Exploring the Relationship Between Modified Output and Working Memory Capacity

Language Learning ISSN 0023-8333

Exploring the Relationship Between

Modified Output and Working Memory

Capacity

Alison Mackey

Georgetown University

Rebecca Adams

University of Auckland

Catherine Stafford

University of Wisconsin–Madison

Paula Winke

Michigan State University

This study examines the relationship between learners’ production of modified outputand their working memory (WM) capacity. The task-based interactions of 42 college-level, native English-speaking learners of Spanish as a foreign language were examined.A relationship was found between learners’ WM test scores and their tendency to modifyoutput. Specifically, greater processing capacity was related to greater production ofmodified output during interaction.

Keywords second language acquisition (SLA); modified output; working memory ca-pacity; interaction; task-based research; corrective feedback

Participation in conversational interaction has been argued to facilitate secondlanguage (L2) learning in a number of ways (Gass, 1997, 2003; Gass & Mackey,2007; Long, 1996, 2007; Mackey, 2007; Pica, 1994). A good deal of empirical

The authors would like to thank Harriet Wood Bowden, Kara Morgan-Short, Nina Moreno, and

Takako Egi for their assistance with this article. We would also like to thank Robert DeKeyser

and the anonymous Language Learning reviewers for their insightful comments. Any mistakes,

however, are our own.

Correspondence concerning this article should be addressed to Alison Mackey, Department of

Linguistics, Georgetown University, Washington, DC 20057. Internet: [email protected]

Language Learning 60:3, September 2010, pp. 501–533 501C© 2010 Language Learning Research Club, University of MichiganDOI: 10.1111/j.1467-9922.2010.00565.x

Mackey et al. Modified Output and Working Memory Capacity

work, carried out with both children and adults in classroom and laboratorysettings, has demonstrated the benefits not only of interaction as a whole butalso of a variety of specific interactional processes (Braidi, 2002; de la Fuente,2002; R. Ellis, Tanaka, & Yamazaki, 1994; Gass & Varonis, 1994; Han, 2002;Iwashita, 2003; Leeman, 2003; Mackey, 1999; Mackey & Philp, 1998; Mackey,Gass, & McDonough, 2000; Mackey, Oliver, & Leeman, 2003; McDonough,2005; Oliver, 1998, 2000, 2002; Philp, 2003; Storch, 2002; for a summary,see the meta-analysis by Mackey & Goo, 2007). Long’s (1996) interactionhypothesis proposes that corrective feedback obtained through interaction pro-motes interlanguage development because it “connects input, internal learnercapacities, particularly selective attention, and output in productive ways” (pp.451–452). Current theoretical work in second language acquisition (SLA) sup-ports this notion, suggesting that the feedback that learners receive on theirutterances during interaction can provide an important means of encourag-ing these connections and promoting learners’ modifications to their linguisticoutput (Swain, 1995, 1998, 2005).

Modified output during interaction often occurs when a learner is providedwith corrective feedback on a given utterance and follows this feedback bymodifying his or her original production. The linguistic focus of the correctivefeedback may vary; for example, feedback can be provided on grammaticalaccuracy, lexical choice, or communicative success. In most sorts of interac-tions, whether with instructors or peers in classrooms, or with native speakersoutside classrooms, corrective feedback occurs in a variety of forms as well. Forexample, it may be provided as a simple repetition of a learner’s incorrect utter-ance, as a clarification request, where an interlocutor asks for a reformulationto help better understand an utterance, or as a recast, in which an interlocutorprovides a more targetlike reformulation of what the learner said. Correctivefeedback is always reactive, occurring in response to a learner’s production at-tempts. Corrective feedback often provides learners with opportunities to focuson the linguistic aspects of their output and sometimes learners modify theiroutput following such feedback. Different types of corrective feedback havebeen claimed to differentially affect L2 learning (e.g., Ammar, 2008; Ammar &Spada, 2006; R. Ellis, Loewen, & Erlam, 2006; Lyster, 2004). Lyster (2004), forexample, distinguished between feedback moves such as prompts that requirelearner responses (i.e., clarification requests, repetitions, metalinguistic clues,and elicitations) and feedback moves such as recasts for which learner responsesare optional. He found that there was some advantage for prompts over recasts.

Modified output following corrective feedback is one of the foci of thecurrent study. Example 1 (taken from Mackey, 2007) illustrates modified output.

Language Learning 60:3, September 2010, pp. 501–533 502


Two nonnative speakers (NNSs) are collaborating on a task that involves findingroutes on a map. After producing an initially problematic utterance, “turnanother side,” NNS 1 receives feedback that appears to result in a realizationthat her utterance was not understood. She reformulates her initial utterance,modifying her output from “turn another side” to (the more targetlike and morespecific) “turn to the left side.” NNS 2 comprehends and the conversation (andtask-completion process) continues.

Example 1. Feedback and modified output.

NNS 1: In front of library, turn another side from grocery store.NNS 2: Which side from the grocery? ← clarification requestNNS 1: Ah, er turn to the left side. ← Modified outputNNS 2: Ok turn left, I did it, now which way to turn?

A number of researchers have argued that modified output plays a role in theL2 learning process (e.g., R. Ellis & He, 1999; McDonough, 2005; Shehadeh,2002; Swain, 1985, 1995, 1998). Empirical studies have pointed to a relation-ship between modified output and L2 learning (McDonough, 2005; Swain,2005). It is important to note that researchers examining the question of howmodified output works have argued that positive developmental effects occurregardless of whether a learner’s modification is more or less or equally as non-targetlike as the original utterance (Gass, 1997; Swain, 1995, 2005). It may bethat the process of producing modified output is useful for learning because it(a) encourages the cognitive processing of both morphosyntax (grammar) andsemantics (meaning) (Swain, 1995), (b) strengthens existing knowledge repre-sentations (Nobuyoshi & Ellis, 1993) and promotes automaticity (Anderson,1993; de Bot, 1996; N. C. Ellis, 2005; Logan, 1992), (c) serves as a site forhypothesis testing about language (Gass, 2003; Mackey, 2002; Swain, 1995,2005), and/or (d) indicates (or encourages) a level of metalinguistic aware-ness facilitative of language development. Several researchers have claimedthat modified output operates by drawing learners’ attention to grammaticalstructures, making them salient and thereby creating a context for L2 learn-ing (Izumi, 2002; McDonough, 2005; Swain, 1995), and that this can occurimmediately or in subsequent turns (McDonough & Mackey, 2006). In moregeneral terms, Swain (1993) has proposed that output is important for learningbecause learners “need to be pushed to make use of their resources; they needto have their linguistic abilities stretched to their fullest; they need to reflect ontheir output and consider ways of modifying it to enhance comprehensibility,appropriateness, and accuracy” (p. 160).

503 Language Learning 60:3, September 2010, pp. 501–533


One thing that is clear from previous research is that some learners modifytheir output during interaction and others do not. The same learner may varyresponse types at different times within the same interaction, as was the casein McDonough and Mackey (2006), for example. Suggestions to the effect thatmodified output may promote interlanguage development in several differentways (e.g., Swain, 1995, 2005) mean that there is still much research to be donein this area. It has been proposed, for instance, that modified output can allowlearners to reflect upon their own utterances in light of native-speaker feedbackand that modifying their output helps them to notice the gap (Schmidt & Frota,1986) between what they know of their developing L2 and what remains tobe acquired. Izumi (2002) has suggested that beyond noticing simple surfaceforms, output can also serve “effectively both as the stimulator of integrativeprocessing and as the glue to connect individual form elements” (p. 571).

In addition to encouraging learners to notice the gap, the production ofoutput may also push them to notice the place of structures in the linguisticsystem. This may, in turn, help learners to test their interlanguage hypotheses,allowing for modifications of these hypotheses as the interlanguage develops.Modified output might also allow for more elaborated processing of linguisticinformation that has been taken in, but that learners have not yet begun toproduce spontaneously. As Swain (1998) has claimed, modified output mayreflect the “cutting edge” of a learner’s interlanguage knowledge.

In the current study, we sought to extend the output line of research andclarify the role of modified output in learning by investigating the internalfactors hypothesized by Long (2007) as important. We focused specifically onthe relationship between learners’ working memory capacities and modifiedoutput. Because working memory (WM) is posited to involve the ability toregulate attention during the performance of complex cognitive tasks (as will bediscussed below), it seems that this internal factor might be related to learners’ability to attend to corrective feedback, to modify their output accordingly,and, thus, to benefit from an interactional learning context. The next sectionprovides an overview of previous L2 research on WM capacity.

Working Memory and L2 Learning

In this study, we followed Miyake and Friedman’s (1998) definition of workingmemory as “a computational arena or workspace, fuelled by flexibly deployable,limited cognitive resources (or activation) that support both the execution ofvarious symbolic computations and the maintenance of intermediate productsgenerated by these computations” (p. 341). In line with this definition, we



assume that WM encompasses the dynamic and simultaneous processing andstorage functions of memory (Baddeley, 2007).

