Comprehensible Output, From Occurrence to Acquisition:An Agenda for Acquisitional Research

Language Learning 52:3, September 2002, pp. 597–647 Comprehensible Output, From Occurrence to Acquisition: An Agenda for Acquisitional Research Ali Shehadeh University of Aleppo King Saud University After over a decade of research into Swain’s (1985) comprehensible output (CO) hypothesis, there is still a severe lack of data showing that learner output or output modifications have any effect on second-language (L2) learning. Izumi and Bigelow (2000, p. 245) argued that this is because, in most cases, researchers assumed rather than showed whether and how output helps with language learning. In this article, I will argue that this, in turn, is because existing research on CO was mostly descriptive in nature, focusing primarily on occurrence per se rather than acquisition or whether and how output can be a source of competence in the L2. I will outline a research agenda that makes acquisitional research central to the study of CO. The goal of this article is to propose a research agenda that makes acquisitional research central to the study of comprehensible output. It begins by looking at the context in which the comprehensible output (CO) hypothesis was proposed, namely, the need to look beyond comprehensible input as a condition for second language acquisition (SLA), which surfaced in the need to explain I acknowledge the help I received from Nick Ellis and a number of anonymous reviewers, and I would like to thank them all for their advice and many insightful suggestions and comments. Correspondence concerning this article may be addressed to Ali Shehadeh, College of Languages and Translation, King Saud University, P.O. Box 87907, Riyadh 11652, Saudi Arabia. Internet: ashhada@ksu.edu.sa 597 598 Language Learning Vol. 52, No. 3 the lack of theorized effect on SLA on a clear case of comprehensible input as witnessed in the Canadian immersion data. The article then surveys and assesses the research conducted to date into CO. I will show that much of existing research has been descriptive in nature, focusing primarily on occurrence rather than acquisition. Finally, I will outline areas and directions for further investigation. In a seminal article, Swain (1985) argued that comprehensible input is not sufficient for successful SLA but that opportunities for nonnative speakers (NNSs) to produce CO are also necessary. She based her conclusions on findings from studies she conducted in immersion contexts in Canada. Swain (1984, 1985) found that although immersion students were provided with a rich source of comprehensible input over a period of 8 years, their interlanguage (IL) performance was still off-target; that is, they were clearly identifiable as nonnative speakers or writers. In particular, Swain found that the expressive performance of these students was far weaker than that of same-aged native speakers (NSs) of French. For example, they evidenced less knowledge and control of complex grammar, less precision in their overall use of vocabulary and morphosyntax, and lower accuracy in pronunciation. Swain (1985, p. 249) argued that the IL performance of these students was still off-target because they lacked opportunities for output in two ways: “First, the students are simply not given— especially in later grades—adequate opportunities to use the target language in the classroom context. Second, they are not being ‘pushed’ in their output.” Swain goes on to say that “there appears to be little social or cognitive pressure to produce language that reflects more appropriately or precisely their intended meaning: there is no push to be more comprehensible than they already are” (p. 249). In other words, what immersion students needed was not just comprehensible input, but also opportunities for CO in order to be both fluent and accurate in the second language (L2). Thus, Swain claimed that understanding target language (TL) utterances is not enough and that learners must also be given the opportunity to produce them. She therefore Shehadeh 599 doubted that interactions and comprehensible input on their own are sufficient for SLA: Conversational exchanges . . . are not themselves the source of acquisition derived from comprehensible input. Rather they are the source of acquisition derived from comprehensible output: output that extends the linguistic repertoire of the learner as he or she attempts to create precisely and appropriately the meaning desired. (Swain, 1985, p. 252) Swain (1985) proposed a hypothesis relating to the L2 learner’s production comparable to Krashen’s comprehensible input hypothesis (for this, see, for example, Krashen, 1985), arguing that the role of learner production of CO is independent in many ways of the role of comprehensible input. This she termed the “comprehensible output hypothesis” for SLA (p. 249). The basic premise of the CO hypothesis postulates that producing the L2, especially when learners experience difficulties in communicating their intended messages successfully, “pushes” learners to make their output more precise, coherent, and appropriate and that this process contributes to SLA. By the same token, Swain posited that output “may be the trigger that forces the learner to pay attention to the means of expression needed in order to successfully convey his or her own intended meaning” (p. 249). Swain acknowledged the role of comprehensible input in SLA but argued that CO is also necessary because it aids SLA in many ways: “Its role is, at minimum, to provide opportunities for contextualized, meaningful use, to test out hypotheses about the target language, and to move the learner from a purely semantic analysis of the language to a syntactic analysis of it” (p. 252). CO has also been an area of concern in studies of negotiation. Several SLA researchers have drawn attention to the modifications that NNSs make to their IL utterances (phonological, morphosyntactic, and