Working with working memory and language

Working memory (WM) in psychology is generally understood to refer to a temporary storage facility available during online processing and subject to strict capacity limitations, the nature of which is open to different interpretations. This special issue on working memory in second language acquisition (SLA) comes at a time when the field is becoming increasingly outward looking. For example, researchers who formerly focused exclusively on theoretical linguistic aspects of second language development are now regularly taking into account research into real-time performance. Already there are numerous studies in the recent second language acquisition and bilingualism literature where working memory is either the focus or an important part of the investigation. Indeed, an interest in the role of memory in SLA is not completely new (see, for early examples, Cook, 1977, 1979). Memory just has to be a vital component of any theory of learning since without the ability to retain any changes in knowledge or ability, all learning would be rendered ephemeral and therefore useless. Working memory, as it is generally understood, is also crucial to understanding how individuals are differentially able to exploit their current resources in one or other of their languages irrespective of any longer term effects. It is odd in retrospect how long memory has been taken for granted in second language (L2) research and how questions regarding the precise nature of language learning mechanisms have been set aside for future consideration.

Increasing levels of cross-disciplinary engagement pose a significant challenge to busy SLA researchers; hence the need for regular updates on the state of the art. For this reason, the contents of this special issue not only include three interesting new empirical studies, the first of which is actually a meta-analysis based on empirical studies (Grundy and Timmer, 2017; Kim and Christenson, 2017; Sagarra, 2017). Also included are two shorter discussion papers on theoretical issues relating to the nature of working memory: Baddeley reflects on the development of his own well-known views on WM while Truscott proposes a quite new approach, each author being concerned with the notion of modularity but in a quite different way (Baddeley, 2017; Truscott, 2017). More extensive collections of L2 working memory studies are discussed in Rankin’s review article that brings this special issue to a close (Rankin, 2017).

The more researchers in language development and performance apply working memory and associated concepts to the design and interpretation of their empirical investigations, the more it is important to acquire a basic familiarity with the relevant debates in psychology and neuroscience. For some years now a seminal anthology by Miyake and Shah (1999) has for many been a useful source providing a big picture of the various contemporary approaches to working memory. There are now more recent examples where different approaches are discussed; see, for example, Cowan, 2015; Eriksson et al., 2015; Oberauer et al., 2016 (see also Sagarra, 2017).

Oberauer et al. distinguish three basic views on the nature of WM capacity, each of which may be further subcategorized, and which they characterize using the following basic categories: (1) decay, (2) resources and (3) interference (Oberauer et al., 2016: 761–65). The first type is based on the idea that WM representations decay rapidly over time. Representations in WM can still be refreshed in various ways, subvocal articulation being a phonological process already familiar to students of Baddeley’s work (Baddeley et al., 1975). The second type of approach revolves around the idea of limited resources: the authors single out for special attention a more constrained version, namely ‘slot theory’, according to which the resource underlying short-term maintenance of information consists of a limited number of discrete units or slots that can be allocated to individual items or chunks (Fukuda et al., 2010). The third and final type of approach is interference whereby representations in WM at any given moment interfere with one another: compensatory mechanisms include the filtering out of irrelevant material. The authors consider all three approaches in turn as well as combined approaches. Awareness of critical debates of this kind should provide less seasoned L2 researchers with an incentive to keep pace as much as possible with WM research and avoid the danger of relying on theories, concepts and measuring instruments that may have been updated several times over or even superseded by alternative versions since they were originally proposed.

Measures to test WM differ both with regard to the assumed type of memory they are targeting, e.g. digit-span tests versus test of verbal working memory, and with regard to the assumed cognitive load placed on the participants (for a discussion of taxing versus non-taxing tests, see Sagarra, 2017): all these tests make theoretical assumptions about the nature of working memory and its role in the mind as a whole; as already mentioned, there is at present nothing like a consensus about the answer to this fascinating but very challenging question no more in fact than with regard to questions about how lexical, grammatical and other aspects of linguistic ability are represented in the mind of the language user.

Returning now to contributions in the present issue, the opening discussion paper in the theoretical issues section is by Baddeley who, apart from being a pioneer in working memory research is also the one with whose multicomponent model of WM most SLA researchers are probably most familiar. Here, he presents his current views on the modularity of working memory and an account of the path he has taken to get to his current position which includes an additional, fourth component, the episodic buffer. Working out the role of this particular component has occupied Baddeley and associates for quite a few years. In the concluding part of his contribution, he discusses how his conceptualization of working memory relates to other ones discussed in the literature, including that of Nelsons Cowan, frequently contrasted with his own, and he notes how, in both his and other approaches, strict Fodorian modularity (Fodor, 1983) has given way to less rigid ways of defining functionally distinct systems.

Following on from Baddeley’s contribution, comes Truscott’s discussion paper in which he discusses alternative approaches to working memory and makes a special case for a research programme that is based on a particular modular conceptualization of the mind as a whole. This view of the mind contrasts with the common assumption he sees in many contemporary approaches to memory that the mind is largely homogeneous. Baddeley’s model of working memory is indeed modular but this relates specifically to its internal architecture and not to the architecture of the mind as a whole. Truscott advocates an approach that is in tune with the state-based views of Cowan (see related discussion on the differences between this position and Baddeley’s in Baddeley, 2017): in Cowan’s approach, working memory is not separate from long-term memory but is simply a subset of elements in long-term memory that are currently in an activated state (Cowan, 2015). However, in the wider perspective of the mind as a whole, Truscott’s domain-specific memory is reflected in all the various functionally distinct systems that constitute the mind as a whole, each of which has its own memory store. Furthermore, if all these components of the mind were taken into account, argues Truscott, the internally modular make-up of working memory proposed by Baddeley and associates would require considerably more components than the few it currently contains. However, in addition to the memory that resides in each functionally distinct mental system, Truscott also proposes a domain-general global working memory (GWM), which is a combined product of the multiple local memory stores and, importantly, is conscious and effortful. This is closely related to the conception of WM.

