effects of proficiency differences and patterns of pair interaction on

23
Effects of proficiency differences and patterns of pair interaction on second language learning: collaborative dialogue between adult ESL learners Yuko Watanabe and Merrill Swain Ontario Institute for Studies in Education of the University of Toronto, Canada This study investigated the effects of second language (L2) proficiency differences in pairs and patterns of interaction on L2 learning, making use of both qualitative and quantitative data. We designed the study in such a way that four different core participants interacted with higher and lower proficiency non-core participants. These learners engaged in a three-stage task involving pair writing, pair comparison (between their original text and a reformulated version of it) and individual writing. The core participants also engaged in a stimulated recall after the task. We analysed each pair’s collaborative dialogue in terms of language-related episodes and patterns of pair interaction (Storch, 2002a) as well as each learner’s individual post-test score. The findings suggested that the patterns of pair interaction greatly influenced the frequency of LREs and post-test performance. When the learners engaged in collaborative patterns of interaction, they were more likely to achieve higher post- test scores regardless of their partner’s proficiency level. It seems that proficiency differences do not necessarily affect the nature of peer assistance and L2 learning. I Introduction In this article, we examine collaborative dialogue between adult ESL learners of different L2 proficiency 1 levels. Collaborative dialogue is ‘dialogue in which speakers are engaged in problem-solving and knowledge-building’ (Swain, 2000). This concept was extended from the output hypothesis (Swain, 1985; 1995) which was based within an information-processing framework of learning. However, Vygotsky’s (1978; 1986) sociocultural theory of mind prompted us to move beyond output (i.e. speaking and writing) as a mere message to be conveyed, to collaborative dialogue – a tool of cognitive activ- ity that mediates L2 learning (Swain, 2000). A number of recent studies have shown peer–peer collaborative dialogue to be an important aspect of L2 learning (see Swain et al., 2002 for a review). Yet, several studies have found that the patterns of interaction vary across Language Teaching Research 11,2 (2007); pp. 121–142 © 2007 SAGE Publications 10.1177/136216880607074599 Address for correspondence:Yuko Watanabe, Modern Language Centre, OISE/UT, 252 Bloor Street West, Toronto, ON M5S IV6, Canada; e-mail: [email protected]

Upload: others

Post on 19-Mar-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Language Teaching Research 11,2 (2007); pp. 121–142

Effects of proficiency differences andpatterns of pair interaction on secondlanguage learning: collaborative dialoguebetween adult ESL learnersYuko Watanabe and Merrill Swain Ontario Institute forStudies in Education of the University of Toronto, Canada

This study investigated the effects of second language (L2) proficiency differencesin pairs and patterns of interaction on L2 learning, making use of both qualitative andquantitative data. We designed the study in such a way that four different coreparticipants interacted with higher and lower proficiency non-core participants.These learners engaged in a three-stage task involving pair writing, pair comparison(between their original text and a reformulated version of it) and individual writing.The core participants also engaged in a stimulated recall after the task. We analysedeach pair’s collaborative dialogue in terms of language-related episodes and patternsof pair interaction (Storch, 2002a) as well as each learner’s individual post-test score.The findings suggested that the patterns of pair interaction greatly influenced thefrequency of LREs and post-test performance. When the learners engaged incollaborative patterns of interaction, they were more likely to achieve higher post-test scores regardless of their partner’s proficiency level. It seems that proficiencydifferences do not necessarily affect the nature of peer assistance and L2 learning.

I Introduction

In this article, we examine collaborative dialogue between adult ESL learnersof different L2 proficiency1 levels. Collaborative dialogue is ‘dialogue inwhich speakers are engaged in problem-solving and knowledge-building’(Swain, 2000). This concept was extended from the output hypothesis (Swain,1985; 1995) which was based within an information-processing framework oflearning. However, Vygotsky’s (1978; 1986) sociocultural theory of mindprompted us to move beyond output (i.e. speaking and writing) as a meremessage to be conveyed, to collaborative dialogue – a tool of cognitive activ-ity that mediates L2 learning (Swain, 2000).

A number of recent studies have shown peer–peer collaborative dialogue tobe an important aspect of L2 learning (see Swain et al., 2002 for a review).Yet, several studies have found that the patterns of interaction vary across

Address for correspondence: Yuko Watanabe, Modern Language Centre, OISE/UT, 252 Bloor StreetWest, Toronto, ON M5S IV6, Canada; e-mail: [email protected]

© 2007 SAGE Publications

10.1177/136216880607074599

peer groups, and more importantly, certain patterns of interaction are claimedto be more conducive to L2 learning than those of others (e.g. Guerrero andVillamil, 1994; Kowal and Swain, 1994; 1997; Lockhart and Ng, 1995;Nelson and Murphy, 1993; Storch, 2001a; 2001b; 2002a; 2002b).

Proficiency differences have been debated as one of the influential factorsin the nature of peer–peer interaction. Although peer–peer interaction involv-ing different proficiency learners is commonly observed in a L2 classroom,surprisingly little research has documented how learners with different profi-ciency levels interact with each other, and whether such a grouping is usefulfor L2 learning (but see Kowal and Swain, 1994; 1997; Leeser, 2004; Yule andMacdonald, 1990). If we accept van Lier’s (1996) claim that students learnthrough teaching other peers, more proficient students are likely to benefitfrom working with less proficient peers. But is this really the case? A furtherquestion concerns whether less proficient learners truly benefit from receiv-ing more proficient learners’ help as they may not be developmentally readyto discuss some linguistic problems (Leeser, 2004). The present study, there-fore, seeks to understand the effects of proficiency differences and patterns ofinteraction on L2 learning.

II Background

1 Peer–peer interaction and patterns of pair interaction

Although the role of interaction involving L2 learners has been greatlydebated in the second language acquisition (SLA) literature for over twodecades (see Gass, 1997; Pica, 1994 for a review), much of the researchfocused heavily on quantitative accounts of linguistic behaviour with littleattention to the socially constructed nature of interaction. Recently, however,the nature of interaction and its significance to L2 learning has started toreceive some attention.

One of the key studies is Storch’s (2001a; 2001b; 2002a; 2002b) longitudi-nal investigation into the nature of pair interaction in an adult ESL classroom.Based on her pair talk data, she identified four patterns of interaction amongstpairs. In the collaborative pattern, both learners work together throughout thetask completion process and assist each other. Dominant/dominant pairs, onthe other hand, show an unwillingness or incapability to engage with eachother’s contribution. Dominant/passive pairs involve a dominant participantwho takes control of the task with an authoritarian stance, and a passive peerwho maintains a subservient role. Finally, in expert/novice pairs, the moreknowledgeable learner (expert) actively encourages the less knowledgeablelearner (novice) to engage in the task. Storch also found that more instancesof knowledge transfer took place in the pairs with a collaborative orientation(collaborative and expert/novice) than the pairs with a non-collaborativeorientation (dominant/dominant and dominant/passive). More importantly,a greater number of instances in the non-collaborative pairs showed either

122 Collaborative dialogue between adult ESL learners

no transfer of knowledge or missed opportunities – the lack of engagementabout language items in the pair talk results in errors on individual perform-ance (Storch, 2002a).