lexical) when interlocutors signal difficulty in understanding, arguing that when NNSs are asked by NSs to clarify their output in the course of negotiation of meaning, they will produce CO that provides comprehensible input to NSs. They 600 Language Learning Vol. 52, No. 3 will do this by modifying their IL performance toward more accurate, comprehensible production (e.g., Hatch, Flashner, & Hunt, 1986; Lyster & Ranta, 1997; Pica, 1994; Pica, Holliday, Lewis, & Morgenthaler, 1989; Sato, 1986; Van den Branden, 1997). For instance, in her review article on negotiation, Pica (1994, p. 498) noted the importance of modified output (MO) to learners’ participation in negotiation, pointing out that negotiation research showed that when NNSs are asked by interlocutors to clarify their output, they will reprocess and modify their IL utterances (Swain, 1993) in the direction of greater message comprehensibility, as in Example 1, which shows syntactic modification through embedding and elaboration in clauses. 1. NNS: NS: NNS: you have aa three which is . . . white square of which appears sharp huh? you have aa three houses one is no-no-not- one is not square and one is square, but with a little bit-a small house (Pica et al., 1989, p. 89) Similarly, Van den Branden (1997, pp. 626–627) emphasized the need to look at output as he addressed questions of learner accuracy and completeness in his study of negotiation outcomes. In particular, Van den Branden noted that during negotiations “learners can be pushed to produce far more than merely CO; they can be pushed to the production of output that is more complete and accurate” (p. 630). Lyster and Ranta (1997) also drew attention to learners’ ability to achieve self-repair that leads to more accurate and more CO as they participate in negotiation that involves the provision of corrective feedback, as in Example 2. 2. Student: Teacher: La marmotte c’est pas celui en haut? [Error-gender] “The groundhog isn’t the one on top?” Pardon? “Pardon?” Shehadeh Student: 601 La marmotte c’est pas celle en haut [Self-repair] “The groundhog isn’t the one on top?” (Lyster & Ranta, 1997, p. 65) Research into CO and IL Modification: Review and Assessment After well over a decade of research into Swain’s (1985) CO hypothesis, no definitive conclusions can be made, because the question of whether and how learners’ output, or output modification, helps with L2 learning is still largely unanswered. For example, Swain and Lapkin (1995) wrote that “no one has yet shown directly that these modified, or reprocessed, responses are maintained in the learner’s interlanguage” (p. 373). Similarly, Swain (1998) speculated that “it may be [emphasis added] that the modified, or reprocessed, output can be considered to represent the leading edge of a learner’s interlanguage” (p. 68). Indeed, in most cases, researchers assumed, rather than showed, that learners’ output, or output modification, contributes to language learning. For instance, Izumi and Bigelow (2000), in their comment on research on CO, pointed out that “even though student output is a prevalent feature of many language teaching practices, exactly whether and how it helps with language learning has often been assumed rather than vigorously tested” (p. 245). Lack of definitive conclusions is not surprising, because research on CO has been mostly cross-sectional in nature, focusing primarily on the production of MO per se rather than on whether and to what degree MO can be a source of linguistic competence. In other words, existing research focused on occurrence rather than acquisition. As will be shown below, much of the empirical research confined itself to examining the different learner and contextual factors that affect learners’ production of MO (e.g., gender difference, task type, and discourse variables, including signal encoding and signal source) rather than investigating whether and how MO can help with L2 learning. 602 Language Learning Vol. 52, No. 3 A number of studies considered the effect of gender and found that gender difference plays a role in the NNSs’ production of CO (e.g., Gass & Varonis, 1986; Pica et al., 1989; Pica, Holliday, Lewis, Berducci, & Newman, 1991; Shehadeh, 1994). These studies found that there is evidence from cross-gender conversations in the contexts of NS-NNS and NNS-NNS interaction to suggest that men and women tend to utilize conversation differently, such that men appear to take advantage of conversation in a way that allows them to produce a greater amount of CO, whereas women utilize conversation to obtain a greater amount of comprehensible input. For example, in a study specifically designed to investigate the effect of gender in NNS-NNS interaction, Gass and Varonis found differences between men and women in the amount each participated in conversation and in the control each had over the direction of the interaction: “Men took greater advantage of the opportunities to use the conversation in a way that allowed them to produce a greater amount of comprehensible output, whereas women utilized the conversation to obtain a greater amount of comprehensible input” (p. 349). Similarly, Pica et al. (1989) found that in male NNS–female NS dyads on a discussion task, there was more frequent male morphosyntactic modification than in female NNS–female NS dyads and proportionately more than the other two tasks employed in their study (information gap and jigsaw). These results imply that gender differences may also play a role in NS-NNS interaction, in particular on tasks whose successful completion involves the provision of optional information and in which both parties have equal opportunities to participate (such as a discussion task), rather than on those tasks whose successful completion involves the provision of specific information from each interlocutor, thereby creating a more restricted environment for turn allocations (such as information gap and jigsaw tasks). Shehadeh (1994) also