Working Memory and Second Language Development: A Dynamic Systems Approach to Academic Writing Performance
Kathleen C. Scherer S3593452
M.A. Thesis, Department of Applied Linguistics Faculty of Arts
Rijksuniversiteit Groningen Supervisor: Dr. Merel Keijzer Second Reader: Dr. Hilde Hacquebord
Word count: 17, 549
Table of Contents
Table of Contents 2
Abstract 4
Introduction & Background 6
Research Question(s) 12
Hypotheses 14
Working Memory Constructs 15
Domain-general 15
Visuospatial Working Memory 17
Multi-sensory Working Memory 19
Working Memory and Second Language Development: A Dynamic Systems Approach 23
Development over time 23
Individual differences 28
Variability across task contexts 32
A Dynamic Systems Approach to Writing 35
Models of writing processes and SLD-WM 35
Multisensory/multimodal input and SLD 38
Proposal for Study 42
Research Design 44
Population 44
Materials 45
Procedure 48
Analysis 49
Conclusion 53
References 58
Appendix A 79
Appendix B 80
LIST OF ABBREVIATIONS
WM Working Memory
VSWM Visuospatial Working Memory DST Dynamic System(s) Theory LLA Language Learning Aptitude SLD Second Language Development SLA Second Language Acquisition CBTT Corsi Block Tapping test
L2 Second Language
L3 Third Language
MLAT Modern Language Aptitude Test CLT Communicative Language Teaching TOEFL Test of English as a Foreign Language
TESOL Teaching English to Speakers of Other Languages BICS Basic Interpersonal Communication Skills
CALP Cognitive Academic Language Proficiency OST Operation Span Task
EAP English for Academic Purposes CLAN Computerized Language Analysis FVR Finite Verb Ratio
MWL Mean Word Length
ASL Average Sentence Length
ANPL Average Noun-Phrase Length
Abstract
Working memory (WM) has been frequently invoked to predict second language
acquisition (SLA) success. The majority of past work has looked at the capacity estimates of the phonological (verbal) component of WM in relation to SLA. However, the same WM-related research has almost entirely overlooked the visuospatial component of the memory system because its direct relevance to second language development (SLD) has not been firmly
established (Baddeley, 2015). Furthermore, a great majority of WM research views memory as
an innate/fixed system once and individual reaches maturity. More recently, however, WM
functioning and SLA aptitude have been attributed to an interaction with instructional conditions
as well as prior language experience (Krashen, 1981; Bialystok, 2003; Cook, 2012; Huang,
2019). Therefore, the degree to which certain aspects of language aptitude are innate/fixed or
conditionally amendable is a focus of much research. In order to better understand the vast
complexities of WM and its role in second language development, this paper approaches WM as
a dynamic system (DST), which considers the complete connectedness of the memory system as
it develops over time, where language “emerges” at the convergence of human cognitive ability
and interaction with the language environment. The review of the literature prompts a much-
needed blueprint for a study that explores the visuospatial component of working memory
functions aided by multisensory input, and its role in the formulation and planning stages of the
writing process. Since multisensory input reduces the processing load on the limited capacity
(memory) system and/or enhances WM functioning, it should encourage more cognitively
integrated teaching methodologies and assessment models.
Keywords: working memory, visuospatial, domain general, dynamic systems,
multisensory input, individual differences, second language development
Working Memory and Second Language Development: A Dynamic Systems Approach to Academic Writing Performance
Introduction & Background
There are a great number of factors, both internal and external, that contribute to a person’s general ability to learn and such factors vary dramatically among individuals. The general ability to learn has been labeled learning aptitude (Skehan, 2015). Aptitude tests are often designed for use in various academic and professional contexts in order to assess what an individual is capable of producing under a given set of educational, instructional and
environmental circumstances (Smith, 2017). When extended more specifically to the context of second language learning, aptitude includes the degree to which an individual possesses the ability to recall sounds, remember words, and induce grammatical rules. Carroll (1958) was the first to classify these abilities as language learning aptitude (LLA). LLA has been a
controversial topic, most especially in educational settings where it has been viewed as outdated, irrelevant, and anti-egalitarian (Wen, 2016). After all, the main assumption underlying the definition of language aptitude is that some individuals are innately better at learning languages than others. Currently, LLA is seen as one of several individual differences that modulate language learning trajectories and outcomes and has influenced subsequent teaching
methodologies (Wen, Skehan, and Biedroń, 2017), but as a construct remains insufficiently
explored in second language development (SLD) research. Despite the insufficiently explored
parameters of LLA and its degree of ecological validity in educational settings, aptitude tests
such as the Modern Language Aptitude Test (MLAT; Carroll, 1959) persist in making claims about how well an individual will acquire a (second) language relative to others under the same circumstances. Such an approach to language learning gave rise, perhaps, to an
oversimplification of merely a “gift for languages”
Alternatively, aptitude has been attributed to an interaction with instructional conditions as well as prior language experience (Krashen, 1981; Bialystok 2003; Cook 2012) and therefore, the degree to which certain aspects of LLA, such as working memory (WM) capacity, are innate/fixed or conditionally amendable is a focus of much research in cognitive science (Sáfár and Kormos 2008; Singleton 2014; Singleton 2017; Dehn, 2015; Wen, 2019). This line of work has important consequences in the second language classroom (Dehn, 2015; Huang, 2019).
While other individual differences such as motivation, attitude, and personality are more
apparently susceptible to change and variability over a period of time (Dörnyei, 2010), the
prevailing understanding has been that LLA is a fixed and mostly stable cognitive construct
once an individual reaches maturity (Dörnyei and Skehan, 2003). However, there is converging
evidence which suggests that certain components of LLA, such as working memory capacity
(WM), may improve over the course of the language learning trajectory based upon increased
exposure to linguistic structures and educational conditions (Kormos, 2012; Dehn, 2015; Huang,
2019). The advances in understanding, more holistically, WM functioning and development as
well as behavioral tendencies of individuals in the context of second language learning, have
implored researchers and educators to reconsider and reconceptualize WM and LLA and its
significance in SLD and L2 instruction (Wen, 2016). This includes strategies for educational
interventions such as increasing working memory capacity via explicit, specialized training as
well as implicit use of multisensory input to trigger the most efficient WM functioning (Dehn, 2015).