Daneman and Carpenter’s (1980) sentence span task, or some variation ofit, is often used in WM research. This sort of task requires participants to eitherread or listen to series of sentences, processing the verbal input in some waywhile attempting to maintain specific verbal material in short-term memory.Executive functioning is important here; the input needs to be filtered andselectively attended to, and activation and rehearsal of information needs to becontrolled in a way that minimizes depletion of the limited attentional resourcesthat are available.

In first language (L1) research, relationships have been found between WMcapacity and such constructs as fluency of speech (Daneman, 1991) and theability to learn new words from context clues (Daneman & Green, 1986).Withinthe framework of cognitive and psycholinguistic accounts of SLA, WM capacityhas been ascribed an important role in the processing of verbal input andL2 learning (e.g., Ando et al., 1992; N. C. Ellis, 1996, 2005; Erlam, 2005;Geva & Ryan, 1993; Harrington & Sawyer, 1992; Mackey, Philp, Egi, Fujii,& Tatsumi, 2002; Miyake & Friedman, 1998; Palladino & Cornoldi, 2004;Payne & Whitney, 2002; Robinson, 1995, 2002, 2005; Sawyer & Ranta, 2001;Segalowitz & Lightbown, 1999; Skehan, 1998; Tokowicz, Michael, & Kroll,2004). In fact, Miyake and Friedman (1998) suggested that because research hasshown that individual differences in the availability of WM resources influencethe efficiency and quality of real-time L2 processing, WM may be central toL2 aptitude. Others have also stressed that working memory may be the keyto understanding L2 aptitude better (N. C. Ellis, 2001; Sawyer & Ranta, 2001;Skehan, 2002).

In the context of interaction research, WM capacity is quite relevant. Theinteraction hypothesis proposes that learner-internal processes such as thoseinvolved in selective attention mediate between input and learning. The pro-vision of feedback and production of modified output during interaction mayhelp to focus learners’ attention and memory resources on what is particularlyrelevant to advancing their L2 development. Many empirical studies of interac-tion have reported that learners differ in whether and how much they are ableto learn from feedback, and existing studies of output note interlearner and in-tralearner variation (McDonough & Mackey, 2006). Payne and Whitney (2002)have suggested that as learners attempt to understand and produce language,their internal processing involves holding representations of the input in short-term memory while retrieving information about L2 grammar from long-termmemory. At the same time, learners need to relate the input to both L2 and



world knowledge and to make choices to determine what new information willbe incorporated into their existing knowledge.

As a number of researchers including Gass (2006) have pointed out, adultlearners are exposed to more input than they can process, and they need amechanism to help them sort through the input during ongoing conversationalinteractions. The claim that receiving feedback and subsequently producingmodified output creates contexts for learning through drawing attention togrammatical structures implicates individual differences in WM as a potentiallyimportant factor in L2 development. Claims about the relationship between WMand SLA have also been made by Skehan (2002), who discussed a variety of L2aptitude constructs and suggested that the noticing and pattern identificationstages of SLA are related to attention management and WM operations.

In empirical studies that examine the communicative interaction-L2 learn-ing relationship and the role of WM capacity in that relationship, Ando et al.(1992) and Mackey et al. (2002) found that learners with higher WM capacityexhibited more delayed benefits from communicative instruction and feedbackthan their counterparts with lower WM capacity. Ando et al.’s (1992) studyexplored the relationship between learners’ cognitive capacity and how muchthey benefited from different instructional approaches. They assessed 90 L1Japanese fifth graders’ verbal WM capacity by means of L1 and L2 sentencespan tasks (both reading span and listening span). Looking at immediate anddelayed posttests, which the learners took immediately and 2 months follow-ing 9 hours of explicit form-focused instruction, they found that WM capacityreliably predicted the participants’ relative success in L2 English learning;however, this relationship did not emerge until the time of the delayed posttest.Another interesting finding was that learners with low WM capacity benefitedmore from a more implicit communicative approach involving learner-learnerinteractions than they did from an instructional approach that involved explicitlyform-focused production activities without communicative tasks.

The findings of Mackey et al. (2002) are compatible with those of Ando et al.(1992) in that significant relationships between L2 outcomes and WM capacitywere observed only in delayed test results. Mackey et al. investigated the rela-tionships among learners’ WM capacities, their noticing of interactional feed-back, and their subsequent L2 development. They assessed the L1 and L2 WMcapacities of 30 adult Japanese learners of English who received feedback ontheir question use during dyadic, task-based, communicative interactions withnative speakers. Interestingly, the learners with lower WM capacity appearedto benefit more immediately from communicative interaction; all six of thelow WM learners showed development on immediate posttests. Nevertheless,



they did not maintain this development on the delayed posttests 2 weeks later.In contrast, the learners with high WM capacity did not tend to demonstrateinitial progress but did show improvement on the delayed posttests. Mackeyet al. (2002) suggested that although the learners with higher WM capacity mayhave taken longer to “consolidate and make sense of the feedback given them,reflecting change only after an interval” (p. 204), they may have been betterable to make cognitive comparisons between the target language forms andtheir own versions of question forms, with lasting effects for L2 development.On the other hand, the learners with lower WM capacity may not have beencognitively comparing the target language forms with their own forms in thesame manner as those with higher WM capacity. In other words, they may nothave processed the input provided in the form of feedback in a sufficientlyelaborative way such that encoding in long-term memory would occur. Conse-quently, the lower capacity learners were less able to access information fromlong-term memory at the time of the delayed posttest.

Recent research by Sagarra (2007) has also shown that the effects of feed-back may be moderated by WM capacity. In her study, 65 L1 English collegestudents in first-semester Spanish courses completed fill-in-the-blank exercisesonline that required the appropriate use of Spanish adjectives. After enteringincorrect responses, the learners either received computer-delivered oral re-casts on their errors (the feedback group) or did not receive recasts (the controlgroup). The WM capacities of learners in the feedback group (as measuredby a reading span test) predicted not only their linguistic accuracy on writ-ten posttests but also the amount of targetlike modified output they producedfollowing feedback in subsequent face-to-face (FTF) interactions.

Payne and Whitney (2002) likewise employed both computer-mediatedand FTF interaction tasks to examine L2 development in relation to the WMcapacity of 58 L1 English learners in their third semester of Spanish courses.This research compared the posttest performance of learners who had engagedin computer-mediated communication (CMC) tasks with that of learners whohad engaged in FTF interactions on the same topics over the course of asemester. Greater improvements in oral proficiency were found for learnersin the CMC group compared to the FTF group. Furthermore, the correlationbetween WM and L2 development was stronger for the FTF participants in away that suggested that CMC may have been a particularly helpful “equalizer”for learners with lower WM capacities. Payne and Whitney reasoned that thechatroom environment of CMC might have served as a support for students whohad less ability to maintain verbal information in memory; after all, the rate ofconversational exchange is slower in CMC, and given that learners can refresh



their memory traces through reading, interlocutors’ language production is lessfleeting in CMC compared to the situation of oral interactions.

Other research has suggested that learners with different WM capacitiesmay benefit differentially from various kinds of feedback. For instance, inthe context of a study of memory and implicit L2 learning, J. N. Williams(1999) suggested that learners with lower WM capacity might benefit morefrom reformulations that provide them with opportunities to make cognitivecomparisons without overly taxing their attentional resources by additionallyrequiring modifications to their output. Robinson (2002) has also claimed thatmemory for contingent speech “may be particularly relevant to distinguishingbetween learners who benefit from implicit negative feedback provided by tar-geted recasts during oral interaction [and those who do not]” (p. 118). Williamsfurther pointed out that these “individual differences in memory ability that areapparent even in the earliest stages of exposure have consequences for ultimatelevels of learning” (p. 22).

Modified Output and Working Memory Capacity

Taken together, the existing research provides both indirect and direct empiricalsupport for a relationship between WM capacity and the amount and typeof benefits learners garner from interaction by modifying their output. It isalso possible that WM capacity influences what learners do at the momentof output modification. For example, modifying one’s language in responseto an undirected clarification request such as “What?” likely invokes both theprocessing and storage functions of WM. First, a learner must shift attentionaway from meaning-based task completion and toward the particular linguisticstructures that are being used to accomplish the task. Then, if the learner decidesto respond with a linguistic modification of the original utterance, the learnermight “scan” what s/he said immediately prior to the feedback in order to either(a) compare the utterance to existing L2 structures in long-term memory andidentify the mismatch between what s/he has said and what s/he knows to becorrect or (b) identify the locus of a likely error and “take a stab” at correcting it.Because any attempt at modifying output in this manner involves some degreeof metalinguistic awareness and reprocessing, changes made to an utterancemight be related to interlanguage development in some way, regardless ofwhether the resulting modification is more, less, or equally as targetlike as theoriginal, as discussed earlier.

Acknowledging the variety of ways in which feedback may be providedand output may be modified in a meaning-focused interactional context, it



seems clear that WM capacity likely exercises influence on learners’ abilitynot only to maintain their interlocutors’ input and keep their own output activefor further processing but also to regulate attention during that processing. Inorder to examine this influence, the current project investigates the relationshipbetween WM capacity and the production of modified output in task-basedinteractions.