2 Peer–peer interaction and proficiency differences

Although Storch’s focus was not on proficiency differences, her 2001a studyrevealed that the pair with the highest proficiency difference (low and upperintermediate) was most collaborative in engaging in the task compared to theother two pairs. Furthermore, the pair with some degree of homogeneity (lowand intermediate) was found to be a non-collaborative dominant/dominantpair, which showed less transfer of knowledge and more missed opportuni-ties. Storch therefore suggested that proficiency differences may not be themajor reason for a non-collaborative orientation.

However, Kowal and Swain (1994; 1997) revealed contradictory findings.Their data with grade eight French immersion students documented that in ahighly heterogeneous grouping (e.g. upper-middle and low), the stronger stu-dent tended to carry out most of the work either because the weaker studentwas too intimidated to say anything, willing to let the stronger student do thetask, or was not allowed to do any of the task whether their opinion was validor not. Successful scaffolding requires the group members to respect oneanother’s perspectives and trust each other’s opinions (Stone, 1993). This maybe difficult to achieve when proficiency differences are too large.

Yule and Macdonald (1990) investigated whether different proficiencypairs could work successfully if each member was given appropriate interac-tive roles. In their task with adult ESL pairs, the more dominant role was toprovide map directions and the less dominant role was to identify the direc-tions with a slightly different map. They found that when the lower profi-ciency member was responsible for the more dominant role, there was morenegotiation of meaning and a successful resolution of referential conflicts.Conversely, when the higher proficiency member played the more dominantrole, they engaged in little negotiation. The higher proficiency membersseemed to ignore their lower proficiency partners’ contribution while thelower proficiency partners often assumed a passive role.

Leeser (2004) focused on the impact of learner proficiency on language-related episodes (LREs)2 – instances of collaborative dialogue (Swain,2001) – in an adult L2 Spanish class. An LRE is defined by Swain and Lapkin(2002) as ‘any part of the dialogue where learners talk about the languagethey produced, and reflect on their language use’ (p. 292). Leeser analysed thefrequency, type (i.e. lexical or grammar-based) and outcome of LREs (i.e.problem solved correctly, not solved or solved incorrectly) produced by threedifferent groupings: high-high, high-low and low-low. He found that as theoverall proficiency of a pair increases, the learners produce a greater numberof LREs, correctly resolve more LREs, and focus more on form than on lex-ical items. Because the high-low pairs fell between the high-high and low-low

Yuko Watanabe and Merrill Swain 123

peers in their performances, Leeser was led to wonder if the high-proficiencylearners actually benefited from their interaction with their low-proficiencypartner, and what the basis was of the improved performances of the low-proficiency partners (relative to those in the low-low groupings). What roledid the nature of the interactions within each pair play? These questions,however, cannot be answered from the quantitative analysis of LREs alone.

Our literature review revealed that despite its significance, the issue ofpeer–peer learning between learners of different L2 proficiency levels elicitedfew studies in the field of SLA. Moreover, the few studies that addressed thisissue examined the interaction of pairs involving different proficiency learn-ers. Such a research design fails to consider how the same learners interactwith higher and lower proficiency peers. The present study was conducted inan attempt to provide insight into this under-explored area, using a moreappropriate research design. The overall research question is how proficiencydifferences and patterns of interaction affect L2 learning. We broke down thisquestion into the following four specific questions:

1) What is the relationship between proficiency differences in pairs andthe frequency of LREs produced?

2) What is the relationship between proficiency differences in pairs and thelearners’ post-test results?

3) What is the relationship between patterns of pair interaction and the frequency of LREs produced?

4) What is the relationship between patterns of pair interaction and thelearners’ post-test results?

III The study

1 Participants

The participants were 12 Japanese learners in a non-credit ESL programme ata Canadian university. We recruited only Japanese learners so that the firstauthor could conduct interviews in Japanese, their first language, whichallowed them to express their thoughts and feelings without any language bar-riers. Furthermore, controlling the L1 background helped to reduce differ-ences in communication styles, which ‘may lead to conflict among . . . groupmembers’ (Allaei and Conner, 1990: 20).

The study involved two different types of participants: core and non-core.Each core participant interacted with two non-core participants whose Englishproficiency was higher and lower than their own. There were four core andeight non-core participants (four higher and four lower). We placed each par-ticipant in one of three proficiency groups (lower, intermediate and higher)using their scores of the shorter version of a model TOEFL.3 Our criterion wasto have about a 50-point difference between each core and non-core participanton the model TOEFL score. We also considered their current ESL class level,

124 Collaborative dialogue between adult ESL learners

Yuko Watanabe and Merrill Swain 125

Table 1 Participants

Non-core: Lower Core: Intermediate Non-core: Higher

Male Name Shu Jun GouModified TOEFL 385 447 538

Name Sota Yoji KenModified TOEFL 390 502 553

Female Name Aya Emi NanaModified TOEFL 405 483 523

Name Rina Mai ChieModified TOEFL 387 440 523

English learning history and available scores for English proficiency tests. Weintentionally formed pairs of the same gender, considering Gass and Varonis’s(1986) findings that Japanese men seem to dominate conversations whenworking with women. We also intended to organize all pairs so that they werefamiliar with each other, but it was impossible to form eight friend–friend pairsdue to their similar proficiency levels. Consequently, only one core participant(Jun) and both of his partners were friends whereas the other core participantsand their partners were not acquainted. Table 1 presents each participant’sname (pseudonym), gender, and test score.

2 Methodology

We collected the data in a university seminar room after classes. Table 2describes the timeframe and events of the data collection procedure for onedyad. The data presented in this paper come from stages 2 to 5.

Stage 1: Pre-task interview. The first author interviewed each core partici-pant to elicit their attitudes towards peer–peer learning in ESL/EFL classes.

Stage 2: Pre-test. At the beginning of the session, each pair engaged in a 5-minute warm-up activity. Following this, the first author described theupcoming writing task and provided a practice session using a sample writ-ing topic. Each pair then jointly wrote a target essay on a specific topic. Thelearners were advised that they could not have access to any aids such as adictionary or the researcher during task completion so that we could see howtwo learners with different proficiency levels jointly solved linguistic prob-lems. The pairs took between 34 and 74 minutes to complete their composi-tion. Their joint text was considered as a pre-test because it represented howmuch the pair could do together without using any external help (Swain andLapkin, 2002).