To date, a great majority of tasks used to measure WM capacity in order to make predictions about aptitude and potential learning gains target capacity estimates of the
phonological (verbal) component of WM. At the same time, tasks which attempt to examine the visuospatial component of WM, such as the Corsi block tapping test (CBTT), have been mostly absent in WM-SLD research because its direct relevance to SLD has not been firmly established (Baddeley, 2015). In either case, WM span tasks attempt to assess the function of particular components of the (working) memory system in isolation from the other components of WM and as a result, may fail to capture the full scope and dynamic processing potential of the memory system in assessment across various contexts, including those that require simultaneous processing of both visual and verbal information. More specifically, (phonological) WM span tasks, for capacity estimates of verbalizable information, manifest more clearly (via verbal responses) than the same information processed, perhaps, visually (Dingfelder, 2006). Even more, the simultaneous processing of verbal, visual, and even tactile information is more
cognitively laborious than the processing of item occurrences in isolation (only verbal). In other
words, more complex language tasks would likely require a reallocation of cognitive resources,
which may require the visuospatial component of WM. Writing, for example, would likely more
heavily elicit the visuospatial component of WM as well as additional functions of the memory
system due to what Galbraith (2005) refers to as the “transformation of knowledge.” During the
writing process, the spatial component has a specific effect on the generation of new ideas at the
point of planning/organization as well as other language skills that are not (or can never be) fully
automatized (Galbraith, et. al 2005). In this way, it becomes important to consider what aspects
of language development (speaking, vocabulary, writing, reading, etc.) are being investigated with a particular kind of WM span task as well as investigating the degree to which the memory system collectively contributes to task performance as it varies in complexity. Thus, it has been proposed that WM is a dynamic system (Sáfár and Kormos 2008; Singleton 2014; Singleton 2017; Thompson 2013) and its (executive, verbal, and visual) subsystems do not exist in isolation and that components of memory capacity cannot be “tapped”; rather capacity abilities are “an emergent product of various components of cognitive and behavioral systems organizing themselves over a period of time within a specific task context” (Simmering, 2012, p 11).
In an attempt to understand the full scope of WM-SLD, it becomes important to ask: what is the aim of studying WM-SLD in the context of LLA? For research it may be helpful to
reconceptualize WM from a dynamic system theory approach (DST; Cameron and Larsen- Freeman 2007; De Bot, Lowie, and Verspoor 2007; De Bot and Larsen-Freeman 2011), primarily because of the varying degree of interaction among its components, all of which are themselves heavily dependent on internal and external resources, likely influenced by the sensory modality of the stimulus of the task at hand. As such, exploring WM from a DST
approach might help account for the wide variation in WM capacity estimates among individuals in certain task contexts.
From a more applied and pedagogical perspective, it has been proposed that
modifications and enhancements to L2 learning material, such as multiple modes of multisensory
input may ease the processing and integration load on the limited capacity memory system
(Dehn, 2015) and thus enhance the processing capacity of individuals across task contexts and
throughout development. Perhaps, WM functions most efficiently when individuals are exposed
to multisensory input, reflecting the multisensory nature of WM itself. From this perspective, the
discussion surrounding LLA and WM-SLD would do well to focus more heavily on the kind of input the system receives (i.e. educational conditions) rather than WM functioning rooted in innate abilities. Indeed, such modifications would more accurately reflect the multisensory environment through which individuals are accustomed to experiencing the world, which continually demands integration of new stimulus into the existing schemata in the face of interactive and converging cognitive processes (Skehan, 2012; DeKeyser 2012). Such an approach to WM is especially important in the context of complex tasks such as L2 writing, which demand a plethora of cognitive resources (Kellogg, 2001, 2006) that are, perhaps, less apparent in more simple tasks. Further investigation into the usefulness of multisensory input in the context of cognitively complex writing tasks, reconfigured within the dynamic,
developmental processes of WM-SLD might provide a more extensive, complete and practical
understanding of WM-SLD in the context of LLA, thus resulting in more effective pedagogical
interventions to L2 learning.
Part 1: Literature Review
Research Question(s)
In order for the theoretical framework of WM-SLD to have ecological value and aid in designing effective interventions for tasks which heavily elicit WM functioning, such as L2 academic writing, the cognitive and behavioral systems underlying language learning must be understood from a dynamic approach that considers individual differences, internal and external resources as well as the subsystems of WM and their collective contribution to language
development. As such, an alternative view of WM capacity that embraces the vast complexities of neural, cognitive, and behavioral systems interacting across task contexts may provide new insights into the sources of both variation and stability throughout development and task contexts.
Informed by a dynamic approach to WM (Simmering, 2012) and building on this approach, the current paper presents a proposal for future investigation which highlights the variability that persists in WM performance among individuals across tasks contexts, as well as emphasizes that the kind of input system receives in real time may better index WM functioning and subsequent task performance rather than innate ability measured by (verbal or visual) span tasks in the laboratory. The proposal calls upon existing findings from interdisciplinary and integrated approaches to WM and SLD in cognitive psychology as well as findings of SLD research with a primary purpose of exploring the different components of working memory and their collective contribution to complex cognitive processes as well as the sources of variation among individuals. To meet this aim, the present paper comprises two parts. In the first part, a literature review is presented, where gaps in the current literature are highlighted and interpreted.
This synthesis is then followed by a concrete proposal for future research, one that attempts to
further explore WM as a complex, dynamic system, where WM capacity does not function as a
compartment in the head, and where capacity is instead “an emergent product of cognitive and behavioral systems interacting in a real-time task context” and that emergence exists only through the meeting of the components of the system (Simmering, 2012, p. 11 ) In an attempt to overcome limitations that persist in assessment of WM, the proposal that is explicated in part 2 of this paper will encompass tasks to reflect the complexity of the system and realize the impact it has on real world behaviors, including those in language learning classrooms.
This thesis thus discusses the theoretical backdrop of WM-SLD and proposes a study design, which aims to incorporate the functioning of the less explored visuospatial component of WM into the current understanding of second language development, more specifically the role of the visuospatial sketchpad in L2 academic writing among (highly) proficient, adult
multilinguals.