Research Question and Hypothesis

Theoretical and empirical SLA research suggests the following: (a) One of thebenefits of interaction, important in the interaction-learning relationship, is theproduction of modified output; (b) WM capacity is related to L2 learning; and(c) learners with different WM capacities might vary in their production ofmodified output. In the few studies that have been carried out to date, learnerswith higher WM capacities have tended to demonstrate more lasting benefitsfrom communicative instruction and feedback; however, the relationship be-tween WM and modified output has not been investigated empirically. Thepresent experiment was designed to investigate the following question: Whatis the relationship between learners’ WM capacities and their modificationsof output? Based on the existing literature, it seems that learners with higherWM capacities, who presumably are better able to simultaneously process andstore information, might also be better able to reprocess or reconstruct theirutterances with new grammatical encodings. Therefore, we hypothesized thatlearners with higher WM capacities would be likely to produce more modifiedoutput than those with lower WM capacities.

Method

OperationalizationsOpportunities for Modified OutputLearners were prompted to reflect on their ungrammatical production in away that was as naturalistic as possible during the task-based interactions.The Spanish-speaking interlocutors provided signals to the learners during theinteractions to suggest that an utterance was nontargetlike or had not been un-derstood. The feedback encompassed mainly clarification requests (e.g., askingin Spanish, “Pardon?” “What?” or saying, “I don’t understand”) and repetitions(i.e., repeating part or all of the error with rising intonation or a questioninglook), which have been treated as one category of feedback, prompts, in pre-vious literature (Lyster, 2004). Because what happens after feedback was the



focus of the current study, not the nature of the feedback, we did not distinguishamong the different types of feedback provided. The feedback was as natural-istic as possible, with the goal of providing opportunities for learners to modifytheir output. Turns where learners made partial or complete changes to theiroriginal utterances were coded as modified output. As noted earlier, modifiedoutput included utterances that were more, equally, or less targetlike than theoriginals. An example of a learner modifying output is provided in Example 2.

Example 2. Change to original utterance.

NNS: necesita ∗doble a la derecha(you) need ∗turn to the right

NS: necesita. . .? ← Feedback(you) need. . .?

NNS: necesita doblar a la derecha ← Response(you) need to turn to the right

In Example 2, the learner attempted the phrase “You need to turn to the right”and used an inflected form where an infinitive form is required. Followinga prompt from the native speaker (NS), in the form of a partial repetitionof the error with rising intonation, the learner modified her output in thedirection of the target. Learners sometimes modified their output away from thetarget or produced nontargetlike utterances that were different from the originalutterance but were equally ungrammatical. Again, more and less targetlikemodified output were both coded as modifications because accuracy was notthe issue in the current study, and, as noted earlier, a number of researchers haveargued that the process of modifying output may be as important as the product(McDonough & Mackey, 2006). Furthermore, learners’ ability to modify in thedirection of the target following feedback is likely to be more related to theirL2 competence than to any cognitive capacities underpinning the tendency tomodify output, which was the relationship of interest for the present study.

Another way that learners made changes to their original utterances was byrepeating the feedback, either partially or completely. An example of repeatingthe feedback is provided in Example 3.

Example 3. Repetition of feedback.

NNS: en ∗la dibujoin ∗the picture

NS: el dibujo ← Feedbackthe picture



NNS: el dibujo ← Responsethe picture

In Example 3, the learner repeated the corrected form, modifying his originalutterance.

Not included in the operationalization of modified output were turns wherelearners simply acknowledged the NS’s feedback, as shown in Example 4.

Example 4. Acknowledgment.

NNS: Ella tiene ∗pelaShe has ∗hair

NS: Tiene pelo? ← FeedbackShe has hair?

NNS: Sı. ← ResponseYes

In Example 4, the learner produced a nontargetlike attempt at the lexical item“hair.” The NS interlocutor used a clarification request that included the target-like lexical item. The learner’s response seems to confirm that “hair” was hisintended meaning; however, he did not explicitly modify his original output orrepeat the form, and it is not quite clear what his response indicates. This sortof “moving along” utterance was not one of the constructs of interest for thecurrent study.

We designed the study to investigate modified output as a whole constructand to examine it in the context of WM capacity. Isolating and comparing arange of different feedback types that potentially lead to a range of differentmodified output types and testing the associations between each individualcomponent of feedback-modified output and WM, for a range of controlledlinguistic forms, would take a study much larger in scale and scope than theone carried out here. Rather, our research represents a starting point for workon interaction and WM. For this reason, our NS interlocutors were trained touse feedback in the form of prompts to maximize opportunities for output, andonly feedback turns that provided opportunities for modified output to occurwere considered in the dataset.

Working Memory CapacityFollowing the conceptual framework of WM that asserts that processing andstorage draw from the same supply of resources (e.g., Miyake & Friedman,1998), WM capacity was operationalized as the ability to process verbal in-formation for form and meaning while storing lexical items. It was assessed



by means of an L1 listening span test adapted from Daneman and Carpenter’s(1980) sentence span test. WM is argued by a number of cognitive psycholo-gists to be a language-independent system, and these arguments motivated ourchoice of a WM test administered in learners’ L1 rather than the L2, a point towhich we return below.

Transcription and CodingThe treatment sessions were transcribed by two of the researchers. Next, thetranscripts were examined for instances of interactional feedback that providedopportunities for a response. When the learner had the opportunity to respond,the response was coded following the possible responses to feedback discussedearlier. It is important to point out that, like much interaction data, these data arenoisy. Feedback types, like recasts and clarification requests, often co-occuredin the same turn. Many problematic learner utterances involved multiple errors.Finally, changes to an original utterance occasionally co-occurred with repeti-tions of original utterances, with some parts of the response being more, andsome less, targetlike than the original. Each occurrence of a coding categorywas coded individually.

ParticipantsSpanish Foreign Language LearnersParticipating learners were 42 undergraduates enrolled in fourth-semesterSpanish courses at a major U.S. university. In order to enroll in this inter-mediate course, students had either successfully completed the third-semesterSpanish course at the same university and/or placed into the class on the basis ofperformance on a department-internal placement exam. Learners volunteeredfor participation in the study, which was one of several optional opportunitiesfor earning extra credit in their course. All learners were native speakers ofEnglish. Learners who had studied other Romance languages were excludedfrom the study. Learners’ ages ranged from 18 to 20, with the exception of twolearners, one of whom was 24 and the other 30. The mean age was 19.

Spanish SpeakersThe interlocutors for the interactions were four bilingual speakers of Span-ish and English (NSs) who ranged in age from 27 to 32. All four NSs werePh.D. candidates in Spanish linguistics at the same university. The NSs weretrained to provide feedback during interaction in the form of clarification re-quests and repetitions, both considered in previous research to be similar toone another in terms of beneficial roles and functions, and known as prompts



(Lyster, 2004). The use of these prompts during interaction was a means tocontrol for consistency in the amount and types of feedback that was providedacross dyads. Training consisted of receiving written information on modifiedoutput, practicing the recognition of the quantity, nature, and linguistic objec-tives of feedback, and monitoring the amount and type of feedback providedduring interaction. Interlocutors were shown video and live demonstrations ofa learner and a Spanish-speaking interlocutor completing oral communicationtasks of the same types as those used in the study. During the videotapedtasks, after each instance of feedback and modified output, the videotape wasstopped and the instance was discussed. After the video demonstration, the NSswatched another learner-interlocutor pair complete a task in real time and thendiscussed the interaction extensively. Each NS then engaged in individual prac-tice through role-playing of several interactional tasks with intermediate-levellearners of Spanish. After the role-plays, the NSs and researchers discussedstrategies for the appropriate delivery of feedback that would push for output(i.e., prompts, as explained earlier). They were trained to avoid providing cor-rect forms through either recasts or direct error correction wherever possible.The training sessions lasted 2 hours. E-mail follow-ups with questions andanswers were also conducted to be sure the NSs understood and were able toprovide feedback that led to modified output.

MaterialsInteraction Materials: Communicative TasksResponses to feedback were elicited through four interactive tasks carried out indyads, where learners were paired with a NS. Four tasks were carried out by eachdyad in order to elicit the greatest variety of forms within different contexts ofspontaneous speech. The tasks included the following: (a) a map task, adaptedfrom Terrell, Andrade, Egasse, and Munoz (1994); (b) a picture drawing task;and (c) a spot-the-difference task, both adapted from McDonough and Mackey(2000); and (d) a story completion task, adapted from Bretz, Dvorak, andKirschner (1997). For the map task, the learner and the interlocutor each hadone of two slightly different college campus maps, which they used to work outhow to get to different locations on the campus. In the picture drawing task, thelearner used a drawing of three people and described it to the interlocutor sothat she could draw the picture without looking at it. In the spot-the-differenceactivity, the learner and interlocutor each looked at a different drawing of astreet scene and they worked together to identify differences between the similarscenes. Finally, in the story completion task, the learner and interlocutor bothlooked at a picture of three people in a restaurant. The learner asked questions



about the people in the picture in order to elicit information from the NSinterlocutor and thus construct a story.