Reformulation: A native speaker of English reformulated the joint textwritten in stage 2. We asked the reformulator ‘to revise the students’ text toreflect target-language usage while preserving the students’ original meaning’

126 Collaborative dialogue between adult ESL learners

Tab

le 2

Dat

a co

llect

ion

tim

e fr

ame

and

eve

nts

Wee

k 1

Wee

k 2

Th

urs

day

Mo

nd

ayW

edn

esd

ayFr

iday

Sta

ge

1S

tag

e 2

Sta

ge

3S

tag

e 4

Sta

ge

5S

tag

e 6

• P

re-t

ask

inte

rvie

w•

Wri

tin

g (

pre

-tes

t)

• N

oti

cin

g•

Wri

tin

g (

po

st-t

est)

• S

tim

ula

ted

rec

all

• P

ost

-tas

k in

terv

iew

• C

ore

on

ly•

Pai

r•

Pai

r•

Ind

ivid

ual

• C

ore

on

ly•

Co

re o

nly

• in

Jap

anes

e•

in E

ng

lish

• in

En

glis

h•

in E

ng

lish

• in

Jap

anes

e•

in J

apan

ese

• 20

min

.•

no

lim

it•

no

lim

it•

no

lim

it•

30m

in.

• 30

min

.

(Swain and Lapkin, 2002: 291). Reformulation involves having a nativewriter of the target language rewrite the learner’s essay, preserving all thelearner’s ideas, making it sound as nativelike as possible (Cohen, 1983: 4).Learners then compared their original draft and the reformulated version of it.Several studies suggested that this comparison stage promotes learner’s notic-ing (Adams, 2003; Lapkin et al., 2002; Mantello, 1997; Qi and Lapkin, 2001;Swain and Lapkin, 2002; Thornbury, 1997).

Stage 3: Noticing. Each pair was given their original text and the reformu-lated version of it, both typewritten prior to the session. They were asked tonotice and discuss the differences between the two texts. Each pair spent 10to 29 minutes. During stages 2 and 3, the first author took observation notes,noting any salient features of the interaction. Their interactions in these stageswere audio- and video-recorded. We employed video-recording in addition toaudio-recording so that the learners’ paralinguistic expressions such as ges-tures and facial expressions during their interaction could be incorporatedinto the analysis of their interactional patterns. Informal conversation with theparticipants following the data collection revealed that most of them felt thatthe videotaping did not affect their performance or did affect it but only at thebeginning. However, one participant mentioned that he felt as though he wasin an ‘artificial’ atmosphere because of the video-recording.

Stage 4: Post-test. Each pair member received a typed copy of their origi-nal text and was asked to write the essay again, making any changes theywanted. This individually written text was considered to be a post-testbecause the changes the learners individually made to their rewrite of theiroriginal story represent what they learned from noticing the feedback (in thereformulated text) and from their dialogue about it. In many cases, they aredirectly traceable to what the learners noticed in the reformulation relative totheir own text.

Since the core participants were required to go through the same full setof stages twice, after the first time, they would know that they would be testedin stage 4, which could influence their second post-test results. To avoid this,all participants were informed of the upcoming post-test before engaging inthe noticing task with their first pair partner to make the data as comparableas possible. The pair members worked individually and spent 10 to 37minutes.

Stage 5: Stimulated recall. The first author conducted a stimulated recallinterview with each core participant after working with each of their profi-ciency partners.4 Stimulated recall is a type of introspective method in whichprompts such as videotaped interaction of themselves are used to stimulatethe learners’ recall of their thoughts at the time of the activities originally tookplace (Gass and Mackey, 2000). We used this method in order to incorporatethe core participants’ perspective of their behaviour during their interaction sothat we could better understand the nature of their interaction which may notbe apparent from the recordings and transcripts alone. We developed the stim-ulated recall questions based on the core participants’ behaviour observed

Yuko Watanabe and Merrill Swain 127

during their interaction. In particular, we were interested in characteristicsrepresentative of a novice, expert, collaborative, dominant and passive part-ner as per Storch (2001a; 2001b; 2002a; 2002b). For example, according toStorch, a long monologue characterizes dominant behaviour. Thus, if a coreparticipant produced a long monologue during his/her interaction, we showedthem the monologue and asked what he/she was thinking at that time.

Stage 6: Post-task interview. After the stimulated recall, the first authorinterviewed each core participant to elicit their feelings about their interac-tions.

Each core participant performed stages 2 to 4 with two different partners.Since they were required to write an essay with two partners, we prepared twowriting topics as shown in Table 3. The topics were adapted from ‘writingtopics’ in TOEFL to ensure equivalence. In addition, we did not inform theparticipants about their partners’ proficiency levels in order to avoid themhaving preconceived ideas about their partners’ proficiency. However, sinceJun and his partners were all friends, their proficiency levels were known toeach other.

3 Data analysis

a Language-related episodes (LREs): We transcribed the pair talk andinterviews. The transcribed pair talk was coded for LREs in terms of fre-quency.5 We first counted the frequency of LREs per pair in each stage,followed by counting the frequency of turns per LRE. Each time one personspoke, it was counted as one turn. Finally, we calculated the ratio of turns perLRE by each pair in each stage. The first author coded all the transcripts andthe second coder coded 25% of the transcripts. Intercoder reliability was 94%.We compared the frequency of LREs between core-low and core-high pairs.Because of the small sample size (N � 12), this comparison could not beexamined statistically. Therefore, in this paper, we consider any differenceunder 5% as equivalent due to possible error in coding. A difference of 5% to

128 Collaborative dialogue between adult ESL learners

Table 3 Dyad grouping

Core participant Week Partner Writing topic*

Male–Male Jun 1 Shu (Lower) A2 Gou (Higher) B

Yoji 1 Ken (Higher) B2 Sota (Lower) A

Female–Female Emi 1 Aya (Lower) B2 Nana (Higher) A

Mai 1 Chie (Higher) A2 Rina (Lower) B

Note: *See Appendix A.

10% is considered as a trend and a difference greater than 10% is regarded asa difference. We also applied this to the post-test results.

b Patterns of pair interaction: We also analysed the transcribed pair talkfor patterns of interaction. For this analysis, we used Storch’s (2001b) ‘pat-terns of dyadic interaction and associated traits’ as a guideline. Storchdescribes four distinct patterns of interaction and their traits in terms of:1) pattern of contribution; 2) decision-making behaviour; 3) nature of assis-tance; and 4) discourse and linguistic features.6 We coded the pair talk tran-scripts for these traits. Based on the coding for each category, we attemptedto identify each pair as fitting one of the four patterns. The observation notestaken by the first author and the stimulated recalls were used to supplementthis analysis.