1More specifically, by means of the literature synthesis (part 1) and the research proposal (part 2), this paper attempts to address the following research questions:
1. Is there an effect of multisensory input on writing task performance when compared to visuospatial WM span scores? (i.e. does multisensory input in WM-related tasks such as L2 academic writing offset [lower] WM span scores?);
2. If so, is the effect of multisensory input in WM-related tasks such as academic writing manifested in complexity measurements such as finite verb ratio (FVR) and mean word length (MWL)?;
1
High proficiency is measured by at least a TOEFL iBt 90 (with a minimum of 21 on all items), IELTS 6.5 (with a
minimum of 6 on all items), and/or Cambridge C1 Advanced or C2 Proficiency with a minimum score of 180.
3. Is the effect of multisensory input in WM-related tasks such as academic L2 writing manifested when qualitatively measured by a standardized writing rubric (see Appendix A)?;
4. If a relationship and/or an effect of input is found, what would be the best design to tap the multifaceted nature of WM, especially as it pertains to L2 pedagogy?
Hypotheses
As discussed above, the phonological component of WM has dominated the majority of research in language acquisition and development, and has been robustly shown to influence learning, especially at the early stages of language learning (Gathercole, 2006; Majerus et al., 2013). However, the role of the visuospatial component has received little to no attention in language related research, especially in an L2 context. In the event that L2 learning among adults does indeed demand additional cognitive resources (Wen, 2016), it is of particular importance to explore all the components of WM as part of an integrated framework, including the visuospatial sketchpad. Furthermore, an increasing amount of research from communicative language teaching (CLT) emphasizes the importance of multisensory input, which aids in rehearsal, a crucial component of WM (Bonsignori, 2017). However, standardized use of multisensory input is rarely used as a medium to facilitate SLD related WM tasks, nor is it regularly used in multilingual, academic environments.
In line with prior research surrounding the discussion of LLA and WM-SLD as well
as preliminary findings from a pilot study concerning visuospatial WM and SLD (see
section: Visuospatial Working Memory), it is hypothesized that there will be an significant,
positive relationship between visuospatial WM capacity estimates and L2 academic writing
(Kellogg, 2001, 2006). It is further hypothesized that there will be a significant positive
effect of input (i.e. multisensory tasks will produce higher academic writing scores than those that provide only written (visual) text as the primary source of input (Dehn, 2015).
Furthermore, it is important to at least consider potential increases in writing performance as it pertains to the quality of the writing as well as the complexity of the writing to inform future directions for pedagogical intervention strategies.
Working Memory Constructs Domain-general
Working memory (WM) as an element of cognition has been fiercely debated across disciplines since the early 1900’s and has emerged as perhaps the most important component of LLA (De Bot, 2008). More recent approaches to WM, especially in the context of
LLA and SLD, have sought to highlight the commonalities between the findings of both functional and structural neuroimaging
2as well as psycholinguistic analyses of WM capacity estimates, built up across disciplines over the past 100 years of inquiry (Wen, 2016). As a result, a number of characteristics of WM have now been firmly established: its limited capacity, increased function throughout adolescence, and variation of capacity estimates across studies, tasks, and contexts is well-documented (Simmering, 2012). Less well-established at the moment is the degree of interaction between WM components during different tasks, at which points and under which conditions in late-adolescence and adulthood WM continues to change, and the kind of environmental conditions that may lead to capacity increases (i.e. memory training and instruction) (Dehn, 2015; Huang, 2019).
2
Structural neuroimaging examines the structure of the brain (i.e. the contrast of different tissues such as
cerebrospinal fluid, grey matter, white matter, etc.) and functional neuroimaging is used to measure in-task brain
functions (i.e. localization of activation during a particular task) (Mourao-Miranda, 2005).
WM is often seen as part of executive functioning, an umbrella term used for several interconnected cognitive processes such as inhibitory control, planning, and self-monitoring (Baddeley and Logie, 1999). The complex nature of executive control makes its exact parameters uncertain, but there is general consensus of WM components. It is generally accepted that the central executive is a supervisory system responsible for allocating and controlling attentional resources necessary for problem solving and learning, and thus indexes a need for great
efficiency as individuals acquire, integrate, and store information (Goldstein and Naglieri, 2014).
The structure of the central executive is thought to consist of two main slave/buffer systems: the phonological loop and the visuospatial sketchpad. Most often, the phonological loop is thought to facilitate the temporary storage of sound-based information (i.e. hearing a word) via an articulatory rehearsal process (Wen, 2016) and the visuospatial sketchpad is responsible for the temporary storage of visual and spatial information (i.e. seeing an item in a particular spatial context). In 2002, Alan Baddeley proposed a fourth component to his original triage of WM: the episodic buffer. This addition would serve to temporarily hold, integrate, and link “chunks” of information across domains and multiple modalities from many sources. This relatively new inclusion addressed the apparent interaction between visual WM and verbal WM (see Figure 1) and shifted the discussion of WM away from isolated and strictly domain-specific visual and phonological subsystems and more towards the processes involved in the integration of
information. In doing so, the episodic buffer provides a better basis for addressing more complex
dimensions of executive control in working memory processes (Baddeley, 2011).
Figure 1. Working Memory
Representation of the working memory construct (Baddeley and Hitch, 1974) and Baddeley’s (2002) proposed interaction with long-term memory functions (bottom box).
Visuospatial Working Memory
Domain-specific theories of the four-component WM construct postulate that some elements of cognition develop independently of others and thus certain tasks rely on specific components of the memory system in isolation of the others (i.e. the phonological loop is for language learning and the visuospatial for navigational-type tasks) (Shah and Miyake, 1999). As such, the majority of research treats the visuospatial component of WM (VSWM) as a unitary, domain-specific system, largely unrelated to language processing. Contrastively, a
multicomponent, domain-general analysis of WM (Baddeley, 2002) attributes the most efficient
WM to the system as a whole, with structurally and functionally interconnected subsystems.