The tasks were designed to maximize opportunities for interaction as wellas to elicit a wide range of morphosyntactic forms and vocabulary. Whilecertain structures tended to be used in different tasks, questions in the spot-the-difference task, for example, the goal was for the meaning-focused interactionsto provide many opportunities for production of a wide range of forms. The taskswere not targeted a priori at specific forms to avoid the learners interpretingthe interaction as a practice activity for a specific form, which might havealtered their modified output. The approach used allowed the interlocutors togive feedback appropriate for each individual learner’s errors. In other words,prompts for modified output were automatically tailored to a learner’s individualdevelopmental level in relation to the form(s) with which they were havingproblems.

Working Memory TestThe learners’ WM capacity was assessed using a verbal WM span test basedon those employed in previous studies (Daneman & Carpenter, 1980; Turner &Engle, 1989; Waters & Caplan, 1996). The span test format was chosen becauseits underlying construct assumes that WM involves storage and processing,which research in cognitive science has shown relates to how individuals can“carry out various central executive processes such as updating information,inhibiting incorrect but natural responses, and filtering out irrelevant infor-mation” (Cowan, 2005, p. 62). The test was administered in the L1 becauseWM has been argued to be language independent (Kane et al., 2004; Osaka& Osaka, 1992; Osaka, Osaka, & Groner, 1993; Trofimovich, Ammar, &Gatbonton, 2007) and because we believe that measuring WM in the L2 couldbe obscured or filtered by nonautomatized linguistic knowledge, especiallywhen administered to learners at nonadvanced levels of L2 proficiency.

The test consisted of 36 unrelated sentences, presented to learners aurally(via cassette tape) at 3-s intervals. Half of the sentences were grammaticallycorrect and half were semantically plausible. There was an equal distribution ofthe four resulting sentence types (i.e., grammatical and plausible; grammaticaland implausible; ungrammatical and plausible; ungrammatical and implausi-ble), and the order of presentation of sentence types was randomized.

The sentence-final words, which served as recall items, were common,noncompound concrete nouns of one to three syllables in length. Abstractwords were not included as recall items because they have been found to affectthe difficulty of recall (Turner & Engle, 1989). No sentence-final word of a



given set was semantically associated with another word in that set, as judgedby three of the researchers, and no words within a set rhymed. The 36 sentenceswere grouped into nine sets for presentation to the students; these includedthree sets of 3 sentences, three sets of 4 sentences, and three sets of 5 sentences,a grouping similar to that used by Daneman and Carpenter (1980) and Turnerand Engle (1989).

The sets were presented in ascending order, which may have made it pos-sible for learners to anticipate how many recall items would be included in theupcoming set. This was not considered problematic, however, because scoringwas not absolute, but rather aggregate (see Conway et al., 2005, for furtherdiscussion and critique of WM scoring procedures). In absolute scoring suchas that used by Daneman and Carpenter (1980) and Waters and Caplan (1996),WM capacity is determined based on the maximum set size for which par-ticipants can recall all sentence-final words. If participants know the size ofthe upcoming set, they might be able to invoke some trade-off strategy thatends up affecting an absolute score. In contrast, in our study, WM capacitywas determined with aggregate scoring—that is, based on sums of all correctlyprocessed sentences and correctly recalled sentence-final words. This type ofscoring made it less important whether the learners could anticipate how manywords they would be required to recall and thus strategically allocate attentionto the recall component of the test.1 Furthermore, learners’ plausibility, gram-mar, and recall subscores were not negatively correlated, as would be expectedif they were trading off the processing and recall components of the test. Infact, there was a significant positive correlation between processing and recallsubscores, r = .33, p < .05.

Each learner took the WM test individually with one of the researchers.Before the span test, participants were presented with prerecorded instructionsthat included a practice set of three sentences. As they listened to the sets oftest items, they made judgments about the plausibility and grammaticality ofeach sentence, marking these judgments on a paper answer sheet. This wasmeant to represent the processing component of WM and was done to ensurethat the participants were processing the sentences for meaning in addition tofocusing on the retention of recall items (see Walter, 2004, for a discussion ofthis issue). At the same time, participants were required to remember the finalword of each sentence, representing the storage component of WM. After eachset of three, four, or five sentences, the auditory presentation was stopped bythe researcher, and learners repeated aloud the words they remembered fromthat set. Following scoring methods that have been used in previous research(e.g., Harrington & Sawyer, 1992; Mackey et al., 2002; Turner & Engle, 1989;



Walter, 2004; Waters & Caplan, 1996), the order in which words were repeateddid not affect scores.

Exit QuestionnaireAn exit questionnaire was used to check for the level of the learners’ involve-ment in the tasks, to uncover information about whether they had identified anyof the goals of the study, and to allow for qualitative post hoc analyses of thedata in connection with levels of attention in future research. The questionnairecontained eight questions. These are provided in the Appendix.

ProcedureLearners took the WM test, then interacted with a NS on the four tasks (orderwas randomized using the Latin square method, which ensures that each taskappears in each position—first, second, and so on—an equal number of times),and then filled out the exit questionnaire. Completion of the experimentalsession took approximately 2 hours.

Coding and ScoringWorking Memory DataRecall that the operationalization of WM in this study includes both a process-ing and a storage component. Following Walter (2004), scores took both ofthese components into account and were calculated as follows: Because twojudgments, one of plausibility and one of grammaticality, contributed to overallprocessing, one-half point was awarded for a correct judgment of plausibilityand one-half point for a correct judgment of grammaticality for each sentence.For the recall scores, one point was awarded for each correctly recalled word.Recalled words that added or deleted a plural morpheme were awarded onepoint as well. This scoring procedure was also used by Turner and Engle (1989)and by Mackey et al. (2002). There were 36 sentences on the test, so therewere 36 possible recall points and 36 possible processing points, for a total of72 possible points. Cronbach’s alpha was .8 for recall and .6 for processing.Table 1 displays descriptive statistics for the WM test.

Interaction DataAs discussed earlier, each dyad’s interaction data were coded by first markingwhere opportunities for output modification occurred and then categorizing thelearners’ verbal responses to interactional feedback as follows: (a) completeor partial repetitions of the feedback, (b) changes of original utterances eithertoward or away from the target form, and (c) simple acknowledgments of the



Table 1 Descriptive statistics for the WM span tests

Mean SD Min Max

WM Total (Recall + Processing) 58.70 5.40 45.50 67.00Recall 27.50 4.77 17.00 35.00Processing (Plaus + Gram) 31.20 1.84 27.50 34.50

(Plaus 1/2 pt each) 16.55 0.93 14.00 18.00(Gram 1/2 pt each) 14.65 1.39 11.00 17.00

feedback. Instances of (a) and (b) were scored as modified output. Interraterreliability was 90% and was measured on 15% of the interaction data codedby five researchers. The researchers discussed and recoded the 10% of the dataabout which there was initial disagreement. Because feedback by definition isprovided in response to errors, learners received different amounts of feedbackaccording to the errors they made, so proportional scores were calculated,dividing each learner’s number of actual modified output responses to feedbackby the total number of response opportunities available to that learner.

Results

This experiment was designed to investigate whether there was a relationshipbetween learners’ production of modified output and their WM capacity. It washypothesized that learners with higher WM capacities would be more likely toproduce modified output than learners with lower WM capacities.

Responses to FeedbackOverall, learners had 1,204 opportunities to modify their output (mean: 28.67episodes per learner). In response to these 1,204 opportunities, learners changedtheir original utterances just under half (47%) of the time (these changes weremore targetlike 91.94% of the time). They also completely or partially repeatedwhatever the NS had said just over a quarter (26%) of the time. Simple turnacknowledgment, which was not classified as modified output, occurred 27%of the time. Table 2 displays this information about learners’ responses tofeedback.

Modified Output ScoresProportion scores were calculated for each learner’s production of modifiedoutput. The analysis considered the total proportion of modified responses toopportunities to modify as well as the proportion of repetitions, the proportion ofchanges to original utterances, and the proportion of more targetlike changes.



Table 2 Learners’ (N = 42) responses to feedback when output modification waspossible

Mean rateResponse Mean of responsetype frequency (%) SD Min Max

Modifiedoutput (MO)

Repetitions 7 26.08 6.06 0 27Utterance changes 13 46.85 5.86 0 25

More targetlike 11.95 91.94a 5.56 0 23Less targetlike 1.05 8.06b 1.32 0 5

Total MO responses 20 72.93 6.77 10 35

Acknowledgment of turn 8.67 27.07 7.33 0 29

Total opportunities to modify 28.67 100 11.5 14 64

a,bThese percentages represent proportions of the total number of utterance changes(more + less targetlike).