The stimulated recall comments helped us to better understand the coreparticipants’ intended role and behaviour in their interaction. For example, atfirst glance, Yoji and Sota’s pattern of interaction during the noticing stagecould be considered as a typical dominant/passive pair because it consisted ofYoji’s long monologues and Sota’s echoic repetition and acknowledgements(Excerpt 1). However, a closer examination of the dialogue led us to specu-late whether Yoji was actually trying to encourage Sota to speak. Yoji’s stim-ulated recall comments (Excerpt 2) confirmed that he was indeed attemptingto involve Sota in the task.

Excerpt 1: Yoji and Sota in stage 37

Yoji: The composer, subject is the composer … so should be ‘expresses’ (looks at Sota)need ‘s’? … “The feelings” … feelings [instead of feeling] … because their, their[feelings]? … “was thinking … thoughts.” Thoughts … because thinking is …Ahh! … Do you know why this one? He [the reformulator] put thoughts [insteadof thinking]?

Sota: ThoughtsYoji: Yeah.Sota: Ohhh … (silence)Yoji: It’s not a gerund, feelings or thoughts, noun. So this is kind of parallel . . . so we

have to put two nouns, not one noun with and a one gerund, ah gerund. So that’swhy he put thoughts [. . .] “So, singing a song is also important because it is usefulto express our feelings.” Again . . . so . . . mmm, our feelings, right? Not my feel-ing, so we have to put ‘s’, right?

Sota: Okay.Yoji: “In short because music affects on our life.” Not on our life. First . . . do you know

why?

Excerpt 2: Yoji’s stimulated recall

I know I asked him persistently because I really wanted him to say something. [. . .] Iknew the answers. But if I was solving them by myself, it’s like, ‘why is Sota here?’ [. . .]This activity is pair work isn’t it? So, if I was doing it by myself, what is this for? (trans-lated from Japanese)

Yuko Watanabe and Merrill Swain 129

c Pre- and post-tests: We scored the pre-test by counting the number ofitems that the reformulator changed. Since the number of words in each pair’stext was different, the number of reformulated items was calculated as apercentage of the total number of words. This percentage is considered to bethe pre-test score. We coded the post-test by noting whether each participantgot the reformulated items right or wrong. Items were considered correct if: 1)they match the reformulated version, and 2) they were an acceptable correctalternative (Swain and Lapkin, 2002). After counting the number of correctitems, this number was calculated as a percentage of the total number of refor-mulated changes from the pre-test. This percentage is regarded as the post-testscore. The first author coded all the tests and a second coder coded 25% of thepre- and post-test data. Intercoder reliability was 93%.

IV Findings

1 What is the relationship between proficiency differences in pairs andthe frequency of LREs?

Table 4 presents the mean frequency and range of LREs and turns, and theratio of turns per LRE for the core-low and core-high pairs in stages 2 (writ-ing) and 3 (noticing). We see that the core-high pairs produced a higherfrequency of LREs (M � 29.3) and turns (M � 114) than those of the core-low pairs (Mean LREs � 19; Mean turns � 80.3) during stage 2. However, instage 3, the core-low pairs generated about the same frequency of LREs(M � 20.5) as the core-high pairs (M � 19.8). This may be related to theamount of reformulation that each group received. That is, the trend indicatedthat the core-low pairs received slightly more reformulations (18%, seeTable 5) than the core-high pairs (11.5%, see Table 5) and hence, the core-lowpairs had more items to discuss during the noticing stage.

In terms of the ratio of turns per LRE, little difference is observed in thehighest range between the core-low (4.6 in stage 2; 14.2 in stage 3) and core-high (4.6 in stage 2; 14.1 in stage 3) pairs. However, when examining thelowest range, we can see that the core-high pairs produced more turns (2.9 instage 2; 7.8 in stage 3) than the core-low pairs (1 in stage 2; 3.7 in stage 3).This suggests that on average, the core-high pairs talked longer per episodethroughout the two stages than the core-low pairs.

Table 4 also shows that both core-low and core-high pairs produced farmore turns per LRE during stage 3 (M � 9.8 and 9.2 respectively) than theydid in stage 2 (M � 4.2 and 3.9 respectively). This suggests that thenoticing task stimulated longer collaborative dialogues than the writing task.When considering the average time that both core-low and core-high pairsspent for each task (57 and 58 minutes for writing; 21 and 17 minutes fornoticing), it is apparent that the noticing task generated more LREs than thewriting task.

130 Collaborative dialogue between adult ESL learners

Yuko Watanabe and Merrill Swain 131

Tab

le 4

Mea

n f

req

uen

cy a

nd

ran

ge

of

LRE

s an

d t

urn

s, a

nd

th

e ra

tio

of

turn

s p

er L

RE

fo

r th

e co

re-l

ow

an

d c

ore

-hig

h p

airs

in S

tag

es 2

an

d 3

Freq

uen

cyC

ore

– L

ow

C

ore

– H

igh

(M

ean

fo

r 4

dya

ds)

(Mea

n f

or

4 d

yad

s)

LRE

sTu

rns

Turn

/Ti

me

LRE

/LR

Es

Turn

sTu

rn/

Tim

e LR

E/

LRE

on

tas

k m

inu

te

LRE

on

tas

k m

inu

te(m

in)

(min

)

S2*

(W

riti

ng

)19

80.3

4.2

570.

329

.311

43.

958

0.5

S3

(No

tici

ng

)20

.520

09.

821

119

.818

2.3

9.2

171.

2

Ran

ge

Co

re –

Lo

wC

ore

– H

igh

S2:

Hig

hes

t41

190

4.6

70–

3616

44.

674

–Lo

wes

t1

11

36–

1976

2.9

34–

S3:

Hig

hes

t24

337

14.2

29–

4034

414

.130

–Lo

wes

t16

773.

717

–8

113

7.8

10–

Not

e:*S

2�

Sta

ge

2.

2 What is the relationship between proficiency differences in pairs andthe learners’ post-test results?

Table 5 compares the post-test scores between core-low and core-high pairs.We see that the core-high pairs as a group scored higher on average (63%) thanthe core-low pairs as a group (50%). In spite of this, the average of the coreparticipants’ individual post-test scores after working with their higher profi-ciency partners was slightly lower (58%) than their scores after working withtheir lower proficiency partners (64%). In other words, although the differencewas small, the trend indicated that the core participants achieved on averagehigher scores when working with their lower proficiency partners.