found that men took greater advantage in a group activity (a mixed-sex task) of the opportunity to use the interaction in a way that allowed them to produce a greater amount of CO than women. But Shehadeh’s study also revealed Shehadeh 603 that same-sex dyads offered women comparatively greater opportunities to produce CO than men. More specifically, Shehadeh found that mixed-gender group interaction provided better contexts for males to request clarification, self-initiate repair, and produce CO than females, whereas same-gender dyadic interaction provided better contexts for females to self-initiate repair and produce CO than males. Research further found that task type can provide learners with varied opportunities toward MO (Iwashita, 1993, 1999; Pica, Lincoln-Porter, Paninos, & Linnell, 1996; Shehadeh, 1999a). Pica et al. found that the storytelling task employed in their study provided NNSs with higher percentages of modification of their output in both NS-NNS and NNS-NNS interactions than the house sequence task. Iwashita (1999) also found that one-way tasks provided learners with greater opportunities to modify their output toward comprehensibility than two-way tasks. Similarly, Shehadeh found that a picture description task (one-way task) provided significantly greater opportunities than an opinion exchange task (two-way task) toward CO (356 instances or 62% vs. 217 instances or 38%). Not only did the type of task play a role in opportunities for MO, but signal type was also found to influence learners’ production of MO significantly. Pica et al. (1989) found that the NS signal type had a significant impact on the type of response NNSs made to it, regardless of task. Specifically, Pica and her team found that across the three tasks employed in their study (information gap, jigsaw, and discussion), NNSs tended to modify their output most often when NSs signaled an explicit need for clarification (as in Example 3) rather than when they provided a model utterance for confirmation (as in Example 4). 3. NNS: NS: NNS: no a triangle, inside the triangle inside the triangle, there’s another triangle? above square, there’s a triangle, and that sign is inside the triangle 604 Language Learning 4. NNS: but I didn’t know how drew so we are very confused to draw? yeah (Pica et al., 1989, p. 89) NS: NNS: Vol. 52, No. 3 Contexts that involve learners’ production of MO in NS-NNS vs. NNS-NNS interaction were considered too in order to trace the effect of the type of speech partner (Pica et al., 1996; Shehadeh, 1999a). Research found that in most cases, NNSs produced comparable modified versions of their previous utterances when they negotiated with NSs or with each other. For instance, Pica et al. found that NS-NNS interaction and NNS-NNS interaction produced a comparable number of MO occurrences on both of the tasks employed in their study (a house sequence task and a storytelling task). However, Shehadeh (1999a) found that NNSNNS interaction produced more modified versions than NS-NNS interaction on a picture description task (91 occurrences or 63% vs. 54 occurrences or 37%) but not on an opinion exchange task (13 occurrences or 50% vs. 13 occurrences or 50%). Finally, the degree to which MO is brought to the learner’s attention by external feedback (i.e., other-initiation) or internal feedback/noticing (i.e., self-initiation) has been considered as well (Shehadeh, 1999a, 2001). The evidence suggests that the occurrence of MO can be largely determined by the source of initiation (other or self). For instance, Shehadeh (1999a) found that although NNSs encoded their own modification toward comprehensibility in response to both other- and self-initiations (as in Examples 5 and 6, respectively), NNS-based MOs resulting from self-initiation were significantly greater than those resulting from other-initiation. Specifically, Shehadeh found that NNSs produced an average of 2.5 MO occurrences per minute in response to self-initiation and 1 occurrence per minute in response to other-initiation. This means that instances of MO produced by NNSs resulting from self-initiation were two-and-a-half times more frequent than those resulting from other-initiation.1 Shehadeh 5. NNS1: NNS2: NNS1: NNS2: 6. NNS: 605 two small bottle two small what? bot (1.0) small bottles yeah yes because if the woman is (0.8) the wife always go out (0.6) goes out and left his his husband eh (1.0) her husband and her son in the home (0.7) at home it’s it’s not reasonable for for . . . [Note: ( . . . ) indicates pauses in seconds/fractions of a second.] (Shehadeh, 1999a, p. 645) In addition to examining the effect of gender difference, task type, type of interlocutor, and discourse variables like signal encoding and source, NNSs’ production of MO has been considered in other contexts too. The proficiency level of learners has been found to affect the occurrence of MO, such that intermediate proficiency level learners were found to produce more instances of MO than low proficiency level learners (Pica, 1988; Van den Branden, 1997; Shehadeh, 1999a). For instance, Pica and Van den Branden found that the low proficiency level NNSs who participated in their studies produced an average of four instances of MO per hour and an average of one instance every 5 min, respectively, whereas Shehadeh (1999a) found that the intermediate proficiency level NNSs who participated in his study produced an average of one MO instance per minute (I will return to this point below). Finally, opportunities for MO with younger learners have been investigated. Van den Branden (1997) found that 11- to 12-year-old pupils were able to modify their output interactionally when confronted with negative feedback, as in Example 7. 