Although the measurable capacity of VSWM is most often unaffected by simultaneous rehearsal of small amounts of verbal information (Morey and Cowan, 2004), remembering larger amounts of verbal information, which may require a reallocation of resources for maintenance, has been shown to reduce the storage capacity of VSWM (Morey and Cowan, 2004; Saults and Cowan, 2007). Thus, VSWM capacity seems to reflect an executive resource that is important for remembering information in more than one memory domain, such as executive functioning (Kane and Engle, 2002). A pilot study conducted in pursuit of the current research proposal found there to be a significant positive relationship between VSWM via the Corsi block tapping task (CBTT) and novel word learning of Basque aided by (audiovisual) multisensory input r(18)
= 0.48, p = 0.03) among a sampling of Dutch/English bilinguals between the ages of 21-31 (Scherer, 2020). As such, the pilot study suggests that VSWM shares a common processing component with executive functioning and verbal working memory. Basque was specifically chosen for this task in order to prevent an advantage in the event participants had more/less exposure to other, more commonly spoken European languages (Spanish, German, or French) nor would participants be able to reply upon cognates shared among languages, which would aid in word recall ability. In this way, by making Basque the target language, the words in the recall task would be equally novel for all participants and ideally maximize reliance on WM
functioning. A crucial point of consideration is the fact that multisensory (audiovisual) input was
introduced to the word learning condition in order to facilitate the learning of Basque. Such
evidence further contributes to the notion of a domain-general approach to WM and that spatial
WM tasks might be helpful in understanding the full scope of language learning styles (i.e. visual
input to trigger schema) as well as understanding, more completely, the collective contribution of
both the spatial and verbal WM sub-systems. As such, more consideration given to the influence of VSWM in language learning may help to account for the wide variation in WM capacity estimates among individuals under certain task conditions as well as the changes to the language system over time. More in-depth theoretical approaches to the domain-general WM construct can be found in Conway and Kane, 2001; Engle; Kane, Bleckley, Conway, and Engle, 2001; Kane and Engle, 2002; Kane et al., 2004.
Multi-sensory Working Memory
Additionally informative to the discussion surrounding WM-SLD and its interconnected subsystems is a multisensory approach to WM, which may provide a more ecologically valuable understanding of how WM functions have evolved to maintain and manipulate information in a multisensory experience of the world. Indeed, cognitive philosophers and anthropologists have engaged in discussions about WM as a key aspect of the cognitive revolution, unique to human species (Wen, 2012). Such discussions suggest that the evolution of WM was/is largely
attributed to the need to perform sequence-dependent processes as well as accommodate and integrate novel stimuli into the memory system for survival (Kolodny, 2018). In this way, human cognition has evolved primarily to accommodate multisensory stimuli in meaningful, relevant, and high-stake, survival contexts. Thus, performance estimates in domain-specific, unisensory lab tasks may not be indicative of how WM has evolved to accommodate multisensory
perceptions of information. Instead, comparing performance estimates across task contexts, some
of which include multisensory stimuli, might better reflect WM functioning in real-life contexts
and such an approach might shed light on the WM processes throughout the developmental
trajectory of language acquisition.
Multisensory processing refers to the interaction of stimuli entering the cognitive system nearly simultaneously from different sensory modalities (Quak, 2015). To date, WM-related research has primarily focused on studying sensory perception and related WM components via unimodal and unisesory input. However, it has become increasingly clearer that information from different domains (i.e. visual and verbal) show more interaction in WM than one would expect on the basis of a strictly domain-specific perspective (Jiang et al., 2000; Prabhakaran et al., 2000). This is further illustrated by Baddeley’s (2000) inclusion of the episodic buffer to the domain-specific visuospatial and phonological storage components, which serves as a link between components to integrate memory traces that come from multisensory stimuli into a coherent perceptual scene (Baddeley, 2000). Furthermore, research on multisensory WM has shown that recall is better for cross-modal objects compared to modality-specific objects
3(i.e.
performance benefits for the recall of audiovisual information compared to the recall of only visual information) (Goolkasian and Foos, 2005; Delogu et al., 2009) and WM capacity estimates, under audiovisual conditions for example, are higher in tasks which include such cross-modal information, (Saults and Cowan, 2007; Fougnie and Marois, 2011). When
multisensory WM is coupled alongside a more integrated and functional understanding of more general WM approaches, it becomes more apparent that the senses (visual, verbal, auditory, and tactile) interact throughout the stages of input processing to form integrated, unified, and
meaningful representations (Quak, 2015). While the exact point at which a cross-modal stimulus is fully integrated into the memory system is unclear, it is apparent that cross-modal information does interact in WM more than what domain-specific functioning suggests. In either case, it
3
Cross-modal refers to perception that involves the interaction between two or more different sensory modalities.
Cross modal plasticity within the human brain is increasingly studied to gain a more thorough understanding of
large-scale and long-term cognitive functioning (Shams, 2010).
could be proposed that information from one modality can influence information processing in another modality in the form of necessary reallocation of cognitive resources for efficiency of processing and that information from different sensory modalities can also be combined into a single multisensory event - a process known as multisensory integration (Stein et al., 2010).
Due to ongoing and adaptive contributions to WM research, across disciplines, some WM constructs go beyond WM functions as responsible for temporary maintenance of novel
information, and extend long-term memory (LTM) contributions to WM by organizing and grouping information in a way that reduces the working memory load (Miller, 1956; Ericsson and Kintsch, 1995, mentioned in Cowan 2008). In this way, while WM deals primarily with the temporary maintenance of novel stimuli, the most efficient WM may also elicit long-term memory functions (again, see Figure 1). Consider the differential ease between recalling the series PIGJTYYTW and PTKHTESOL. It is likely that the ending of the second words to include TESOL would have elicited a function of the reader’s long-term memory, thus making the storage required for recall easier. In sum, what has emerged from a more interdisciplinary understanding, is that WM is a multicomponent and interconnected system (Baddeley and Logie, 1999), which regulates and actively maintains task relevant information in the face of converging cognitive processes, with a primary function of enhancing and facilitating the encoding, storage, rehearsal, and retrieval of familiar and novel information (Dehn, 2015; Wen, 2016), all of which are essential for higher level cognitive tasks such as second language learning.
Thus far, the review of WM-related literature has sought to demonstrate the ways in
which including WM capacity as one of the contributing factors of language learning aptitude
(LLA), has provoked a need for increased discussion and research into the variability and
complexity of WM and its role in second language development (Wen and Skehan 2011, 2012;
Wen 2015, 2016, 2019). Such research suggests that WM capacity greatly influences L2 learning, most especially when coupled with theories that suggest WM plays an even more influential role, or at the very least, a different kind of role in L2 learning than other higher level cognitive functions such as L1 acquisition
4(Wen, 2016).