3 What is the relationship between patterns of pair interaction and thefrequency of LREs produced?

Among the eight pairs in this study, seven fell under one of Storch’s (2002a)patterns of pair interaction, yet one pair did not neatly fit into any of her cat-egories. The seven pairs were identified as the collaborative (3 pairs),expert/novice (3 pairs), and dominant/passive (1 pair) patterns of interaction.No instance of a dominant/dominant pair was observed in this study. Excerpts3, 4 and 5 show examples of each of the three patterns of interaction, eachfollowed by an explanation of characteristics of each pattern as outlined byStorch (2002a).

132 Collaborative dialogue between adult ESL learners

Table 5 Pre- and post-test scores in core-low and core-high pairs

Pre-test* Post-test** Range (%)(% of (% of items matching reformulations) the reformulation) Highest Lowest

Core-Low 18 Core & Low (N � 8) 50(Mean for Core (n � 4) 64 72 574 dyads) Low (n � 4) 36 52 17

Core-High 11.5 Core & High (N � 8) 63(Mean for Core (n � 4) 58 67 494 dyads) High (n � 4) 68 74 65

Notes: *The pre-test was scored by counting the number of reformulated items. Sincethe total words in each pair’s text were different, the number of reformulated itemswas calculated as a percentage of the total words in each text. This percentage isconsidered to be the pre-test score; **The post-test score was coded by notingwhether each participant got the reformulated items right or wrong. Items wereconsidered correct if: 1) they matched the reformulated version, and 2) they were anacceptable correct alternative (Swain and Lapkin, 2002). The number of correct itemswas calculated as a percentage of the total number of reformulations from the pre-test. This percentage is regarded as the post-test score.

Excerpt 3: Collaborative (Jun and Gou in stage 2)

265 Gou: [. . .] diminish, deplete like decreased? But not decreased.266 Jun: Reduced?267 Gou: Reduced, yes. “Although the carrying capacity of airplanes is smaller than

that of ships” . . . airplanes have reduced . . . reduced . . .268 Jun: Time of?269 Gou: Time?270 Jun: I hope271 Gou: I don’t know . . . time. What kind of time?272 Jun: Time um time . . . ah273 Gou: Time of transportation? Not transportation274 Jun: No no no . . . time of the . . . maybe quick275 Gou: Quick?276 Jun: Just time of trade?277 Gou: Um-hmm. Right. Business letter is important for . . . ah but we can use fax.278 Jun: Hmmm. How about, I don’t know, airplanes brought new products uhh

which . . . uh . . .279 Gou: What do you wanna say? Airplanes?

In Excerpt 3, Gou and Jun work together on all parts of the task and are will-ing to offer and engage with each other’s ideas. Furthermore, they engagecritically but constructively with one another’s suggestions (lines 271, 274),leading to resolutions that seem acceptable to both of them.

Excerpt 4: Expert/Novice (Emi and Nana in stage 2)

333 Nana: What do you want to say next?334 Emi: I want, I wanna say um . . . that um . . . actually I wanna say, how he, not how,

I wanna say he’s really um . . . good, good teacher of my life.335 Nana: Yeah, yeah, yeah.336 Emi: He taught me a lot.337 Nana: Ah, I know that how important he is for you.338 Emi: Yeah.339 Nana: But ah . . .340 Emi: Yeah but generally341 Nana: Yes, I know, but ah in this paragraph, we have to write about how important

thinking342 Emi: about him343 Nana: Yes.

In Excerpt 4, Nana seems to assume or is afforded the role of the expertand leads the task. However, unlike a dominant role, Nana actively encour-ages Emi, the novice, to participate in the task (line 333) and providesassistance that will help Emi learn from the interaction (line 341). ThoughNana is authoritative, she is not necessarily being authoritarian (van Lier,1996).

Yuko Watanabe and Merrill Swain 133

Excerpt 5: Dominant/Passive (Jun and Shu in stage 3)

170 Jun: “Music is important to many people to get energy and for healings,” for heal-ings, to get energy for healings . . . because it’s, because it’s noun here.

171 Shu: for healings172 Jun: Yeah, but yeah, I have some, what can I say, I wonder this sentence. Why we

don’t write down like here, like this? We can [write] “people can get somehealing.” Anyway some, some sounds, some [is] a little bit strange. But, peoplecan get healing . . . I don’t know. For our essay, “people can get healing bylistening to music.” Ahh . . . no no no, I don’t know . . . No, I don’t think so.

173 Shu: can get healing . . . mm.174 Jun: Anyway, we can . . . yeah, maybe it’s better.175 Shu: . . . maybe it’s better.

In Excerpt 5, we can observe an obvious unequal contribution between Jun, the dominant participant, and Shu, the passive partner. Jun seems to see the task as an individual rather than joint construction and makes little attempt to involve Shu. In fact, many of Jun’s utterances are self-directed (lines 170, 172). There is little negotiation between them and Shu’s contribution is limited to echoic repetitions (lines 171,172, 175).

The pair that could not be identified as per Storch’s classification wasconsidered to be an expert/passive pattern of interaction. In the expert/passive pair, despite the ongoing encouragement of the more proficient expert participant, the less proficient passive participant’s involvement in the task decreased over time as he became intimidated and reluctant to say anything in front of his expert partner (refer to Excerpt 1). Such apattern of interaction was not reported in Storch’s study. We consider the expert/passive pattern of interaction to be a non-collaborative orientation.

Table 6 shows the relationship between the patterns of interaction found in our study and the frequency of LREs produced in stage 2 by each pair. Since the frequency of LREs in stage 3 was influenced by the amount of reformulation that each pair received, we analysed only the frequency of LREs in stage 2. There are two distinctive trends: 1) both collaborative (n � 3) and expert/novice (n � 3) pairs produced a higher frequency of LREs (M � 35 and 28 respectively) than the expert/passive pair (n � 1; the frequency of LREs � 1) and dominant/passive pair (n � 1; the frequency of LREs � 4); and 2) a similar tendency can beobserved for the ratio of turns per LRE. Whereas both collaborative andexpert/novice pairs produced an average of 4.0 to 4.1 turns per LRE,expert/passive and dominant/passive pairs generated only 1.0 to 1.5 turns per LREs. In short, the pairs with a collaborative orientation produced agreater frequency of LREs and more turns per LRE than the pairs with a non-collaborative orientation.