7. NNS pupil: NS pupil: NNS pupil: NS pupil: he was wearing a mass huh? what? what is he wearing? a mask oh I see (Van den Branden, 1997, p. 604) 606 Language Learning Vol. 52, No. 3 Overall, empirical research has shown that learners are given varied opportunities to modify their output toward comprehensibility. Research has also shown that a number of learner and contextual factors determine the frequency/occurrence of these opportunities. But can this research be considered acquisitional in nature? Does it enable us to make definitive conclusions as to whether and how these MO occurrences affect learners’ achievement in the L2? In the absence of any investigation that attempts to show that these modifications are actually retained in the learners’ IL or have an impact on SLA, the answer is, clearly, no. For example, observations pertaining to the effect of gender must remain preliminary, because only a small amount of focused research has been done to date into the effect of gender difference on L2 learning. It is not yet clear, for instance, whether and how much gender differences affect classroom interaction, progress, and final attainment in the L2. Nor is it quite clear to what degree task type or proficiency level affects learners’ achievement or can be a source of varied competence in the L2. Swain and Lapkin’s (1995) think-aloud protocols also focused on occurrence rather than acquisition. Swain and Lapkin conducted an introspection study in which they tried to shed more focused light on the processes and mechanisms that learners follow to reprocess and modify their IL utterances. In particular, they sought “to try to arrive at the mental processes . . . reflected in the changes students made to their output” (p. 381). The researchers examined the ability of 18 grade 8 immersion students learning French to consciously reprocess their IL output without any sort of external feedback when faced with a performance problem. The task given to the students was to write a report on some environmental problem. The students were instructed to think aloud while writing and especially when they were faced with a problem.Swain and Lapkin found that there were 190 occasions in which students encountered a linguistic problem (a gap) in their output. On each occasion, they forced themselves to modify their output toward comprehensibility or accuracy, as in Example 8. Shehadeh 607 8. Student 8: ‘I was gonna write les droits uhm d’animaux, but it doesn’t sound right so I said les droits des animaux (animal rights).’ (Swain & Lapkin, 1995, p. 381) Swain and Lapkin argued that the activity of producing the TL enabled the learners to notice a gap in their existing IL capacity. This noticing pushed them to reprocess their performance consciously in order to produce MO. They stated that in producing the L2, a learner will on occasion become aware of (i.e., notice) a linguistic problem (brought to his/her attention either by external feedback (e.g., clarification requests) or internal feedback). Noticing a problem “pushes” the learner to modify his/her output. In doing so, the learner may sometimes be forced into a more syntactic processing mode than might occur in comprehension. Thus, output may set “noticing” in train, triggering mental processes that lead to modified output. (pp. 372-373) Swain and Lapkin (1995) further argued that “on each occasion, the students engaged in mental processing that may have generated linguistic knowledge that is new for the learner, or consolidated existing knowledge” (p. 384). Consequently, Swain and Lapkin maintained that when learners reprocess and modify their output to make it more comprehensible, they are engaged in mental processes that are part of the process of language learning: “’pushing’ learners beyond their current performance level can lead to enhanced performance, a step which may represent the internalization of new linguistic knowledge, or the consolidation of existing knowledge” (p. 374). Again, it is not possible to make claims of acquisition or argue that this process of modification engaged learners in some restructuring of system that affected their access to the knowledge base or that this restructuring process was part of L2 learning, because no evidence was provided showing that student reformulations led to any acquisition of the L2. It is therefore difficult to argue, in the absence of investigation that captures acquisitional aspects, that the enhanced performance that learners achieved 608 Language Learning Vol. 52, No. 3 actually represents the internalization of new linguistic knowledge or the consolidation of existing knowledge or that it constitutes some sort of language development. Noticing and Beyond Researchers involved in the immersion programs went beyond noticing to explore how dialogue as a cognitive tool enables learners to notice gaps in their IL system and helps them in internalizing linguistic knowledge (e.g., Kowal & Swain, 1997; LaPierre, 1994; Swain, 1995, 1997, 1998). First of all, Swain (1995) proposed three functions of output in SLA: It promotes noticing, it serves the L2 learning process through hypothesis testing, and it serves as a metalinguistic function for language learners. Swain (1997, p. 119) first argued that each of these functions of output represents a cognitive activity: the cognitive activity of identifying knowledge gaps, the cognitive activity of generating and testing hypotheses, and the cognitive activity of solving problems. She further argued that dialogue, which she calls “collaborative dialogue,” is sometimes the source of these activities, because it creates a context that enables learners to identify knowledge gaps in their IL performance, to verbalize and explicitly test their hypotheses about the TL, and to solve linguistic problems jointly by negotiating about TL forms (i.e., metalinguistic talk; pp. 118–119). The evidence Swain (1997) provided in support of her arguments was obtained from dialogues based on a dictogloss task. In the task, students listened to a text read aloud twice to them at normal speed by their regular classroom teacher. During the reading, students were allowed to jot down familiar words and phrases. Following this they were instructed to collaboratively (in pairs) reconstruct the text they had just heard from their shared resources. The students were encouraged to reconstruct the text as closely to the original as possible with respect to both content and form. Shehadeh 609 Example 9 shows two students, George and Keith, jointly reconstructing the first sentence of the dictogloss task (En ce qui concerne l’ environnement, il y a beaucoup de problèmes qui nous tracassent [as far as the environment is concerned, there are many problems which worry us]). 