Table 3 Regressions: The impact of WM capacity on the production of modified output

R β F t p

Model 1 .41 8.24 .01WM composite .41 2.87 .01

Model 2 .43 4.41 .02WM processing .25 1.72 .09

WM recall .31 2.09 .04

Note. n = 42 for all variables. R2 = .17 for Model 1; R2 = .18 for Model 2.

Working Memory and Responses to FeedbackIn order to determine whether learners’ WM test scores predicted the amount ofmodifications learners produced during interaction, a simple linear regressionanalysis was conducted. It was hypothesized that learners who had higherWM capacity would provide more modifications than learners with lower WMcapacity. As can be seen in Table 3, this hypothesis was confirmed: Linearregression analysis of the composite WM scores significantly predicted learnerproduction of modified output.

The regression analysis showed a significant, positive relationship of mod-erate strength, indicating that learners with higher WM scores produced moremodified output. The value of the standardized coefficient β is .413, mean-ing that each increase of one standard deviation for WM scores predicted



an increase of approximately one half a standard deviation for modifications.The R2 value (measuring effect size) shows that 17% of the variation in theproduction of modified output can be explained by variation in WM.

A second regression model was used to investigate the separate effects ofprocessing and recall on the production of modified output. These results arealso displayed in Table 3. Only the recall portion of the WM scores signif-icantly predicted the production of modified output. The value of β is .31,indicating that each increase of one standard deviation in recall scores pre-dicted an approximately one-third standard deviation increase in modifications(the variable effect size of .06, however, indicates that this effect is a weakone). The model R2 value of .18 indicates that this model accounts for 18%of the variation in the production of modified output. Although the relation-ship between processing scores and modified output was also positive, it wasnot significant. It should be noted that whereas the regression analysis sug-gests that WM is an important factor in the production of modified output,the relatively low effect sizes for both models (both under 20%) indicate thatvariables other than WM are also important to the production of modifiedoutput.

Discussion

This study was designed to test the hypothesis that WM capacity is related tolearners’ tendency to produce modified output, and the results provided somesupport for this claim. Learners’ WM test scores significantly predicted theirproduction of modified output. When analyzing the components of the WM testscore, the recall scores were found to significantly predict learner productionof modified output. The predictive power of learners’ processing scores alonewas not statistically significant.

The relationships found here between WM and modified output resonatewell with Skehan’s (2002) claims about the potential of WM to contributeproductively to SLA research. They also support Trofimovich et al.’s (2007)statement that the “cognitive constructs of attention, memory, and languageaptitude ‘shape’ L2 interaction on a minute-by-minute basis” (p. 192). Asdiscussed earlier, conversational interaction requires the balancing of a complexset of abilities. In comprehending and producing language, learners may alsobe maintaining input and output in short-term memory, retrieving relevantknowledge about the L2 from long-term memory, and attempting to processand compare the two. This requires selective attention as well as the executivecontrol of what is activated so that limited attentional resources will not be



overwhelmed. Accordingly, we were interested in WM as a dynamic capacitysystem encompassing both processing and storage functions.

Although these results indicate overall that WM may be an important factorin the production of modified output, it is important to note that all types ofmodified output may not be equally related to all types of WM capacity. Asnoted in several places earlier in this article, the purpose of the current studywas to investigate the possibility that higher WM capacity was associated withmore output modification. As such, modified output was defined quite broadly,as discussed in the Operationalizations section, and interlocutors were trainedto provide prompts to get learners to produce modifications. In the case of othersorts of feedback, including recasts, interaction researchers have claimed thatlearners might sometimes “parrot” or repeat the feedback they received, whichmay involve immediate recall from a short-term memory store (Gass, 2006).At other times, learners may make a cognitive comparison of their output withthe recast, notice differences, and produce novel output.

In contrast, when responding to prompts, learners are not given a modelto parrot, and they may need to recall their earlier utterance while processingpossibilities for modifications. Thus, it is possible that the WM demands thatoccur during repeating recasts may differ from the demands of making changesto utterances in response to prompts. As noted in the Operationalizations sec-tion, it was beyond the scope of the current study to focus on a complete rangeof different types of modified output. However, because of the interesting pos-sibility that different types of output may be differentially related to WM, wecarried out a post hoc analysis to compare the role of WM in the productionof different types of modified output. A General Linear Model (GLM), usingoutput in the form of some sort of repetition and output in the form of some sortof change to the original production as response variables with WM processingscores and recall scores as covariates was used. The results of the analysis aresummarized in Table 4.

Table 4 WM capacity and different responses to feedback

Repetitions Changes to original utterances

Source df MSQ F Partial η2 df MSQ F Partial η2

WM processing 1 .02 .45 .01 1 .16 4.45∗ .11WM recall 1 .28 6.67∗∗ .15 1 .04 1.29 .03

∗p < .05. ∗∗p < .01.



Results of the GLM analysis show a significant effect of WM processingscores on the tendency to change original utterances following feedback but noton the tendency to repeat a recast. Additionally, there was a significant effectof WM recall scores but with the opposite pattern (on the tendency to repeatrecasts but not on the tendency to modify original utterances). This indicates asplit effect, where different aspects of WM measured by the test were relatedto different types of modified output, lending credence to the possibility thatthe effect of WM on modified output may be different for different types ofmodifications and suggesting that it would be useful for future researchers todesign a study to investigate this.

The results of this study also suggest it might be useful in the future tolook into whether separate “short-term memory” and “grammatical sensitivity”measures may have been built into the sentence span test we employed. Phono-logical short-term memory (PSTM), typically operationalized as the ability torepeat novel phonological input (e.g., pseudowords) correctly, is a subcompo-nent of Baddeley’s (2000, 2003a) WM system and represents a hearer’s abilityto retain spoken sequences temporarily. Several studies focusing specificallyon this aspect of the WM system have discovered relationships between PSTMand L2 learning, finding PSTM to be related not only to vocabulary acquisition(e.g., Daneman & Case, 1981; Papagno, Valentine, & Baddeley, 1991; Papagno& Vallar, 1992; Service, 1992; Service & Craik, 1993; Service & Kohonen,1995) but also to grammar learning (e.g., N. C. Ellis & Schmidt, 1997; N. C.Ellis & Sinclair, 1996; J. N. Williams, 1999; J. N. Williams & Lovatt, 2003).The recall component of the WM test used in the current study was not de-signed or intended to be used as a stand-alone measure of PSTM; in orderto remember the sentence-final words on our WM tests, participants had toallocate attentional resources to that part of a more complex task, which alsoinvolved making online plausibility and grammar judgments. Nonetheless, ifthe recall component of our WM tests served as an approximate measure ofthe phonological loop function plus the involvement of the central executive inallocating resources to it, it makes sense that this might be related to a tendencyto repeat feedback in the context of interaction.

In contrast, making changes to original utterances in response to feedbacklikely draws on both storage and processing abilities. In making output modi-fications, learners might be attempting to maintain what they have already saidin an active state while retrieving L2 structures from long-term memory so thattheir original utterances can be compared against L2 knowledge. As describedearlier, only the processing component of the test (not the recall component)influenced the tendency to modify output by changing an original utterance



(but not the tendency to modify by repeating modeled speech). It may be im-portant here to note that the recall ability measured by our WM test involvedremembering words from sentences that were not produced by the participantsthemselves. Thus, there may have been nontrivial differences between the typesof storage measured by our WM test and those involved in modifying output.

Perhaps more importantly, however, results of bivariate correlations indi-cated that the aspect of processing that was found to be significantly relatedto learners’ output modifications was the learners’ grammar judgments (r =.32, p < .05), not their judgments of plausibility (r = .13, n.s). As mentionedearlier, Skehan (2002) has highlighted the importance of the ability to attachsignificance to certain aspects of language in relation to others when regulatingthe allocation of focal attention. Trofimovich et al. (2007) have argued that ifanalytical ability has a role to play in L2 interaction, it is likely in influencing alearner’s ability to focus on the structural properties of language—somethingthat they claim may mediate the benefits of feedback. In the context of the cur-rent study, it makes sense that learners who were skilled at allocating attentionto processing and making correct grammar judgments on the WM tests wouldalso be able to focus attention on grammatical form in interaction and modifytheir output accordingly.

Additionally, there may be nontrivial similarities between the processingcomponent of the test and the language processing involved in the productionof changes to utterances, particularly as the WM test employed in this studyincluded processing for grammaticality. Indeed, N. C. Ellis (2005) has drawnattention to a danger of circularity in the operationalization of WM in SLAresearch, pointing out that because sentence span tests of WM involve muchlanguage processing in the first place, learners’ demonstrated ability on suchtasks will naturally predict their language processing ability on other tasks. Forresearch intended to focus specifically on the influence of memory abilities inlanguage (and language acquisition) processes, such concerns would seem toargue for less language-dependent measures of WM.