134 Collaborative dialogue between adult ESL learners

4 What is the relationship between patterns of pair interaction andtheir post-test results?

Table 7 presents the relationship between the learners’ patterns of interactionand their post-test scores. In terms of the core participants’ scores in the core-high pairs, Jun (65%) and Mai (67%) performed better than other core partic-ipants (Emi, 50%; Yoji, 49%). Both of these high achievers’ roles in theirinteraction were shown as collaborative, whereas both of the low achievers’roles were found to be novice. As for the core participants’ scores in the core-low pairs, we can see Emi (72%) and Mai (67%) attained higher scores thanthe other core participants (Jun, 59%; Yoji, 57%). These high achievers’ rolesin their interaction were either expert (in the expert/novice pattern) or collab-orative. Conversely, those who achieved lower scores were either dominant orexpert (in the expert/passive pattern). In short, when the core participants’ rolewas either collaborative or expert (in the expert/novice pattern), they weremore likely to achieve higher post-test scores than those who acted as domi-nant, novice or expert (in the expert/passive pattern). On the other hand, whenthe core participants’ role was dominant, novice or expert (in the expert/pas-sive pattern), they were more likely to attain lower post-test scores than thosewho acted as collaborative or expert (in the expert/novice pattern).

As for the non-core participants, when examining the higher proficiencypartners’ post-test scores, they all scored well. Their roles in their interactionwere found to be either collaborative or expert (in the expert/novice pattern).Similarly, when looking at the lower proficiency partners’ post-test scores,Rina, who was collaborative, achieved the highest score (52%). Other learn-ers who were passive or novice scored lower than Rina.

Overall, when the learners engaged in a collaborative pattern of interaction,both core and non-core participants achieved higher post-test scores thanother pairs with different patterns of interaction. In contrast, when the learn-ers engaged in the dominant/passive or expert/passive pattern, both core and

Yuko Watanabe and Merrill Swain 135

Table 6 Patterns of interaction and the frequency of LREs in Stage 2

Pair Pattern of interaction LRE Turns Turn/LRE

Yoji – Sota (Lower) Expert/Passive 1 1 1.0Yoji – Ken (Higher) Expert/Novice 35 134 3.8Emi – Aya (Lower) Expert/Novice 30 124 4.1Emi – Nana (Higher) Expert/Novice 19 76 4.0

Mean for Expert/Novice 28 111.3 4.0Jun – Shu (Lower) Dominant/Passive 4 6 1.5Jun – Gou (Higher) Collaborative 36 164 4.6Mai – Rina (Lower) Collaborative 41 190 4.6Mai – Chie (Higher) Collaborative 27 80 3.0

Mean for Collaborative 35 144.7 4.1

136 Collaborative dialogue between adult ESL learners

Tab

le 7

Pat

tern

s o

f in

tera

ctio

n a

nd

po

st-t

est

sco

res

Co

re-H

igh

Co

re p

arti

cip

ant

Hig

her

pro

fici

ency

par

tner

Pai

r (P

atte

rn o

f in

tera

ctio

n)

Nam

eR

ole

Sco

re (

%)

Nam

eR

ole

Sco

re (

%)

Yoji

– K

en (

Exp

ert/

No

vice

)Yo

jiN

ovi

ce49

Ken

E

xper

t74

Em

i – N

ana

(Exp

ert/

No

vice

)E

mi

No

vice

50

Nan

a E

xper

t67

Jun

– G

ou

(C

olla

bo

rati

ve)

Jun

C

olla

bo

rati

ve65

G

ou

Co

llab

ora

tive

65M

ai –

Ch

ie (

Co

llab

ora

tive

)M

aiC

olla

bo

rati

ve67

Ch

ie

Co

llab

ora

tive

67

Co

re-L

ow

Co

re p

arti

cip

ant

Low

er p

rofi

cien

cy p

artn

er

Pai

r (P

atte

rn o

f in

tera

ctio

n)

Nam

eR

ole

Sco

re (

%)

Nam

eR

ole

Sco

re (

%)

Yoji

– S

ota

(E

xper

t/P

assi

ve)

Yoji

Exp

ert

57S

ota

Pas

sive

32

Em

i – A

ya (

Exp

ert/

No

vice

)E

mi

Exp

ert

72A

yaN

ovi

ce17

Jun

– S

hu

(D

om

inan

t/P

assi

ve)

Jun

Do

min

ant

59S

hu

Pas

sive

41M

ai –

Rin

a (C

olla

bo

rati

ve)

Mai

Co

llab

ora

tive

67R

ina

Co

llab

ora

tive

52

non-core participants attained lower post-test scores than those who acted ascollaborative and expert (in the expert/novice pattern). In the expert/novicepairs, the expert achieved higher scores while the novice attained lowerscores. Additionally, all the dominant and expert participants performedbetter than their passive or novice partners.

V Discussion

The purpose of this study was to explore the effects of proficiency differencesand patterns of interaction on L2 learning through the examination of collab-orative dialogue and post-test performance. The first two research questionsaddressed how proficiency differences affect LREs and post-test scores. Interms of LREs, the core-high pairs produced a greater frequency of LREsthan that of the core-low pairs. This is consistent with the previous studiesthat as the overall proficiency of the pair increased, learners produced agreater frequency of LREs (Leeser, 2004; Williams, 2001). Since severalstudies agreed that LREs represent L2 learning in progress (e.g. Basturkmenet al., 2002; Leeser, 2004; Swain, 1998; Swain and Lapkin, 1998; Williams,2001), this could suggest that the core participants benefited more from work-ing with their higher proficiency partner. However, the trend indicated that thecore participants achieved on average higher post-test scores when workingwith their lower proficiency partners than their higher proficiency partners.8

In other words, the core participants learned more when working with lowerproficiency peers than higher proficiency peers, suggesting that there is cer-tainly value for more proficient students to be paired with less proficientpeers. This led to the third and fourth research questions, which addressed theeffects of patterns of pair interaction on LREs and post-test scores.

In terms of LREs, we found that the pairs with a collaborative orientation(collaborative and expert/novice) produced more LREs than the pairs with anon-collaborative orientation (dominant/passive and expert/passive). Thisbrings us back to our findings about the frequency of LREs that the core-highpairs produced on average more LREs than that of the core-low pairs. Onemight question whether this is due to proficiency difference or a pattern ofinteraction. Interestingly, the pairs who produced the most (n � 41) and theleast LREs (n � 1) were in fact both core-low pairs (see Table 4). Moreover,the pair with the most LREs was found to be the collaborative pair while thepair with the least LREs was a non-collaborative, expert/passive pair. Giventhis, in this study, the patterns of interaction rather than proficiency differ-ences seemed to have a more important effect on the frequency of LREs.