9. 58 Keith: 59 60 61 62 63 George: Keith: George: Keith: George: 64 Keith: 65 George: 66 Keith: 67 Teacher: 68 Keith: 69 George: Attends une minute! Non, j’ai besoin du Bescherelle (verb reference book). S’il vous plaît ouvrir le Bescherelle à la page qui. OK à la dernière page (i.e., the index). OK, cherche tracasse, un page, deux pages. Tra, tra, tracer. Tracasser page six. Cherche le s’il vous pla_t. Pas de problème. c’est sur page. verbe, <à la page> six. OK, c’est le même que aimer (i.e., it is conjugated in the same way and aimer is given as the standard example for all verbs with this pattern of conjugation). Laissez-moi le voir s’il vous plaît (reading from the page). Le passé simple, nous tracasse; nous aime (Keith is trying to find a first person plural version of the verb which sounds like “tracasse” the word he has written in his notes, but is unable to find one). Peut-être c’est ici. Non, c’est juste nous aime (pause) ah, le présent. Tracasse, aimons, n’est-ce pas que tracasse (to teacher who has just arrived), ce n’est pas nous tracasse (what he has written down in his notes), c’est nous tracassons? Ce son des problèmes qui nous tracassent (deliberately not directly giving the answer). Nous tracassons. Oh (beginning to realize what is happening). 610 Language Learning 70 Keith: 71 George: Oui? (so what?). Les problèmes qui nous tracassent. Like the (pause) c’est les problèmes (pause) like, that concerns us. Oui, mais tracasse n’est-ce pas que c’est <o-n-z>? Tracasse. C’est pas un, c’est pas un, (pause), oui I dunno (unable to articulate what he has discovered). OK, ça dit, les problèmes qui nous tracassent. Donc, est-ce que tracasse est un verbe? Qu’on, qu’on doit conjuger? Uh huh. Donc est-ce que c’est tracassons? Ce sont les problèmes qui nous tracassent. Nous, c’est, c’est pas, c’est pas, oui, c’est les problèmes, c’est pas, c’est pas nous. Ah! E-n-t (in French –third person plural ending), OK, OK. [Note: ( . . . ) indicates editorial comments added by the authors and < . . . > indicates text added by the transcriber to aid comprehension.] (Swain, 1997, pp. 120–121) 72 Keith: 73 George: 74 Keith: 75 76 77 78 Teacher: Keith: Teacher: George: 79 Keith: Vol. 52, No. 3 Swain (1997, p. 121) explained that Keith jotted down nous tracasse in his notes, but this does not correspond with his knowledge of French that when nous is the subject of the verb, the ending of the verb must be -ons. In turns 58–65, the students are trying to find a first-person plural version of a verb that does not end with -ons in their Bescherelle reference book. But they fail because such a form does not exist in French. Therefore, Keith, in turn 66, appeals to the teacher for help (ce n’est pas nous tracasse, c’est nous tracassons? [you don’t say nous tracasse (what he has written down in his notes), shouldn’t it be nous tracassons?]). Keith has verbalized the problem, and he and his colleague can now work on solving it by engaging in an explicit hypothesis-testing activity. In turn 67, the teacher guides them by providing hints but Shehadeh 611 deliberately does not provide the correct answer. Turn 69 shows that George is beginning to understand how the words are related to one another (Oh). Turns 70–79 show that George, questioned by Keith and guided by the teacher, is trying to explain to Keith why tracasser should be in the third-person plural and not the first-person plural, as in turn 71 (like the . . . c’est les problemes . . . like, that concerns us) and turn 78 (Nous . . . c’est . . . c’est pas . . . oui . . . c’est les problemes . . . c’est pas . . . c’est pas nous [Us . . . it’s . . . it’s not . . . yeah . . . it’s the problems . . . it’s not . . . it’s not us]). This explanation provides Keith with the same understanding that George has had, and this enables Keith to write the verb with the correct ending (-ent). Swain (1997, p. 121) commented that this example shows that Keith’s construction of knowledge is a cognitive activity mediated by objects—the reference grammar, his interlocutor, and the teacher—through dialogue, arguing that this cognitive activity was constituted in dialogue and led to new linguistic knowledge. Describing the activity of the two students, Swain stated that “through their interaction, they established the problem and solved it. Furthermore, through dialogue, they reached a deeper understanding of language in context than what either of them could have done alone on their own” (p. 122). This example shows that collaborative dialogue necessitated output that enabled the students to notice a gap in their IL system. To overcome this gap, they verbalized the problem and jointly started to work on solving it by explicitly engaging in a hypothesis-testing activity and negotiating about the TL forms until a satisfactory resolution was reached. Thus, output triggered the search for solutions; without output, and in this case, without collaborative dialogue, the students might never have noticed their problems in their IL. We do not know, however, if the solutions reached during these dialogues play a role in internalizing and retaining linguistic knowledge or are facilitative of L2 learning without investigation that captures acquisitional aspects. It is not clear from these data 612 Language Learning Vol. 52, No. 3 alone that the function of output as a triggering process that encourages noticing of gaps, hypothesis formation and