The opening contribution to the next, empirical studies section is Grundy and Timmer’s meta-analysis, not strictly speaking an empirical investigation itself but which is nevertheless based on a range of investigations that involve working memory in bilinguals and monolinguals. These authors, faced with mixed results from studies with regard to the effects of bilingualism on working memory capacity, conducted an analysis of 27 independent studies, 19 of which have been carried out since 2010. As their model of WM, they adopt the proposals of Eriksson et al. (2015). These are broadly speaking in the Baddeley tradition, seeing working memory as a multi-component processing system. Their expectation for bilinguals operating with more than one language is that practice over the lifetime in attentional control, selecting and maintaining information in memory, will give them greater WM capacity compared with monolinguals. Overall their analysis supports this expectation especially, but not exclusively, with regard to bilingual children. They conclude their contribution with recommendations for further research and particularly into the effect of task type and a number of linguistic variables.

Sagarra reports on a longitudinal investigation focusing on L2 grammar and reading ability. Her study investigates whether WM influences L2 grammar and reading development in early-stage classroom learners. These findings reveal longitudinal WM effects on L2 grammar and reading development at early acquisition stages. The more taxing tests, imposing increased processing demands on participants, proved to more effective in yielding longitudinal WM effects. One of the conclusions Sagarra draws is that WM plays a more prominent role in lower as opposed to higher proficiency learners. Her results also supply more evidence that task explicitness, as required for example in grammaticality judgement tests, is not responsible for WM effects. She also asserts that her results support resource-sharing views of WM that see cognitive load not in terms of its complexity but in terms of to what degree the attentional resources that are required to execute a given task impose limits on accompanying processing activities such as retrieval and refreshment of memory.

Kim and Christenson’s contribution looks at the L2 processing of globally ambiguous relative clauses by Korean learners of English, that is where the ambiguity remains and is not resolved by the end of the sentence. In contrast to Sagarra’s study, their L2 learners were highly advanced. The model they were applying was the Capacity Constrained Parsing Model (MacDonald et al., 1992). Participants’ working memory capacity was measured by a reading-span test and a paragraph-decision task. The aim was to see how individual working memory capacity at this proficiency level affects processing, and whether these effects differ from those reported in the literature on first language (L1). The authors conclude that there are no qualitative differences when L1 and L2 processing are compared; in both cases just those individual readers who have high WM capacity slow down at a point where the global ambiguity becomes detectable and show themselves able to maintain different possible interpretations in working memory. This provides them with the necessary cognitive resources to integrate various types of information during processing. The authors conclude that the Capacity Constrained Parsing Model is supported by their results and is also able to predict the performance not only on temporarily ambiguous structures but also on globally ambiguous ones. Furthermore, this prediction applies to learners at very advanced levels of proficiency.

In the review article section, Tom Rankin discusses two anthologies, namely Wen et al. (2015) and Altarriba and Isurin (2012), both of which are recommended by him as very useful sources for L2 researchers interested in language processing (Rankin, 2017). Rankin chooses to concentrate on issues relating to L2 processing and development. He singles out Wright’s survey of the field and Wen’s promising contribution to the first anthology as being particularly relevant to the concerns of second language acquisition researchers, although questions of its potential scope and applicability are left unanswered (Wen, 2015; Wright, 2015). In further comments on the first anthology, Rankin notes the relative lack of space devoted to L1 effects, which are highlighted only in Williams’ final contribution (Williams, 2015). The need for a greater discussion of crosslinguistic effects is in fact a general comment on both volumes. The second anthology by Altariba and Isurin targets a wider audience. One feature of the volume that concerns Rankin is the undifferentiated use of the term ‘bilingual’. Further specification of this term is needed to work out potential differences in processing effects. Other contributions that are deemed very useful and particularly noteworthy of praise are the contribution by Bartolotti and Marian (2012), contributions by Szmalec et al. (2012) and by Trofimovich et al. (2012); the latter two receive more extensive commentary. Rankin’s general conclusion reflects a main theme of this special issue as a whole, namely that researchers from different fields where memory is of vital concern need to continually keep abreast of each other’s research in order to benefit their own theoretical and empirical investigations.

In the light of the contributions to this issue, it seems reasonable to conclude that the state of the art on working memory should encourage researchers in L2 acquisition to keep a close eye on relevant developments both in psychology and also in neuroscience. Appropriate caution is advised when making claims about empirical findings that rely at least in rely on results from standardized tests of WM, since these tests will have theoretical implications that need to be taken into consideration (Oberauer et al., 2016). It is also the case that a serious consideration of current memory research will undermine arguments for a rigid separation between psycholinguistic studies of L2 performance, on the one hand, and studies of L2 development over time, on the other: the current trend both in the literature and especially in research meetings to discuss these two dimensions of second language research together is clearly a healthy one. To conclude this introductory editorial contribution, I would like to record a special debt of gratitude not only to the contributors to this special issue for their patience while the issue was assembled over some considerable period of time, but also to the reviewers for the vital service they have provided; and, finally, to the three editors of Second Language Research who have offered valuable support, comments and useful suggestions along the way.