On the other hand, Baddeley (2003b) and Turner and Engle (1989) haveargued that what processing the test-takers do should not matter. For example,Daneman and Carpenter (1980) found a high correlation between WM span andreading comprehension, and this finding held true in later studies even whenthe sentence processing was replaced by other tasks, such as calculating mathproblems (Turner & Engle, 1989) or making color-word associations (Bayliss,Jarrold, Gunn, & Baddeley, 2003). According to Conway et al. (2005), to drawout individual differences on the recall or storage portion of a WM test, a highlydemanding processing task is needed, but when processing accuracy is not close



to ceiling, there may be a processing/storage trade-off for certain individuals,which goes against evidence from past studies showing that processing scoresusually correlate positively with storage performance scores (Kane et al., 2004;Waters & Caplan, 1996).

As stated earlier, in this study relationships were found between learners’prompted changes to original utterances and the processing components of theWM test, but not between such modifications and the recall component of theWM measure. What this tells us is that the particular WM test used in this studymay have a processing component that matters in that it revealed relationshipsthat would not have been observed had the subcomponents of the WM task notbeen taken into account in scoring and data analyses. Thus, such WM tests withcomplex processing demands may be tapping into more than just WM capacity,and it may be plausible that they are tapping into grammatical sensitivity, oneof the constructs of L2 aptitude.

It may also be worthwhile to rethink the scoring procedures that have beenused for sentence span WM tests in SLA research. Some researchers havejustified the inclusion of plausibility and/or grammar judgments primarily asa way of ensuring that their WM tests involve both processing and storage.Such research is less interested in how accurate the participants’ plausibilityand grammar judgments are and more interested in preventing participantsfrom rehearsing the sentence-final words, thereby decreasing the extent towhich a WM test actually measures simultaneous processing and storage. Thus,processing information is often used for scoring purposes mainly as prerequisitefor determining whether the recall for a given sentence should be counted, anda final sentence span score represents the sum of accurate recalls in caseswhere plausibility and grammar judgments were also correct. However, thecurrent study’s finding that the accuracy of processing judgments predictedthe tendency to make prompted changes to utterances in the L2, whereasrecall scores alone did not, suggests that researchers may be missing importantinformation. Furthermore, as Conway and his colleagues (2005) have argued,there are no trade-offs between processing and storage on WM span testsdesigned to avoid this problem; thus, researchers must take into account theirparticipants’ processing scores in relation to the storage scores to present anaccurate picture of the data.

As we discuss later in the section on limitations, these points indicate thatit might be worthwhile to take a closer look at various aspects of existing WMtests and analyze what they are truly measuring in the context of L2 research.In addition to developing new measures, SLA researchers who are currentlymaking use of sentence span tests might want to look at relationships not only



between their dependent variables and composite WM scores but also withscores on subcomponents of the WM tests. It may also be fruitful to employadditional tests, such as measures of PSTM, attentional control, grammaticalsensitivity, and, certainly, WM measures that are less language dependent, suchas digit span tests. After all, as N. C. Ellis (2005) has suggested, differentcomponents of working memory may be differentially involved in differentkinds of learning; PSTM may contribute to the ability to retain feedback (e.g.,recasts) for further processing, whereas executive functioning (or the “supervi-sory attentional system”) may be “more associated with explicit learning andthe analysis of the language that is temporarily represented in the phonologicalloop or episodic buffer as well as in consciously created construction” and theproduction of output (p. 339).

In any case, the relationships found in the current study between learners’sentence span processing scores and their tendency to modify their utterancesdo seem at least to lend credence to the idea that modifications of output involvethe allocation of attention to meaning and grammatical structures. If this formspart of how modified output facilitates SLA, then learners with skills in this areamay be at an advantage in terms of learning from interaction. As McDonough(2005) and Loewen (2003) have shown, the production of modified output isone of the strongest predictors of learning from feedback. The results of thecurrent study may also help to explain the results of some of the studies reviewedearlier, in which learners with higher WM capacities showed greater lastingbenefits from communicative instruction (Ando et al., 1992) and interactionwith feedback especially in relation to the noticing of feedback (Mackey et al.,2002) and in terms of targetlike modified output production (Sagarra, 2007).Studies that have investigated the role of interaction in SLA have frequentlyfound a range of variation in the impact of interaction on L2 development (e.g.,Mackey, 1999; J. Williams & Evans, 1998). Differences in WM capacity andthe relationship between WM capacity and modified output may explain in partwhy L2 interaction is more beneficial for certain learners than others.

Limitations

The results of this research indicate that WM capacity is likely one of thefactors that influence the production of modified output. As noted earlier,the strength of the relationships found between WM capacity and learners’response tendencies was relatively low, reflecting the likelihood that WM is notthe sole predictor. Rather, other learner-internal capacities, perhaps includingtheir stages of developmental readiness (and/or L2 proficiency), awareness



of form, grammatical sensitivity (see Ranta, 2002), and other aptitudes (seeRobinson, 2005) may additionally influence their responses to interactionalfeedback on different linguistic forms. In the current study, all modified output,regardless of whether it was more targetlike or not, was considered as modifiedoutput. This consideration increased the likelihood that learners’ developmentallevel did not impact on the results, as it would were only more targetlikemodifications considered. However, although the learners were at similar stagesof L2 proficiency, there is still a chance that developmental level played a rolein the results (i.e., that learners at higher proficiency levels may be more likelyto modify their output regardless of WM capacity; cf. Iwashita, 2001). It isalso possible that external factors not considered here, such as the type ofinstruction the learners normally received in their classes, may have influencedtheir response tendencies.

In this study, modified output was elicited primarily by prompts. Althoughthe NSs were trained to use prompts, they also provided some recasts, asthe interactions were kept as naturalistic as possible. When recasts resultedin opportunities for modified output, they were also considered part of thisdataset. Closer attention to the relationship among type of feedback, type ofmodified output, and type of WM is clearly an important next step for futureresearch, now that our relatively coarse-grained study has pointed to an initialassociation.

Future research will benefit from more precise measures of WM to furtherrefine our understanding of the relationship between WM capacity and languagelearning processes. WM tests should be motivated by and operationalized ac-cording to specific models of WM and must also be tested for their reliabilityand validity. Furthermore, despite the widespread use of WM span tests suchas the ones used in this study, there is still debate in the field concerning whatWM tests actually measure. For example, Friedman and Miyake (2004) arguedthat further research into theories and models of WM should be carried out tobetter explain what aspects of WM are measured by span tests. It seems likelythat some aspect of grammatical sensitivity might be captured by tests of WMcapacity. Additionally, it has been shown in previous research that grammaticalsensitivity is a strong predictor of L2 learning. It is one of the five subcom-ponents of the Modern Language Aptitude Test (Carroll, 1990; Skehan, 2002;see Dornyei, 2005, for a review) and is often used in SLA research as a testof language analytic ability (DeKeyser, 2000; Skehan, 2002; Walter, 2004).SLA researchers using WM tests must be informed by research in mainstreampsychology if they are to adequately characterize the relationship between WMand SLA (Juffs, 2003). Some of these suggestions include using a battery of



span measures with different processing components (e.g., using operation spanand counting span in addition to reading or listening span) and to operational-ize the common variance among the measures as WM, which may eradicatetask-specific factors and increase the construct validity of what is being mea-sured (Conway et al., 2005). After all, no single test can perfectly capture theunderlying construct meant to be measured. These limitations apply equallyto the current study, especially given the more fine-grained relationships thatwere found between subcomponents of our WM test and the learners’ responsetendencies.

Finally, it should be remembered that no measure of development wasincluded in this research. Rather, our study was designed to examine the rela-tionship between an internal learner capacity and language production. Otherempirical research has linked the production of modified output to L2 develop-ment (e.g., McDonough, 2005). It is possible that processing capacities such asWM factor into language development by influencing learning through promot-ing interactional processes likes modified output production. However, furtherresearch considering both relationships (i.e., that between WM and modifiedoutput and that between WM and L2 development) is needed to determine moreprecisely how WM influences SLA.

Conclusion

Despite the limitations of this study, the present research has provided in-teresting data pointing to a positive relationship between L2 learners’ WMcapacity and the modified output they produced during task-based conversa-tional interactions with NSs. For the 42 college-level learners of intermediateSpanish as a foreign language in this study, the general trend was for learnerswith higher WM capacity to produce more output in response to feedback.We suggested that this moderate relationship reflects the likelihood that WMcapacity, along with other learner-internal capacities, influences responses tointeractional feedback on different L2 forms.

Revised version accepted 8 April 2009

Note

1 This scoring procedure also made it possible to design a test that is shorter thanthose administered in previous studies. Earlier WM tests (e.g., Daneman &Carpenter, 1980; Waters & Caplan, 1996) included 80 or more test sentencesgrouped in sets of between 2 and 6 sentences. If an absolute scoring procedure isused, then such a range of set sizes would be necessary. Given our aggregate scoring



procedure, however, such a range was not considered essential, and in the interest ofavoiding test fatigue, we developed a shorter version of the test that nonethelessmaintained the basic principle of assessing participants’ capacity for simultaneousprocessing and storage of information.

References

Ammar, A. (2008). Prompts and recasts: Differential effects on second languagemorphosyntax. Language Teaching Research, 12, 183–210.