testing, and metalinguistic talk has any developmental effect on the learners’ IL system. From Occurrence to Acquisition: An Agenda for Acquisitional Research As shown above, there is a lack of acquisitional studies investigating the effect of output on language development. Consequently, it is not possible to make claims of acquisition or to argue that the processes of IL modification or collaborative dialogue have actually engaged learners in some restructuring of system or that these processes are part of L2 learning. The second step is to investigate whether and how output contributes to SLA or has any long- or short-term impact on IL development and L2 internalization. But how do we proceed from occurrence to acquisition? How do we investigate, for instance, whether the modifications learners make to their output can be a source of competence in the L2 or the solutions reached during collaborative dialogues are retained in the learners’ IL system? Two sets of research directions will be proposed below in an attempt to take the field forward in this regard. The first set relates to MO and includes examining the effect of MO on L2 learning, examining the direction toward which modifications are made (comprehensible, correct/accurate, or target-like), examining the specific type of linguistic modification NNSs produce (e.g., phonological, morphosyntactic, lexical) and the effect of frequency of MO on L2 learning. The second set relates to the function of output in L2 learning as a triggering process and includes output as a tool for metalinguistic talk, for noticing and focusing learners’ attention on subsequent input, for hypothesis testing, and for syntactic processing. Each of these research areas and directions will be considered separately below. Shehadeh 613 MO and L2 Learning Research has shown that while producing an L2, learners notice gaps in their output brought to their attention by external feedback (e.g., clarification requests) or internal feedback (internal noticing; e.g., Shehadeh, 2001; Swain & Lapkin, 1995). Research has also shown that a number of learner and contextual factors (e.g., gender difference, task type, signal type, signal source, and type of interlocutor) play a role in the learners’ reformulating and modifying their output toward comprehensibility in their attempt to fill these gaps (e.g., Gass & Varonis, 1986; Pica, 1988; Pica et al., 1989; Pica et al., 1991; Pica et al., 1996; Shehadeh, 1999a, 2001; Van den Branden, 1997). Can the modified or reprocessed output be considered to represent the leading edge of a learner’s IL, as Swain (1998, p. 68) speculated? In order to answer this question, research must first investigate if the modified or reprocessed output is retained in the learners’ IL system or has an effect on language development. More specifically, we need to investigate whether output2 in Figure 1 represents the internalization of new linguistic knowledge or the consolidation of existing knowledge or whether it constitutes some sort of language development. First of all, it must be acknowledged that the main difficulty facing scholars studying learner production in general and MO in particular is a methodological one. How do we construct tasks or posttests that investigate whether, for instance, the modified utterances are retained in the learner’s IL system? Indeed, the Figure 1. MO and L2 learning. Note. From “Problems in Output and the Cognitive Processes They Generate: A Step Towards Second Language Learning,” by M. Swain and S. Lapkin, 1995, Applied Linguistics, 16(3), p. 388. Copyright 1995 by Oxford University Press. Reprinted with permission. 614 Language Learning Vol. 52, No. 3 main problem with the currently employed methods of data collection such as the analysis of IL and the use of introspective protocols is that it has been difficult to ascertain whether and how output modifications are actually retained in the learner’s IL system or help with language learning. It is acknowledged, in fact, that tasks that aim at learner production, rather than comprehension, are difficult to construct (Loschky & Bley-Vroman, 1990; Nobuyoshi & Ellis, 1993; Swain & Lapkin, 1998). This is because it is the learner, not the researcher, who has control over the linguistic content, making it difficult to control learner output (Kowal & Swain, 1997; Loschky & Bley-Vroman, 1990; Nobuyoshi & Ellis, 1993). For example, Nobuyoshi and Ellis pointed out that the inherent redundancy of language and the availability of rich contextual clues in many tasks obviate the need for learners to use any particular grammatical structure. For this reason, most production tasks are focused only to the extent that a particular structure is “useful” or “natural” and, as a consequence, may not actually result in its use. (p. 205) The main question therefore is: How do we construct methodologically focused tasks that “push” learner production? Further, in order to be able to make acquisitional claims, we need to see specific or relevant structures appear in the learner’s output, correctly or incorrectly. How do we construct tasks (as tests and posttests) that force learners to use a particular aspect of language so that then we, the researchers, can argue that it has been acquired? How do we get learners to produce a relevant grammatical structure in a subsequent treatment? Nobuyoshi and Ellis (1993, p. 205) suggested that one way is to construct methodologically focused communication tasks that involve grammatical knowledge in various ways and to varying degrees, tasks in which the use of the specific grammatical structure under examination is “essential” (i.e., its use is required by the task). Such controlled situations, specifically contrived to produce specific grammatical structures such as simple past tense, Shehadeh 615 present perfect, future forms, relative clauses, and conditionals, may enable researchers to examine the acquisition of these grammatical structures and, consequently, to make stronger claims about language development. Employing methodologically focused tasks, Nobuyoshi and Ellis (1993) investigated whether “pushing” learners toward greater accuracy in their production leads to more accurate output and whether this contributes to acquisition. Specifically, Nobuyoshi and Ellis (p. 206) sought to answer the following two research questions: 1. Does “pushing” learners by means of requests for clarification result in more accurate use of past-tense verb forms in communication? 