampling across
a variety of social settings that afford opportunities for diverse language rep-
ertoires, as this will enrich the developmental insights we obtain’ (Ortega and
Byrnes 2008: 284). It seems clear that not only are more longitudinal studies of
advanced language learning needed, but more are needed in a variety of con-
texts, such as untutored learners in a target-language setting.
A point of contention in the debate on advancedness seems to be whether
learner language should be compared with native speaker norms (Ortega and
Byrnes 2008). In this article, we take the position that L2 systems should never
be seen merely as deficient versions of native speaker language systems (Cook
2002; Harklau 2008), but it would be difficult to establish any learner’s level of
advancedness without considering target-like language use. For this reason,
the present study uses a native speaker comparison with the intention not of
showing deficiencies, but rather of showing the advanced language capacity of
an untutored learner. This is similar to Verspoor et al.’s (2008) inclusion of a
native speaker comparison, which is a helpful touchstone for interpreting the
performance of non-native speakers. In addition to the native speaker data
collected in this study, we offer comparisons of our participant’s language
with that of non-native English speakers in several previous studies that
have measured naturalistic oral data.
In summary, the study presented in later text seeks to join several strands of
research that have not yet been united but which have the potential to en-
hance our understanding of language learning: DST, CAF, naturalistic learn-
ing, and advanced language capacities. The complexity theory perspective
allows us to analyze various developmental patterns of an untutored but
nevertheless advanced L2 user, and the CAF platform provides a systematic
and conceptually clear set of tools for our investigation.
METHODS
Participants
Focal participant
The focal participant in this study, Alex (a pseudonym), is a native speaker of
Turkish who had lived in the USA for two and a half years at the beginning of
the interview period. Although Alex completed a bachelor’s degree in televi-
sion production at a prestigious university in Istanbul, he describes his English
at the time of his arrival in the USA as very basic. In fact, his experience with
English had been overwhelmingly negative before his interactions with
Americans. According to Alex, he had taken English for four years in high
school—delivered strictly through grammar-translation instruction—but never
managed to pass the class (he was allowed to graduate on the strength of his
B. POLAT AND Y. KIM 189
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
grades in other subjects). At university, he was required to take a year-long
English preparatory program before beginning his degree studies. Alex esti-
mates that he attended only about 30 percent of the English classes during that
year, and he failed to pass the English examination that would allow him to
proceed with his major studies. To help him move on with the degree, his
department allowed him to complete a ‘project’ in lieu of passing the difficult
examination; the project entailed writing 100 words on a sheet of notebook
paper, which Alex accomplished, thus ending his English learning
requirements.
Alex attributes his repeated failures and complete lack of interest in English
to poor teaching methods and to his belief (at the time) that the language was
completely irrelevant to his life. This belief necessarily changed, however,
when Alex moved to the USA at the age of 25. Once in the USA, he used
only English outside the home, although he reports reading newspapers,
watching movies and television shows, and talking weekly with his family
in Turkish. His interview comments reveal a positive orientation toward the
target language community and an openness to the new language and culture:
For me I’m don’t believe I’m belongs to one culture. Basically I am making my own culture . . . You know anytime I learn something, if it’s better than what I have, I get it. That’s my culture now. I found it something like that in English, in United State, and I took it some of them. Now they are my culture. But, something is ridiculous, it will never be my culture . . . I believe everybody have to do that, like this. (February 14)
Although Alex did have some formal language instruction in his home coun-
try, in this study, he is considered an untutored learner because his English
skills were rudimentary at the time of his arrival in the target language con-
text, because he has not taken any language classes in the USA, and because
he has learned English primarily through quotidian interaction (Lightbown
and Spada 2006). At the same time, Alex can be considered an advanced
English user based on what he is able to accomplish through everyday use
of the language (Harklau 2008; Ortega and Byrnes 2008). During this year-
long study, he worked in an English-only context as an assistant department
manager in a supermarket, supervising 25 employees (mainly native English
speakers), complying with strict federal food safety regulations, and managing
high volumes of perishable food inventory. Alex began working part-time in
the supermarket six months after he arrived in the USA and in three years was
promoted three times, in competition with native English speakers. Shortly
after the study ended, he was promoted again, to department manager.