As for the post-test scores, we found that both of the pair membersachieved higher post-test scores when they engaged in the collaborative pat-tern of interaction, whereas both achieved lower scores when they engaged inthe dominant/passive or expert/passive patterns. Given these findings, thedifference in the core participants’ post-test scores after the core-low and

Yuko Watanabe and Merrill Swain 137

core-high interactions were more likely attributable to the difference in pat-terns of interaction rather than their partners’ proficiency levels. That is, pro-ficiency difference in pairs does not necessarily affect the nature of peerassistance and L2 learning. It is important to note, however, that the profi-ciency difference may create a different pattern of interaction.

Our findings substantiate Storch’s (2001a; 2001b; 2002a; 2002b) claim thatpairs with a collaborative orientation are more likely to learn than those witha non-collaborative orientation, except her findings involving theexpert/novice pair. Although Storch’s data indicated that both collaborativeand expert/novice pairs exhibited more transfer of knowledge than other non-collaborative pairs, our findings demonstrated that when a large proficiencydifference exists within a pair, only the collaborative pattern of interactionmight be conducive to L2 learning for both learners. In the expert/novicepairs, only the expert seemed to benefit from their interaction. In fact, theexperts in the expert/novice pairs achieved three of the highest post-testscores (Ken, 74%; Emi, 72%; Nana, 67%) among all participants. Given thatthese expert partners had more opportunities to provide assistance to theirnovice partners, this provides strong evidence that students can indeed learnfrom the act of teaching other peers (van Lier, 1996).

Leeser (2004) questioned whether lower proficiency learners actually ben-efit from receiving higher proficiency learners’ assistance because they mightnot be developmentally ready to discuss some linguistic problems. Our find-ings provide some supportive evidence. In the case of the expert/novice pairswith a large proficiency difference, the less proficient novice learners seemedto have difficulties internalizing all the information coming from the moreproficient expert peers. Consequently, it might have been difficult for less pro-ficient novice participants to change the reformulated items. It is also possi-ble that given Johnson and Johnson’s (1989) claim that ‘the amount of timespent explaining correlates highly with the amount learned’ (p. 57), those lessproficient novice participants had fewer opportunities to explain languageissues to their peers during their interaction, and hence made it difficult forthem to remember these language problems in their post-test. It might also berelated to the novice learners’ affective aspects. Since peer–peer learninginvolves strong emotions that can influence its outcome (Swain and Miccolli,1994), it is conceivable that they might have felt pressured or intimidated towork with the more proficient learner who acted as the expert. In any case, itis important to note that our interpretation is based on a small sample and thatindividual differences also need to be considered.

Overall, we have seen that peers of different proficiency levels could ben-efit from working with one another, which supports the previous peer–peerlearning research (e.g. Ohta, 2001; Storch, 2001a; Swain and Lapkin, 1998).This is a positive finding. It shows that social mediation comes not only froman expert such as teachers but also from peers, and even from less proficientpeers. Therefore, SLA researchers and teachers should be careful not toassume that grouping different proficiency peers is less conducive to L2

138 Collaborative dialogue between adult ESL learners

learning. However, our data also suggested that grouping different proficiencylearners is conducive to L2 learning when they are collaborative. Thus, it isimportant to pay close attention to the pattern of interaction and how thesepatterns form.

From a methodological perspective, the use of a stimulated recall interviewnot only enabled us to better understand the quantitative data but also topursue the intriguing aspects of what was actually going on in the learners’minds during peer–peer collaborative dialogue. Although we provided onlyone example (Excerpt 2) in this article due to the limited space, there were anumber of cases in which stimulated recall elucidated our analysis of the pat-terns of interaction. In future studies, therefore, it will be important to incor-porate the insider’s point of view through stimulated recall or follow-upinterviews.

From a research perspective, this study confirmed the importance of con-sidering peer–peer collaborative dialogue as a mediator of L2 learning. Therelationship between the learners’ post-test scores and their patterns of inter-action highlights how L2 learning ‘occurs in interaction, not as a result ofinteraction’ [italic: authors’ emphasis] (Swain et al., 2002: 173). The studysupports the claim that collaborative dialogue is a useful unit of analysis toexplore the process and product of L2 learning (Swain, 2000; Wells, 2000),which goes ‘beyond output’ (Swain, 2000). Although this study involved a small sample size and was conducted in a controlled research setting for ashort period of time, we hope that it provides insight into the complex natureof peer–peer collaborative dialogue and its importance to L2 learning.

Acknowledgements

We would like to thank all of our participants for taking part in this study. We are also grateful to Sharon Lapkin for her valuable feedback on an earlierversion of this paper.

Yuko Watanabe and Merrill Swain 139

Notes1 Following Leeser (2004), we will use the term proficiency to refer to ‘a learner’s general language abil-

ity in speaking, listening, reading and/or writing based on some kind of criteria or measure’ (p. 58).2 For more detailed discussion about LREs and examples, see Swain and Lapkin (1998) and Leeser

(2004).3 Not all participants were able to spare the time for a full-length TOEFL. We selected one of the

model tests (Sharpe, 2001) and cut each section to approximately half its original length. Their testscores were calculated using the paper-based TOEFL score conversion chart provided in one of theTOEFL study guides (Murakawa, 1997). Since the test was shortened and the scores were not cal-culated by an expert, these scores might not be comparable to paper-based TOEFL scores.

4 We could not conduct stimulated recall interviews with non-core participants due to time con-straints. We consider this to be one of the limitations of this study.

5 We also coded LREs for types and outcomes; however, only the frequency of LREs is discussedin this article.

6 Although Storch (2001b) analysed both discourse (e.g. requests, questions) and linguistic (use ofpronouns) features, we only examined discourse features in this study.

140 Collaborative dialogue between adult ESL learners

7 The transcription conventions used in this study:[ ] Words or phrases that were omitted from the speech or clarification of the information unclearto the reader( ) Paralinguistic information or speaker’s action“ ” A speaker is reading a written text. . . Short pause, between 0.5 and 3 seconds

8 One might suggest that the joint text with their higher proficiency partner might have been morechallenging for the core participants and thus, they might not have been able to correct morereformulated items when working with their higher proficiency partner. However, one coreparticipant, Mai, scored exactly the same with both high and low partners (67%) on the post-test.More interestingly, the other core participant, Jun, scored higher with his higher proficiencypartner (65%) than with his lower proficiency partner (59%). Given the considerable individualdifferences, we do not feel that the quality of the joint text is the major reason for the trenddiscussed here.

VI References

Adams, R. 2003: L2 output, reformulation and noticing: implications for IL develop-ment. Language Teaching Research 7: 347–76.

Allaei, S.K. and Connor, U.M. 1990: Exploring the dynamics of cross-cultural col-laboration in writing classrooms. The Writing Instructor 10: 19–28.

Basturkmen, H., Loewen, S. and Ellis, R. 2002: Metalanguage in focus on form inthe communicative classroom. Language Awareness 11: 1–13.