2. Do learners continue to show improved accuracy in the use of past-tense verb forms in subsequent communication when there is no attempt to “push” them? The researchers collected data from 6 Japanese low-level learners of English (3 comprised the experimental group and 3 the comparison group). The participants performed two picture jigsaw tasks. The tasks were constructed to involve the use of simple past tense (describing events that happened the previous week [for task 1] and the day before at the office [for task 2]). All 6 participants performed the two tasks twice. There was a 1-week interval between the two treatment phases for both groups. In treatment phase 1, the experimental group received requests for clarification every time they produced an utterance in which the verb was not, but needed to have been, in the simple past tense (obligatory occasion); when the verb was incorrectly formed; or when the teacher genuinely failed to understand what the students had said. In treatment phase 2, students received general requests for clarification only when the teacher genuinely did not understand what the students had said. To ensure that the participants did not practice performing the tasks in the intervening week, they were not told that they would repeat them. The participants in the 616 Language Learning Vol. 52, No. 3 control group received only general requests for clarification, none of which followed an incorrect use of past tense by students, in both treatment phases. The researchers found that all 6 learners produced a substantial number of errors in the first administration. In the case of the experimental group, 2 of the learners showed significant gains in accuracy. Clarification requests led these 2 learners to reformulate their output in a way that corrected their past-tense errors. In other words, when the teacher pushed these 2 learners in the direction of greater accuracy in their production, they were able not only to make self-repair but also to achieve a higher accuracy level in their output, as in Example 10. 10. Learner: Teacher: Learner: last weekend, a man painting, painting “Beware of the dog” sorry? a man painted, painted, painted on the wall “Beware of the dog” (Nobuyoshi & Ellis, 1993, p. 205) The results further revealed that the improved accuracy of these 2 learners resulted in improved performance, both immediate and over time. Both learners improved on their initial level of accuracy: Learner 1 moved from 31% to 89% and learner 2 from 45% to 62%. Nobuyoshi and Ellis (1993) commented that the two experimental learners who had successfully reformulated their utterances to increase the use of correct past tense verb forms during the first administration of the task sustained the gain in accuracy during the second administration, even though on this occasion the teacher made no attempt to “push” them into correct use. (p. 208) The third learner, however, showed no overall gain in accuracy; neither did any of the learners in the comparison group. Nobuyoshi and Ellis (pp. 208–209) argued that the third learner did not correct his errors perhaps because he was more concerned with communication and general fluency than with accuracy (pp. 208–209). The researchers suggested that one possible interpretation of the Shehadeh 617 results is that “pushing” learners to make their output more comprehensible leads to linguistic development in some learners, but not others (for a more detailed discussion, see Nobuyoshi & Ellis, 1993, pp. 208–209). Nobuyoshi and Ellis (1993, p. 209) acknowledged that their study can only be considered exploratory because it was based on a very small number of participants. Nevertheless, this study can be considered an empirical baseline for broader and larger investigations. It can be replicated, for instance, by research undertaking a larger sample size and examining different linguistic structures. Such research may enable us to shed more focused light on, and verify the validity of, Swain and Lapkin’s (1995) assertion that “what goes on between the first output and the second .... is part of the process of second language learning” (p. 386). Comprehensible, Correct, or Target-Like MO and L2 Learning Part of investigating the effect of MO on language learning involves examining the direction toward which NNSs make their modifications (comprehensible, correct/accurate, or target-like), because this may reveal how and in what way MO is related to SLA. Most studies on MO have maintained that the adjustments that NNSs make to their IL will be achieved through their attempts to provide more accurate/correct and more CO (see, for example, Pica et al., 1989; Pica et al., 1996; Shehadeh, 1999a; Van den Branden, 1997). It has been argued that this is because the NNSs’ attempts to modify and adjust their output (whether to make it more accurate/correct or more comprehensible) will expand their IL capacity (Pica et al., 1996) on the assumption that “‘pushing’ learners beyond their current performance level can lead to enhanced performance, a step which may represent the internalization of new linguistic knowledge, or the consolidation of existing knowledge” (Swain & Lapkin, 1995, p. 374). However, it must be acknowledged that CO may not necessarily always be correct/accurate and that correct/accurate output 618 Language Learning Vol. 52, No. 3 may not always be comprehensible. For instance, an NNS utterance like “the husband always