Due to the complexity and variability of WM, it is informative to invoke a dynamic systems theory approach (DST) to WM-SLD. Unlike some language research that aims to isolate sub-systems of cognition and eliminate intra-individual variation as a means to generalize
findings to a larger population, DST-informed language research considers, firstly, that the language system itself is part of a larger (external) behavioral and (internal) cognitive system.
And secondly, that the intra-individual variation (“noise”) is a key to understanding, more holistically, the developmental trajectory of language learning and use in real-time contexts (Lowie, 2017). A DST approach to language is primarily concerned with the developmental processes of language development, which emphasizes complete interconnectedness of
variables
5, non-linear development, self-organization of the language system via interaction with the (language) environment, and dependency on both internal and external resources (Larsen- Freeman and Cameron, 2008). In other words, language “emerges” at the convergence of human cognitive ability (internal) and interaction with the language environment (external).
A closer look at WM-SLD from a DST perspective highlights influential cognitive factors unique to SLD, such as multicompetence and metalinguistic awareness among multilinguals and L2 learners (discussed below) but also the instructional conditions through
4
L1 acquisition is differentiated from L2 development where L2 development is an active and conscious process, requiring the retention of information and thus implicates a need for increased memory functions (Krashen, 1981;
Wen, 2019)
5
Variables refer to the cognitive, linguistic, and social systems, which contribute to the processes of language
development over time.
which many (adult) learners encounter a novel language. In this way, WM capacity might function differently and/or demand additional WM resources throughout L2 development and across different tasks than L1 acquisition.
Working Memory and Second Language Development: A Dynamic Systems Approach Development over time
Development of the system over time is a key component of a DST-informed approach to language learning. As such, development is a quality of language learning which emerges from the juxtaposition and interaction of multiple variables over time, influenced by exposure to correct and conventionally used forms as well as frequent application in particular contexts (Ellis, 2002). The degree to which specific variables within the developmental process of
language learning complement and/or compete with one another is often dependent upon internal and external circumstances, also susceptible to change over time (Verspoor, de Bot, and Lowie 2011).
Despite the vast amount of literature on WM, there is a great deal that remains unknown about the exact parameters of WM and even less about the degree to which different components facilitate second language learning independently and/or collectively throughout development.
While WM capacity does tend to show increased stabilization in adulthood (Montez, 2017),
more recent DST approaches to WM capacity and cognition indicate that increased second
language experience is associated with increased working memory capacity (Huang, 2019) and
that certain pedagogical approaches such as a mnemonic-based instruction
. (Dehn, 2015) may be
successful in raising working memory capacity or at least accommodating the memory system at
a particular point of development (Tsai, Au, and Jaeggi, 2016; Huang, 2019). This is significant
primarily because - if WM is indeed a multicomponent system, subject to development overtime, influenced by external variables, and conditionally amendable - a DST approach may prove to be the most helpful in accounting for all the variables that provoke and/or inhibit the most effective WM functioning and development.
In an attempt to investigate the continued development of LLA and WM over time, thus challenging preconceived notions of LLA and WM as stable constructs once an individual reaches maturity, Huang (2019) compared students studying at a Chinese university under one of three different language-learning conditions. The conditions consisted of first and second-year students majoring in English (only one language, but still an L2 for the Chinese student demographic) or majoring in L2+3, either English/Japanese, or English/Russian. Using a pre/post-test design, each subject’s LLA and WM was tested at two times across a nine-month interval. The LLAMA language aptitude test battery (Meara, 2005) was administered to test LLA and the Operation Span Task (OST) was used to assess WM span. In the research question that pertained to the changeability of WM, the results of the pre- and post-test scores showed significant increases among both L2 and L2+3 cohorts for both the WM OST as well as the LLAMA aptitude test over a nine-month interval. Additionally, the first-year students in the L2+3 condition outperformed their L2 counterparts in WM improvement. Interestingly, the effect of language learning intensity (L2 vs. L2+3) was found in the WM OST but not for the LLAMA aptitude test. Moreover, the effect of language learning intensity in the WM OST was found in the first-year cohort but not the second-year cohort. This might suggest that WM is more active and/or more amendable in the earlier stages of language learning prior to
automatization. Even so, Huang (2019) demonstrates that WM and language aptitude is indeed
subject to change over time and demonstrates performance benefits among individuals leaning
additional languages simultaneously. Such findings are in line with previous studies that also investigated developmental changes in WM over time (Klingberg 2010; Holmes, Gathercole, and Dunning 2009). In the specific context of the previously mentioned study, it could be concluded that multilingual learners have cultivated more cognitive resources throughout their language- learning trajectory and thus perform higher on language related tests such as the LLAMA;
however, the simultaneous improvement on the OST suggests general WM capacity increases alongside L2 and L3 language learning. As such, the discussion surrounding WM as a central aspect of LLA, where aptitude connotes an innate ability as well as stability, requires a reexamination, at least insofar as it pertains to adult L2 learners who may call on additionally cultivated cognitive functioning and/or resources (Wen, 2016). More specifically, a dynamic systems approach may shed further light on which skills and abilities exhibit increased development, perhaps at the expense of others, over the 9-month learning trajectory (i.e.
grammatical accuracy v. fluency).
Although not addressed in depth in Huang (2019), it would be interesting to further explore the context conditions and/or circumstances of the language learning environment that led to the attested enhancement of WM capacity over a 9 month period (i.e. the teaching
methods). Such an inquiry may shed light on the kind of stimulus that may provoke and enhance functioning of WM including memory training, intensive learning environments, or (as discussed below), multi-sensory modifications to input.