Cohen, A. 1983: Reformulating compositions. TESOL Newsletter 17: 1–5.Educational Testing Service 2003: The official web site for the Test of English as

a Foreign Language (TOEFL). Princeton, NJ: Educational Testing Service.�http://www.toefl.org/index.html�, accessed October 2003.

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

Gass, S. and Mackey, A. 2000: Stimulated recall methodology in second languageresearch. Mahwah, NY: Lawrence Erlbaum.

Gass, S. and Varonis, E. 1986: Sex differences in NNS/NNS interactions. In Day, R.,editor, Talking to learn: conversation in second language acquisition. Rowley,MA: Newbury House, 327–51.

Guerrero, M.C.M. de. and Villamil, O.S. 1994: Socio-cognitive dimensions of inter-action in L2 peer revisions. Modern Language Journal 78: 484–96.

Johnson, D.W. and Johnson, R.T. 1989: Learning together and alone: cooperative,competitive and individualistic learning, fifth edition. Upper Saddle River, NJ:Allyn and Bacon.

Kowal, M. and Swain, M. 1994: Using collaborative language production tasks topromote students’ language awareness. Language Awareness 3: 73–93.

—— 1997: From semantic to syntactic processing: how can we promote it in theimmersion classroom? In Johnson, K. and Swain, M., editors, Immersioneducation: international perspectives, Cambridge: Cambridge UniversityPress, 284–309.

Lapkin, S., Swain, M. and Smith, M. 2002: Reformulation and the learning ofFrench pronominal verbs in a Canadian French immersion context. ModernLanguage Journal 86: 485–507.

Yuko Watanabe and Merrill Swain 141

Leeser, M.J. 2004: Learner proficiency and focus on form during collaborativedialogue. Language Teaching Research 8: 55–82.

Lockhart, C. and Ng, P. 1995: Analyzing talk in ESL peer response groups: stances,functions and content. Language Learning 45: 605–55.

Mantello, M.A. 1997: Error correction in the L2 classroom. Canadian ModernLanguage Review 54: 127–31.

Murakawa, H. 1997: TOEFL yosou mondai 700 [TOEFL 700 anticipated questions].Tokyo: Obunsya.

Nelson, G.L. and Murphy, J.M. 1993: Peer response groups: do L2 writers use peercomments in revising their drafts? TESOL Quarterly 27: 135–41.

Ohta, A.S. 2001: Second language acquisition processes in the classroom: learningJapanese. Mahwah, NJ: Lawrence Erlbaum.

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

Qi, D. and Lapkin, S. 2001: Exploring the role of noticing in a three-stage secondlanguage writing task. Journal of Second Language Writing 10: 277–304.

Sharpe, P.J. 2001: How to prepare for the TOEFL test: test of English as a foreignlanguage, tenth edition. New York: Barron’s Educational Series.

Stone, C.A. 1993: What is missing in the metaphor of scaffolding? In Forman, E.A.,Minick, N. and Stone, C.A., editors, Contexts for learning: socioculturaldynamics in children’s development, Oxford: Oxford University Press, 169–83.

Storch, N. 2001a: How collaborative is pair work? ESL tertiary students composingin pairs. Language Teaching Research 5: 29–53.

—— 2001b: An investigation into the nature of pair work in an ESL classroom andits effect on grammatical development. Unpublished doctoral dissertation,University of Melbourne.

—— 2002a: Patterns of interaction in ESL pair work. Language Learning 52:119–58.

—— 2002b: Relationships formed in dyadic interaction and opportunity for learning.International Journal of Educational Research 37: 305–22.

Swain, M. 1985: Communicative competence: some roles of comprehensible inputand comprehensible output in its development. In Gass, S. and Madden, C.,editors, Input in second language acquisition, Rowley, MA: Newbury House,235–53.

—— 1995: Three functions of output in second language learning. In Cook, G. andSeidlhofer, B., editors, Principles and practice in applied linguistics: studies inhonour of H. G. Widdowson, Oxford: Oxford University Press, 125–44.

—— 1998: Focus on form through conscious reflection. In Doughty, C. andWilliams, J., editors, Focus on form in classroom second language acquisition.Cambridge: Cambridge University Press, 64–81.

—— 2000: The output hypothesis and beyond: mediating acquisition through collab-orative dialogue. In Lantolf, J.P., editor, Sociocultural theory and secondlanguage learning, Oxford: Oxford University Press, 97–114.

—— 2001: Examining dialogue: another approach to content specification and tovalidating inferences drawn from test scores. Language Testing 18: 319–46.

142 Collaborative dialogue between adult ESL learners

Swain, M., Brooks, L. and Tocalli-Beller, A. 2002: Peer-peer dialogue as means ofsecond language learning. Annual Review of Applied Linguistics 22: 171–85.

Swain, M. and Lapkin, S. 1998: Interaction and second language learning: twoadolescent French immersion students working together. Modern LanguageJournal 82: 320–38.

—— 2002: Talking it through: two French immersion learners’ response to reformu-lation. International Journal of Educational Research 37: 285–304.

Swain, M. and Miccolli, L.S. 1994: Learning in a content-based, collaborativelystructured course: the experience of an adult ESL learner. TESL CanadaJournal 12: 15–28.

Thornbury, S. 1997: Reformulation and reconstruction: tasks that promote ‘notic-ing’. ELT Journal 51: 326–35.

van Lier, L. 1996: Interaction in the language curriculum: awareness, autonomy, andauthenticity. Harlow, UK: Longman.

Vygotsky, L.S. 1978: Mind in society. Cambridge, MA: Harvard University Press.—— 1986: Thought and language. Cambridge, MA: MIT Press.Wells, G. 2000: Dialogic inquiry in education: building on the legacy of Vygotsky. In

Lee, C.D. and Smagorinsky, P., editors, Vygotskian perspectives on literacyresearch, New York: Cambridge University Press, 51–85.

Williams, J. 2001: The effectiveness of spontaneous attention to form. System 29:325–40.

Yule, G. and Macdonald, D. 1990: Resolving referential conflict in L2 interaction:the effect of proficiency and interactive role. Language Learning 40: 539–56.

Appendix A: Writing topics used in the task

Sample topic

Many students choose to attend schools or universities outside their homecountries. Why do some students study abroad? Use specific reasons anddetails to explain your answer.

Topic A

People listen to music for different reasons and at different times. Why ismusic important to many people? Use specific reasons and details to explainyour answer.

Topic B

Choose one of the following transportation vehicles and explain why youthink it has changed people’s lives.

• automobiles• bicycles• airplanes

Use specific reasons and examples to support your answer.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.