Ammar, A., & Spada, N. (2006). One size fits all? Recasts, prompts, and L2 learning.Studies in Second Language Acquisition, 28(4), 543–574.

Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.Ando, J., Fukunaga, N., Kurahachi, J., Suto, T., Nakano, T., & Kage, M. (1992). A

comparative study on the two EFL teaching methods: The communicative and thegrammatical approach. Japanese Journal of Educational Psychology, 40,247–256.

Baddeley, A. D. (2000). The episodic buffer: A new component of working memory?Trends in Cognitive Sciences, 4, 417–423.

Baddeley, A. D. (2003a). Working memory and language: An overview. Journal ofCommunication Disorders, 36, 189–208.

Baddeley, A. D. (2003b). Working memory: Looking back and looking forward.Nature Reviews. Neuroscience, 4, 829–839.

Baddeley, A. D. (2007). Working memory, thought, and action. Oxford: OxfordUniversity Press.

Bayliss, D. M., Jarrold, C., Gunn, D. M., & Baddeley, A. D. (2003). The complexitiesof complex span: Explaining individual differences in working memory in childrenand adults. Journal of Experimental Psychology, General, 132(1), 71–92.

Braidi, S. M. (2002). Reexamining the role of recasts in native-speaker/nonnative-speaker interactions. Language Learning, 52(1), 1–42.

Bretz, M. L., Dvorak, T., & Kirschner, C. (1997). Pasajes: Lengua (4th ed.). NewYork: McGraw Hill.

Carroll, J. B. (1990). Cognitive abilities in foreign language aptitude: Then and now. InT. S. Parry & C. W. Stansfield (Eds.), Language aptitude reconsidered (pp. 11–29).Englewood Cliffs, NJ: Prentice Hall Regents.

Conway, A. R. A., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O., &Engle, R. W. (2005). Working memory span tasks: A methodological review anduser’s guide. Psychonomic Bulletin & Review, 12(5), 769–786.

Cowan, N. (2005). Working memory capacity. New York: Psychology Press.Daneman, M. (1991). Working memory as a predictor of verbal fluency. Journal of

Psycholinguistic Research, 20(6), 445–464.Daneman, M., & Carpenter, P. (1980). Individual differences in working memory and

reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466.



Daneman, M., & Case, R. (1981). Syntactic form, semantic complexity, and short-termmemory: Influence on children’s acquisition of new linguistic structures.Developmental Psychology, 17, 367–378.

Daneman, M., & Green, I. (1986). Individual differences in comprehending andproducing words in context. Journal of Memory and Language, 25, 1–18.

de Bot, K. (1996). The psycholinguistics of the output hypothesis. Language Learning,46(3), 529–555.

DeKeyser, R. M. (2000). The robustness of critical period effects in second languageacquisition. Studies in Second Language Acquisition, 22(4), 499–533.

de la Fuente, M. J. (2002). Negotiation and oral acquisition of L2 vocabulary: Theroles of input and output in the receptive and productive acquisition of words.Studies in Second Language Acquisition, 24(1), 81–112.

Dornyei, Z. (2005). The psychology of the language learner. Mahwah, NJ: Erlbaum.Ellis, N. C. (1996). Sequencing in SLA: Phonological memory, chunking, and points

of order. Studies in Second Language Acquisition, 18(1), 91–126.Ellis, N. C. (2001). Memory for language. In P. Robinson (Ed.), Cognition and second

language instruction (pp. 33–68). Cambridge: Cambridge University Press.Ellis, N. C. (2005). At the interface: Dynamic interactions of explicit and implicit

language knowledge. Studies in Second Language Acquisition, 27(2), 305–352.Ellis, N. C., & Schmidt, R. (1997). Morphology and longer distance dependencies:

Laboratory research illuminating the A in SLA. Studies in Second LanguageAcquisition, 19(2), 145–171.

Ellis, N. C., & Sinclair, S. (1996). Working memory in the acquisition of syntax:Putting language in good order. Quarterly Journal of Experimental Psychology,49A(1), 234–250.

Ellis, R., & He, X. (1999). The roles of modified input and output in the incidentalacquisition of word meanings. Studies in Second Language Acquisition, 21(2),285–301.

Ellis, R., Loewen, S., & Erlam, R. (2006). Implicit and explicit corrective feedbackand the acquisition of L2 grammar. Studies in Second Language Acquisition, 28(2),339–268.

Ellis, R., Tanaka, Y., & Yamazaki, A. (1994). Classroom interaction, comprehension,and the acquisition of L2 word meanings. Language Learning, 44(3), 449–491.

Erlam, R. (2005). Language aptitude and its relationship to instructional effectivenessin second language acquisition. Language Teaching Research, 9(2), 147–171.

Friedman, N. P., & Miyake, A. (2004). The reading span test and its predictive powerfor reading comprehension ability. Journal of Memory and Language, 51(1),136–158.

Gass, S. M. (1997). Input, interaction and the second language learner. Mahwah, NJ:Erlbaum.

Gass, S. M. (2003). Input and interaction. In C. Doughty & M. H. Long (Eds.),Handbook of second language acquisition (pp. 224–255). Malden, MA: Blackwell.



Gass, S. M. (2006). Input, interaction and output: An overview. AILA Review (pp.3–17). Amsterdam: Benjamins.

Gass, S. M., & Mackey, A. (2007). Input, interaction and output in SLA. In J. Williams& B. VanPatten (Eds.), Theories in SLA (pp. 175–199). Mahwah, NJ: Erlbaum.

Gass, S. M., & Varonis, E. M. (1994). Input, interaction, and second languageproduction. Studies in Second Language Acquisition, 16(3), 283–302.

Geva, E., & Ryan, E. B. (1993). Linguistic and cognitive correlates of academic skillsin first and second languages. Language Learning, 43, 5–42.

Han, Z. (2002). A study of the impact of recasts on tense consistency in L2 output.TESOL Quarterly, 36(4), 543–572.

Harrington, M., & Sawyer, M. (1992). L2 working memory capacity and L2 readingskill. Studies in Second Language Acquisition, 14(1), 25–38.

Iwashita, N. (2001). The effect of learner proficiency on interactional moves andmodified output in nonnative–nonnative interaction in Japanese as a foreignlanguage. System, 29, 267–287.

Iwashita, N. (2003). Negative feedback and positive evidence in task-based interaction:Differential effects on L2 development. Studies in Second Language Acquisition,25(1), 1–36.

Izumi, S. (2002). Output, input enhancement, and the noticing hypothesis: Anexperimental study on ESL relativization. Studies in Second Language Acquisition,24(4), 541–577.

Juffs, A. (2003). Working memory: Understanding current issues in mainstreampsychology. Paper presented at the Second Language Research Forum, Tucson, AZ.

Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle,R. E. (2004). The generality of working memory capacity: A latent-variableapproach to verbal and visuospatial memory span and reasoning. Journal ofExperimental Psychology: General, 133, 189–217.

Leeman, J. (2003). Recasts and second language development. Studies in SecondLanguage Acquisition, 25(1), 37–63.

Loewen, S. (2003). The effectiveness of incidental focus on form in meaning-focusedESL lessons. New Zealand Studies in Applied Linguistics, 9(1), 63–83.

Logan, G. D. (1992). Shapes of reaction-time distributions and shapes of learningcurves: A test of the instance theory of automaticity. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 18, 883–914.

Long, M. (1996). The role of the linguistic environment in second languageacquisition. In W. C. Ritchie & T. K. Bhatia (Eds.), Handbook of second languageacquisition: Vol. 2. Second language acquisition (pp. 413–468). New York:Academic Press.

Long, M. (2007). Problems in SLA. Mahwah, NJ: Erlbaum.Lyster, R. (2004). Differential effects of prompts and recasts in form-focused

instruction. Studies in Second Language Acquisition, 26(3), 399–432.



Mackey, A. (1999). Input, interaction and second language development: An empiricalstudy of question formation in ESL. Studies in Second Language Acquisition, 21(4),557–587.

Mackey, A. (2002). Beyond production: Learners’ perceptions about interactionalprocesses. International Journal of Educational Research, 37, 379–394.

Mackey, A. (2007). Interaction. In R. DeKeyser (Ed.), Practice in second languagelearning: Perspectives from linguistics and psychology (pp. 85–110). Cambridge:Cambridge University Press.

Mackey, A., Gass, S., & McDonough, K. (2000). How do learners perceive implicitnegative feedback? Studies in Second Language Acquisition, 22(4), 471–497.

Mackey, A., & Goo, J. M. (2007). Interaction research in SLA: A meta-analysis andresearch synthesis. In A. Mackey (Ed.), Input, interaction and corrective feedbackin L2 learning (pp. 379–452). New York: Oxford University Press.

Mackey, A., Oliver, R., & Leeman, J. (2003). Interactional input and the incorporationof feedback: An exploration of NS-NNS and NNS-NNS adult and child dyads.Language Learning, 53(1), 35–66.