Prior experience: multicompetence and metalinguistic awareness
As L2 learning is often pursued later in life and into adulthood, there have been
somewhat strong correlations between WM span tasks and the results of various L2 proficiency
tests such as the TOEFL (de Bot, 2008), as well as an effect of prior language learning
experience on language aptitude as measured by the MLAT (Modern Language Aptitude Test)
(Dehn, 2015; Tsai, Au, and Jaeggi, 2016; Huang, 2019). This supports the idea/position that
there is indeed a relationship between WM capacity and language learning, but it is less clear
whether larger WM capacity estimates result in higher scores on language aptitude and/or
language proficiency tests such as the TOEFL or whether it is prior language experience that
results in larger WM capacity estimates instead. As there have also been positive correlations
between the number of languages an individual has learned and their WM span scores (Dehn,
2015; Tsai, Au, and Jaeggi, 2016; Zhang and Chang, 2018; Huang, 2019). Such findings are
significant primarily because they suggest that WM is not exempt from development over time,
but may be additionally enhanced under certain circumstances, such as increased language
experience (Jaušovec and Jaušovec 2012; Klingberg 2010; Economidou-Kogetsidis, 2012; von
Bastian and Oberauer, 2013) and educational intervention strategies (Dehn, 2015), such as
modified forms of comprehensible input (Krashen, 1981). These findings support the notion that
individuals who show higher levels of metalinguistic awareness, via the study of additional
languages, perform better on language-related tests such as the MLAT. However, such findings
may index a cultivated aptitude for language learning rather than an innate ability for language
learning. As such, there have been predominantly non-significant correlations between WM span
scores and the MLAT and further conclusions that language aptitude does not necessarily play an
important role in communicative language teaching (CLT) (Safar and Kormos, 2008). This lack
of correlation in some studies may indicate that WM capacity estimates are not necessarily
related to language learning aptitude as assessed by the MLAT and/or language ability in real-
time communicative contexts. In other words, while WM may indeed be related to language
learning, at least in as far as it is measured by proficiency tests such as the TOEFL, the degree to which WM translates to tests of general language learning aptitude and language use in
communicative contexts is unclear.
While WM capacity is equally significant and influential in both L1 and L2 learning, it is necessary to consider how cognitive processes may develop differently due to the time and type of input that the system receives throughout development. As such, applying the same
methodology used to explore the cognitive processes of L1 acquisition to L2 development may fall short when it comes to an accurate portrayal of WM and how it modulates second language development. One reason is that L1 learning is largely implicit, naturalistic, and procedural - seemingly effortless (Krashen, 1981, 1989; Cook, 2005, 2008, 2010). By contrast, L2 learning, which oftentimes occurs later in life, typically develops via explicit input involving conscious processes, which are themselves heavily influenced by the way the system has organized itself over time as well as external circumstances, including individual life experiences, teaching methods, and prior (language) knowledge (Skehan, 1989; McLaughlin, 1995; Dornyei , 2006).
Another influential aspect, unique to SLD and cognition is the idea of multi-competence,
which describes “'the compound state of a mind with two grammars” (Cook, 1991, p 112). There
is a great deal of evidence that supports hypotheses of concurrent activation of the L1 and L2 in
the bilingual mental lexicon during production (Levelt et al., 1999; Poulisse, 1999; Bialystok
2003; Costa and Santesteban, 2004; Cook 2012). Additional analysis of speech errors and code
switching abilities demonstrate that both languages are interconnected in the L2 learner’s mind at
all stages of language production; lexical retrieval, syntactic slotting and phonological encoding
(Wouter and Hartsuiker, 2017) and that proficient bilinguals have acquired cognitive skills to
suppress language interference that takes place through concurrent activation of L2. Such
research theories help to explain positive and negative language transfer during processing and production in ways unique to multilinguals. Furthermore, as linguistic awareness develops among multilinguals via the integration of new linguistic items (i.e. learning a second and/or third language), it becomes easier to integrate novel items into the existing language system(s).
For example, once an individual has mastered a second or even third language, learning a fourth or fifth language generally becomes less effortful (de Bot and Jaensch, 2015). This is perhaps most especially apparent where development is observed as a dynamic process among
multilinguals who possess additional languages and have cultivated additional cognitive resources. Furthermore, metalinguistic awareness and multicompetence may reflect long-term memory contributions of prior language experience to the learning of new items (Zion, 2019).
All of this demonstrates, most significantly, that while the L1 and the L2 are interwoven and subject to cross-linguistic influences, the cognitive resources and processes of the L2 learner are different and must be examined as such. The focus then becomes how to best understand WM as a complex, dynamic system in the unique context of multilingual cognition.
Individual differences
In addition to the cognitive and environmental resources that differ between first
language acquisition and SLD, there is also a great deal of individual differences in WM capacity
and task performance among language learners. Indeed, informal observation alone shows that
some people are able to learn a language faster, with greater ease, and more accurately than
others. Thus, it has been proposed, where all other aspects of L2 learning are equal, a larger WM
capacity is certainly advantageous as it likely allows for a more in depth analysis of information
and the ability to hold the information longer for efficient encoding, storage, and integration
(Wen, 2016). However, such a proposal is a result of predictive research that aims to explore how smaller WM spans in the context of LLA may be detrimental to successful language learning (i.e. lower WM span results in less efficient language processing). However, perhaps more useful in the context of individual differences, is an applied perspective that seeks to understand what accounts for the variation in span scores/task performance among individuals.
In this way, the most useful applied WM research aims to understand how the mind of the individual is working, adjusting, and benefiting from exposure to language, with a primary focus on intervention strategies such as multisensory input that accommodates the behavioral and cognitive circumstances of an individual.
Gardner 1995 suggests a view of “natural human talents”, captured in his Multiple Intelligences Model. Here, Gardner proposes eight types of intelligence: linguistic,
logical/mathematical, spatial, musical, kinesthetic, interpersonal, intrapersonal, and naturalistic.
He maintains that by considering primarily linguistic and logical/mathematical talent, the true
versatility and range of human cognitive ability in a particular task is restricted to the mode of
presentation as well as (limited) mode(s) of cognitive processing. In other words, a test that
evaluates linguistic talent in the absence of the interpersonal component of language processing
likely is an inaccurate portrayal of the task (i.e. language for interpersonal communication) and
at the same time an incomplete reflection of an individual's overall language ability. Intelligence,
in its traditional sense, may have little to do with success in second language learning, where
overall ability measurements account for and are dependent on the way in which the tasks are
presented and the restrictive context in which they are applied. As a result, the very definition of
what constitutes intelligence cannot be separated from the means used to measure it and thus, the
concept of intelligence indexes ability and aptitude only insofar as it has been assessed by
various aptitude and intelligence tests and tasks. After all, individuals with wide ranges of IQs, aptitude test results, and working memory span scores have proven to be successful at second language learning. Perhaps this inconsistency indexes that measurements for evaluation of intelligence and language learning aptitude rest largely on the shoulders of conceptually narrow modes of assessment and are significant primarily in language learning environments which often lack real-life (multisensory) contexts and application (Hagoort & Brown, 2000).