Native speakers
Three native speakers were selected for comparison (two females and one
male), and they were each interviewed under similar circumstances as Alex
190 COMPLEXITY AND ACCURACY
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
(see Procedures for details). These speakers were all undergraduate
students at a university in the same city where Alex lives, with majors of
religious studies, applied linguistics, and education/drama. We believe that
these participants provide an appropriate comparison group because they
are at a similar education level as Alex (obtaining bachelor’s degrees) and
were discussing topics conceptually similar to topics in several of Alex’s
interviews.
Procedures
For exactly one year, Alex was interviewed once every two weeks for approxi-
mately 30 minutes. Several factors contributed to providing an authentic con-
text for language production: (1) Alex is a friend of the first author and has
experience discussing a wide variety of topics with her, (2) the interviews took
place in a familiar and nonthreatening environment, and (3) Alex was encour-
aged to choose topics that he enjoyed and felt comfortable discussing through-
out the unstructured interview. The interviews were carried out by the first
author, whose primary role was simply to be a conversational participant to
elicit speech production from Alex. Because the goal of data collection was to
gather authentic speech, topics varied and the conversation was unstructured
and unplanned (Duff 2008). Alex selected the topics for discussion (such as
politics, childhood memories, or his experience learning English) and could
decide when to move on to a new subject. The interviews therefore provided
realistic and meaning-oriented communicative situations (Hesse-Biber and
Leavy 2011).
Interviews were held every two weeks to capture any microgenetic
changes in Alex’s language. Microdevelopment is important in DST, as it can
provide details on how processes actually develop, particularly during key
moments of transition (Larsen-Freeman and Cameron 2008b). In addition,
the study was designed along the longitudinal timescale of one year, which
was long enough to represent development but still a manageable commit-
ment for the participant (see Ortega and Iberri-Shea 2005 for a discussion of
choosing timescales).
In addition to Alex’s data, three interviews were held with the native speak-
ers to obtain comparison data. These interviews were conducted in much the
same way as the interviews with Alex, with the main difference being that
they were slightly longer, at 45–60 minutes. The three native speakers were
familiar with the first author, who also carried out these interviews. Interviews
were recorded in a familiar environment on a laptop computer, and the topic
discussed was language learning, which was also a topic in Alex’s interviews.
Although native speaker conversation can vary in its complexity, because vari-
ables such as interlocutor, topic, and passage length were controlled for (see
Data Analysis), these data provide an appropriate starting point for target-like
use comparison.
B. POLAT AND Y. KIM 191
Downloaded from https://academic.oup.com/applij/article-abstract/35/2/184/191987 by University of California, Santa Cruz user on 12 December 2017
Data analysis and intercoder reliability
To investigate Alex’s longitudinal development of complexity and accuracy,
100-word passages were taken from each interview transcript (following
Spoelman and Verspoor 2010). The passage selected from each transcript
was the end of the last turn in which Alex spoke more than 100 words. This
was done to eliminate very short interactional utterances and dialogic re-
sponses that may have different levels of complexity and accuracy than
longer utterances. Passages were taken from the end of each interview because
it was assumed that Alex would be speaking more naturally at the end rather
than the beginning of the recording. Due to the organic nature of the interview
setting, discussion topics varied by session. 1
False starts, repetitions, inserts,
and other hesitation phenomena were excluded from the passages, so that the
100-word segments represented Alex’s speech without hesitation phenomena.
As in Spoelman and Verspoor (2010) and other studies, the oral data were
converted to CHAT format to be compatible with CHILDES program software
(MacWhinney 2000). The data were then analyzed for syntactic complexity
(mean length of AS-units, clauses per AS-unit, mean length of clauses), lexical
diversity (D), and accuracy (errors per 100 words, present simple tense).
AS-units
The 100-word passages were divided into analysis of speech units (AS-units),
which are defined as ‘an independent clause, or sub-clausal unit, together with
any subordinate clauses associated with either’ (Foster et al. 2000: 365).
AS-units have been widely used with oral data in SLA studies in the past
decade (Norris and Ortega 2009), and this was deemed the most appropriate
unit of measurement for the present study. To maintain consistent AS-unit
analysis, we followed Foster et al. (2000) as far as possible. Some examples of
the AS-unit in the data include You are not government and Because they knew
ninety percent people say ‘yes’.
In cases where the 100-word passages did not coincide with the boundaries
of AS-units, the entire AS-unit was retained for purposes of counting mean
length of AS-units and clauses per AS-unit. This was done to avoid including
incomplete AS-units in these complexity measures, which would have dis-
torted them.