Mackey, A., & Philp, J. (1998). Conversational interaction and second languagedevelopment: Recasts, responses, and red herrings? Modern Language Journal,82(3), 338–356.

Mackey, A., Philp, J., Egi, T., Fujii, A., & Tatsumi, T. (2002). Individual differences inworking memory, noticing of interactional feedback and L2 development. In P.Robinson (Ed.), Individual differences and instructed language learning (pp.181–209). New York: Benjamins.

McDonough, K. (2005). Identifying the impact of negative feedback and learners’responses on ESL question development. Studies in Second Language Acquisition,27(1), 79–103.

McDonough, K., & Mackey, A. (2000). Form and meaning: Designing communicativetasks to target grammar in Thai classrooms. Foreign Language Annals, 33(1),82–92.

McDonough, K., & Mackey, A. (2006). Responses to recasts: Repetitions, primedproduction and linguistic development. The Modern Language Journal, 56(4),693–720.

Miyake, A., & Friedman, N. P. (1998). Individual differences in second languageproficiency: Working memory as language aptitude. In A. F. Healy & L. E. Bourne(Eds.), Foreign language learning: Psycholinguistic studies on training andretention (pp. 339–364). Mahwah, NJ: Erlbaum.

Nobuyoshi, J., & Ellis, R. (1993). Focused communication tasks and second languageacquisition. ELT Journal, 47, 203–210.

Oliver, R. (1998). Negotiation of meaning in child interactions. Modern LanguageJournal, 82(3), 372–386.

Oliver, R. (2000). Age differences in negotiation and feedback in classroom andpairwork. Language Learning, 50(1), 119–151.



Oliver, R. (2002). The patterns of negotiation for meaning in child interactions.Modern Language Journal, 86(1), 97–111.

Osaka, M., & Osaka, N. (1992). Language-independent working memory as measuredby Japanese and English reading span tests. Bulletin of the Psychonomic Society,30(4), 287–289.

Osaka, M., Osaka, N., & Groner, R. (1993). Language-independent working memory:Evidence from German and French reading span tests. Bulletin of the PsychonomicSociety, 31(2), 117–118.

Palladino, P., & Cornoldi, C. (2004). Working memory performance of Italian studentswith foreign language learning difficulties. Learning and Individual Differences, 14,137–151.

Papagno, C., Valentine, T., & Baddeley, A. (1991). Phonological short-term memoryand foreign-language vocabulary learning. Journal of Memory and Language, 30,331–347.

Papagno, C. & Vallar, G. (1992). Phonological short-term memory and the learning ofnovel words: The effect of phonological similarity and item length. QuarterlyJournal of Experimental Psychology, 44A, 47–67.

Payne, J. S., & Whitney, P. J. (2002). Developing L2 oral proficiency throughsynchronous CMC: Output, working memory, and interlanguage development.CALICO Journal, 20(1), 7–32.

Philp, J. (2003). Constraints on “noticing the gap”: Nonnative speakers’ noticing ofrecasts in NS-NNS interaction. Studies in Second Language Acquisition, 25(1),99–126.

Pica, T. (1994). Research on negotiation: What does it reveal about second-languagelearning conditions, processes, and outcomes? Language Learning, 44(3), 493–527.

Ranta, L. (2002). Learning conditions, aptitude complexes and SLA: A framework forresearch and pedagogy. In P. Robinson (Ed.), Individual differences and instructedlanguage learning (pp. 113–135). Philadelphia: Benjamins.

Robinson, P. (1995). Attention, memory, and the “noticing” hypothesis. LanguageLearning, 45(2), 283–331.

Robinson, P. (2002). Effects of individual differences in intelligence, aptitude andworking memory on incidental SLA. In P. Robinson (Ed.), Individual differencesand instructed language learning (pp. 211–251). Philadelphia: Benjamins.

Robinson, P. (2005). Aptitude and second language acquisition. Annual Review ofApplied Linguistics, 25, 45–73.

Sagarra, N. (2007). From CALL to face-to-face interaction: The effect ofcomputer-delivered recasts and working memory on L2 development. In A. Mackey(Ed.), Conversational interaction in second language acquisition (pp. 229–248).New York: Oxford University Press.

Sawyer, M., & Ranta, L. (2001). Aptitude, individual differences, and instructionaldesign. In P. Robinson (Ed.), Cognition and second language instruction (pp.319–353). New York: Cambridge University Press.



Schmidt, R., & Frota, S. (1986). Developing basic conversational ability in a secondlanguage: A case study of an adult learner of Portuguese. In R. R. Day (Ed.),Talking to learn: Conversation in second language acquisition (pp. 237–326).Rowley, MA: Newbury House.

Segalowitz, N., & Lightbown, P. M. (1999). Psycholinguistic approaches to SLA.Annual Review of Applied Linguistics, 19, 43–63.

Service, E. (1992). Phonology, working memory, and foreign-language learning.Quarterly Journal of Experimental Psychology, 45A, 21–50.

Service, E., & Craik, F. I. M. (1993). Differences between young and older adults inlearning a foreign vocabulary. Journal of Memory and Language, 33, 59–74.

Service, E., & Kohonen, V. (1995). Is the relation between phonological memory andforeign language learning accounted for by vocabulary acquisition? AppliedPsycholinguistics, 16(2), 155–172.

Shehadeh, A. (2002). Comprehensible output, from occurrence to acquisition: Anagenda for acquisitional research. Language Learning, 52(3), 597–647.

Skehan, P. (1998). A cognitive approach to language learning. Oxford: OxfordUniversity Press.

Skehan, P. (2002). Theorising and updating aptitude. In P. Robinson (Ed.), Individualdifferences and instructed language learning (pp. 69–93). Philadelphia: Benjamins.

Storch, N. (2002). Patterns of interaction in ESL pair work. Language Learning, 52(1),119–158.

Swain, M. (1985). Communicative competence: Some roles of comprehensible inputand comprehensible output in its development. In S. Gass & C. Madden (Eds.),Input in second language acquisition (pp. 235–253). Rowley, MA: Newbury House.

Swain, M. (1993). The Output Hypothesis. Canadian Modern Language Review, 50,158–164.

Swain, M. (1995). Three functions of output in second language learning. In G. Cook& B. Seidlhofer (Eds.), Principles and practice in applied linguistics: Studies inhonour of H. G. Widdowson (pp. 125–144). Oxford: Oxford University Press.

Swain, M. (1998). Focus on form through conscious reflection. In C. Doughty & J.Williams (Eds.), Focus on form in classroom second language acquisition (pp.64–81). Cambridge: Cambridge University Press.

Swain, M. (2005). The output hypothesis: Theory and research. In E. Hinkel (Ed.),Handbook of research in second language teaching and learning (pp. 471–484).Mahwah, NJ: Erlbaum.

Terrell, T. D., Andrade, M., Egasse, J., & Munoz, E. M. (1994). Dos mundos: Acommunicative approach (3rd ed.). New York: McGraw Hill.

Tokowicz, N., Michael, E. B., & Kroll, J. F. (2004). The roles of study-abroadexperience and working-memory capacity in the types of errors made duringtranslation. Bilingualism: Language and Cognition, 7, 255–272.

Trofimovich, P., Ammar, A., & Gatbonton, E. (2007). How effective are recasts? Therole of attention, memory, and analytical ability. In A. Mackey (Ed.),



Conversational interaction in second language acquisition: A series of empiricalstudies (pp. 171–195). Oxford: Oxford University Press.

Turner, M. L., & Engle, R. W. (1989). Is working capacity task dependent? Journal ofMemory and Language, 28, 127–154.

Walter, C. (2004). Transfer of reading comprehension skills to L2 is linked to mentalrepresentations of text and to L2 working memory. Applied Linguistics, 25(3),315–339.

Waters, G. S., & Caplan, D. (1996). The measurement of verbal working memorycapacity and its relation to reading comprehension. Quarterly Journal ofExperimental Psychology, 49A, 51–79.

Williams, J., & Evans, J. (1998). What kind of focus and on which forms? In C.Doughty & J. Williams (Eds.), Focus on form in classroom second languageacquisition (pp. 139–155). Cambridge: Cambridge University Press.

Williams, J. N. (1999). Memory, attention, and inductive learning. Studies in SecondLanguage Acquisition, 21(1), 1–48.

Williams, J. N., & Lovatt, P. (2003). Phonological memory and rule learning.Language Learning, 53(1), 67–121.

Appendix

Exit Questionnaire

1) When you were doing the Spanish tasks just now, was there anything thathelped you to complete the tasks?

2) When you were doing the Spanish tasks just now, was there anything thatdistracted you from communicating?

3) When you practice speaking Spanish, do you like people to correct you?Yes/No

4) When people correct you, do you think you learn from the correction? Ifyes, how?

5) Can you think of an example of you learning through correction?6) When you’re speaking Spanish and people correct you, how do you usually

respond?7) What do you think the purpose of this research is?8) When you were doing the Spanish tasks just now, did you notice anything

specific about how the other speaker communicated?


Exploring the Relationship Between Modified Output and Working Memory Capacity

Documents