More recent studies show that intelligence measurements are related only to very specific abilities in second language learning, rather than an indication of overall language ability. There are higher correlations between intelligence and proficiency in cases where proficiency is measured by tests of reading comprehension, dictation, and free writing, but much lower correlations when proficiency is measured by tests of listening comprehension and free oral production (Mousa and Mahdiye 2018). In one study involving French language immersion students, it was found that intelligence measurements were indeed related to the development of French in the contexts of reading comprehension, grammar, and vocabulary, but unrelated to oral productive skills (Genesee, 1976). This suggests that intelligence measurements are considered to be an influential factor in what has been termed Cognitive Academic Language Proficiency (CALP), a kind of proficiency needed in context-reduced and cognitively demanding tasks, but less so in Basic Interpersonal Communication Skills (BICS), which consist of skills required for oral fluency and sociolinguistic applications (Ellis, 2008). Thus, intelligence may be a strong factor as it pertains to language analysis and grammar rules but is less pertinent in the context of communication and interaction (Candlin and Mercer, 2001). As such, lower
intelligence scores may impede second language learning, based on the demands of more
traditional (academic) learning environments but less often do lower intelligence measurements
impede learning, for example, in immersion settings (August and Hakuta, 1997). The fact that traditional language learning environments and modes of assessment are most often concerned with rules and vocabulary (i.e. learning about language), it is no wonder that those who learn through different modalities and whose language experience has been dominated by
communication and interaction in a multisensory context, for example, may exhibit lower proficiency levels via traditional modes of assessment. This is a point of crucial consideration if the ultimate goal of L2 acquisition is successful development of language skills across both academic and sociolinguistic contexts as well as a more thorough understanding of individual differences in language learning environments.
The discussion of intelligence and language learning aptitude has intensified with the inclusion of WM as an important component of language aptitude (Wen, 2016). This is best illustrated by an increasing consensus that WM capacity, as a cognitive indicator of aptitude, is susceptible to the influence of experience and instruction (Williams, 2012; Wen, 2012). While assessment of skills and ability is indeed helpful and necessary, the current linear modes of assessment (and teaching) risk exclusion of language learners whose “kind” of intelligence does not conform to the tests used to measure ability. Perhaps a clearer understanding of intelligence and LLA in the context of SLD will emerge when the tasks and tests used for measurement account for all of the cognitive channels in which language learners receive and integrate novel information in the presence of multiple converging cognitive processes. Consideration of
Gardner’s Multiple Intelligences Model may facilitate a better understanding of the full scope of
human talent, intelligence and aptitude, thus enabling researchers account for behavioral and
cognitive variation across task contexts as they pertain specifically to SLD-WM. For this reason,
the current paper is primarily concerned with experimental WM research in order to investigate
how task performance may be positively affected by different learning conditions (unisensory vs.
multisensory) to accommodate the various behavioral and cognitive differences of the individual across task contexts.
Variability across task contexts
WM capacity and task performance are highly dependent on and influenced by context. It has been proposed that a central limitation of WM-related studies is that they overlook theories of WM that specify how the multicomponent system organizes itself across contexts
(Simmering, 2012). From a dynamic systems approach, the variation across tasks contexts is attributed to the notion of a “view from below”, which exposes the ‘messy’ details: behavior that is variable and heavily influenced, either positively and/or detrimentally, by the nature of the stimulus and the context in which it is presented at a particular point in development (Thelen and Smith, 1994).
Variability across tasks contexts can be observed in both assessment of WM capacity estimates in laboratory settings via span tasks
6as well as language performance in different contexts. For example, an individual may exhibit highly proficient grammatical accuracy in writing, but struggles with verbal fluency in conversation. And both of these contexts regarding accuracy and fluency may not always correlate with the performance in tasks used to measure WM capacity estimates. As such, it is important to examine the ways in which WM capacity estimates are typically measured.
Despite the interconnectedness of the memory system (especially when it comes to L2 learning) and the individual differences and variability that surround the discussions of LLA,
6
WM span tasks make use of word lists, number series, and/or letters to be recalled in a specific order immediately
after presentation. The amount of correctly recalled items is considered to be an individual’s WM span score.
SLD and L2 learning, the majority of tasks which seek to measure WM span and item recall often target WM components and L2 learning stages in isolation. However, Kane et al. 2004 provides empirical support for the position that phonological WM span tasks, such as counting, operation, and reading span-type tasks, load on the same factor in a factor analysis as WM span tasks in which the tasks demand visuospatial processing and storage (Conway, 2005). A great majority of research has shown that the phonological component of WM, which maintains verbal and auditory information, is influential in the early stages of language learning such as novel word recall (Gathercole, 2006; Majerus et al., 2013) and has thus been the point of focus of all language related WM research. However, the most widely used WM span tasks designed to attend specifically to the phonological component of WM, such as the digit recall and reading span tasks call for item recall and order recall. These tasks place an extra cognitive load on the limited capacity system that is often not a point of consideration, nor is it accounted for in analysis (Gupta, 2004). These tasks are also heavily language dependent, making their use in L2 research linguistically biased and limiting. Alternatively, WM span tasks that are specifically considerate of WM as a complex, dynamic system and are invested in the connection between WM and L2 learning might be more useful in accounting for the variability that persists in assessment of WM capacity in laboratory settings as well as language use contexts, but such tasks are very rare.
In review of WM span task research, Simmering (2012) suggests overall increases in
capacity estimates with age but wide variation across studies, tasks, and domains. Of particular
significance is the fact that WM capacity estimates measured by span tasks often vary even
within a single age group. In one study pertaining to 7-year-olds performing simple verbal tasks,
estimates ranged from as low as 1.2 (Hulme et al., 1995) to 5.4 items recalled (Cowan et al.,
1999). Estimates for only one type of stimulus (digits) spanned 3.0 (Huttenlocher and Burke, 1976) to 5.4 (Cowan et al., 1999). This variability demonstrates substantial influence of task contexts on WM estimates even for studies designed to evaluate WM storage alone. It has also been suggested that variability in WM task performance, where there is an attempt to isolate a singular component such as verbal WM, may be a reflection, in some task contexts, of a cultivated ability to store and retrieve specific kinds of information (such as numbers or words) with greater ease under specific task conditions (Simmering, 2012). For example, a multilingual is likely to perform better in a non-word recall span task and a mathematician, perhaps better in a digit span task, but an explanation for the variation as well as the degree to which this actually reflects WM capacity/functioning and not merely a cultivated ability in an isolated context is unclear. As a result, laboratory tasks designed to tax specific components of WM to a greater or lesser degree by either adding specific processing demands or requiring only storage related tasks only rarely yield straightforward modulations of WM capacity (Simmering, 2012). Indeed, such tasks may show a very specific cognitive ability but not as it contributes to the overall functioning of memory capacity (Case, 1995) in real time nor with tasks of increased complexity.
Alternatively, a dynamic approach would suggest that real-time behavior emerges via the interactions of its components in context and the tendency of a dynamic system to exhibit stability across task contexts depends on the kind of interactions among its components (Simmering, 2012).
Advances in understanding the cognitive processes and behavioral influences involved in
WM as a complex, dynamic system requires a reexamination of the tasks used to measure WM
capacity. Such a reexamination may result in reformed pedagogical approaches and assessment
models that have overlooked the full scope of WM and its relationship to SLD. From this
perspective, variability across task contexts can be better understood as the result of interaction between subsystems (both internal and external) involved in the developmental process of language learning (de Bot, Lowie and Verspoor, 2011). Due to the fact that development of the system and its interconnected subsystems are heavily reliant on a limited capacity (cognitive) system (Wen, 2012), careful examination of the variables and their interaction in instances of real-time language use, in contexts that fully tax WM capacity (rather than numbers and words in isolation) would provide a broader understanding of the variables themselves as well as their interactive role in an ecologically valuable task context such as L2 academic L2 writing (de Bot, Lowie and Verspoor, 2011).
A Dynamic Systems Approach to Writing
Models of writing processes and SLD-WM
Examination of the writing process is an ideal context to observe WM as a complex dynamic system (DST), as L2 academic writing is a highly complex task and requires a
culmination of skills, both behavioral (external) and cognitive (internal), cultivated throughout
the developmental trajectory of language learning (i.e. semantics, syntax, pragmatics, planning,
organization, revision, etc.). Indeed, research has shown that L2 writing places a unique and
heavy demand on the WM’s limited capacity to simultaneously attend to multiple task-related
ideas (Kellogg, 1994). This is in addition to managing concurrent activation of both the L1 and
the L2 in the multilingual lexicon throughout the stages of writing. As such, the primary concern
regarding the writing process, in the context of WM-SLD, is to understand firstly how cognitive
(memory) functions of the multilingual collectively contribute to the L2 writing process and,
secondly, explore potential mnemonic strategies such as multisensory writing prompts (tasks) to facilitate the most effective (behavioral) performance (Dehn, 2015).
Two cognitive, theoretical approaches to the writing process can be distinguished. The first is Hayes’ and Flowers’ 1980 model, which originally proposed that the writing process consists of stages pertaining to planning, translating and revising. Hayes (1996) later proposed a modified model, in which he attributed an essential role to long-term memory in the writing task, for sources of knowledge related to the writing task (McCutchen, 2000, mentioned in
Bergsleithner, 2010). In later years, and building on Baddeley and Logie’s (1999)
multicomponent model of WM, Kellogg’s (1996) model of the writing process focused on the role of WM during the stages of formulation, execution and monitoring. More specifically, the limitations of the central executive WM component and its phonological and visuospatial slave- systems as well as their collective contribution to formulation, execution, and monitoring.
Galbraith (2005) suggests a different, but significant interaction between VSWM (visuospatial working memory) and the central executive, with VSWM having a specific effect in the formulation stage of writing. It has also been concluded that individual differences in VSWM capacity may predict an individual’s ability to retrieve verbal information from long-term
memory (Shipstead, Lindsey, Marshall, and Engle, 2014). Kellogg (2001, 2006) further suggests that writing involves multiple stages of representations and cognitive processes, which are likely constrained by an individual’s WM capacity (see Figure 2). While it is thus clear that WM does modulate the L2 writing process, its precise role deserves to be investigated in more detail, especially in relation to the different stages of writing.
The formulation stage of the writing process involves planning (the writer’s goal and
organization of ideas) and then translating (coding such ideas into various lexical items and
syntactic choices). The execution stage includes the production of ideas into coherent language via the motor system (most likely handwriting, typing, and/or other tactile movements). The final stage of the writing process is monitoring (the writer’s reading and rereading of the text) while simultaneously editing (writer’s attention to linguistics and/or organizational errors) (Ellis, 2005). At each stage of the writing process, to varying degrees, WM is active, and the degree of activation at each stage is dependent on the nature of the task at hand. For example, if a time constraint is imposed, the writer might have to reallocate cognitive resources, and place more of a focus on organization of ideas at the expense of linguistic accuracy.
In general, skilled writers require less cognitive effort to manage attentional resources that trigger syntactic and semantic information necessary for writing performance (Kellogg, 2001). On the other hand, less skilled/experienced writers need greater cognitive effort to produce a text (Abu-Rabia, 2003, McCutchen, 1996). Crucial to task performance is that when an individual has sufficient previous knowledge about a topic, a lower cognitive effort in writing performance is needed (Kellogg, 2001). This is of particular interest, as it pertains to the kind of input the writer receives just prior to the stages of formulation and, perhaps, the way in which the information is presented to trigger the most efficient WM functions. In other words, further exploration of multisensory writing prompts (tasks) in the context of academic writing (and assessment) may provide useful insights related to emergent properties of WM (DST:
Simmering, 2012) in a meaningful context, especially where input is enhanced to accommodate
varying degrees of cognitive and behavioral strengths and weaknesses in writing abilities among
individuals. Therefore, it seems critical to examine whether certain task types, especially more
complex ones (i.e. academic writing) which integrate writing with additional skills such as
reading (visual), listening (auditory), and/or tactical navigation might enable the most efficient WM functions and thus facilitate and/or enhance writing performance.
Figure 2. Working Memory and the Writing Process
A representation of Kellogg's (1996) extension of Baddeley's model of working memory and the writing process.
Multisensory/multimodal
7input and SLD
Specific to the cognitive channels for learning, modalities are the sensory pathways through which individuals transmit, receive, and store information (Quak, 2015). These modalities are visual, auditory, and tactile. New technologies, especially computers, mobile
7