go out and left his wife in the home” can be comprehensible, though incorrect in many ways, whereas an NNS utterance like “I’m married and I claim three children” can be considered correct but not totally comprehensible or nativelike (from the British-North American perspective). Apart from Pica (1988), who found that the overwhelming majority (91%) of NNSs’ self-initiated modifications in response to NS signals of nonunderstanding were encoded in more target-like form, no attempt has been made to describe or examine the specific type of modification (comprehensible, correct/accurate, or targetlike) NNSs produce. Pica examined negotiated interactions between an NS and 10 NNSs of English to find out how the NNSs made their IL utterances comprehensible when the NS indicated difficulty in understanding them. She found that NNSs generated their own modifications of initial trigger utterances only 48% of the time, but when they did so, consistently (91%) they showed target-like use of English, as in Example 11. 11. NNS: NS: NNS: I gotta go ten month huh? ten months (Pica, 1988, p. 55) Pica concluded that the results of her study showed that NNSs can modify their IL utterances in response to an NS signal to achieve output that is both more comprehensible and more targetlike: more comprehensible because . . . during 95% of the interactions in which the NS signaled comprehension difficulty, the NNS response to the signal led to a successful resolution; and more target-like because 48% of the NNSs’ total number of responses to the NS signal were encoded, through their own initiation, in more target-like form, a figure which represents 91% of the NNSs’ self-initiated modification. (pp. 59–60) Shehadeh 619 Examining the direction toward which NNSs make their modifications has been overlooked almost completely by other researchers, perhaps because of the difficulty involved in such investigation. Such investigation is difficult partly because interlocutors’ reactions are sometimes unreliable indicators of comprehension or noncomprehension (Aston, 1986). Interlocutors do not always react to a trouble source that exhibits an incorrect/inaccurate or less target-like form unless it impedes comprehension. Even when a trouble source might impede comprehension, interlocutors do not always react. For example, Aston pointed out that sometimes interlocutors feign comprehension in order to keep the conversation going, reaffirm satisfactory communication, and maintain a satisfying rapport (p. 139). In cases in which the trouble source has been ignored, therefore, there is often no way for the investigator to recognize that there has been a breakdown in comprehension or communication, although something later in the discourse may indicate that in fact the listener did not understand (Aston, 1986) or that the speaker did run into difficulty but did not initiate repair (Hawkins, 1985; Varonis & Gass, 1985). Nevertheless, any attempt to isolate and examine the proportion of comprehensible but incorrect/less accurate output versus correct/more accurate but less comprehensible output and the proportion of output that is more target-like versus output that is less target-like is a worthy task for future research in which a more detailed, longitudinal analysis of NNSs’ MO is carried out and which may reveal whether learning has occurred and in what way the specific type of modification (comprehensible, correct/accurate, target-like) affects language development. By the same token, researchers may also contemplate investigating the type of trouble source interlocutors react to: Does it have to do with correctness/accuracy, comprehensibility, native-likeness, or a combination of these and other factors? 620 Language Learning Vol. 52, No. 3 Type of Linguistic Modification and L2 Learning Not only does the direction toward which NNSs make their modifications merit further investigation, but research on the effect of the specific type of linguistic modification they produce (e.g., phonological, morphosyntactic, lexical) on L2 learning is also a desideratum. Apart from Pica (1988) and Pica et al. (1989), the specific type of linguistic modification NNSs produce has received hardly any attention. Pica found that 50% of the total number of linguistic modifications learners produced were semantic, 31% were morphosyntactic, 11% were phonological, and 8% were L2 translation. Pica et al., who analyzed their data in terms of two categories only, found that 48% of the total number of linguistic modifications were semantic and 52% were morphosyntactic. However, because no posttests were conducted in either study, we do not know if the differences in the proportions of the specific type of linguistic modification had any (significant) effect on language achievement. It is important, therefore, that modifications be analyzed in terms of the specific type of linguistic category to show what portions of the modified versions are phonological, morphosyntactic, or lexical in nature, because this may be important in showing in what way MO aids L2 learning. Ideally, researchers must take into consideration, as part of their research design, learnerspecific tests that investigate where most modifications occur, why they occur, and their consequences on SLA, all rolled into one study. If, for example, future research were to confirm that most of the modifications NNSs produce were morphosyntactic, one could conduct a relevant posttest to find out to what degree this affected learners’ achievement in the L2 on this particular category. By the same token, it is important to know whether NNSs make modifications to their IL phonology, morphosyntax, and lexical choice when interacting with other NNSs in the same way as they have been claimed to do when interacting with NSs or whether they make such modifications in a systematic but less comprehensible Shehadeh 621 or target-like way, for instance (for a similar argument, see Pica, 1994, p. 518). Frequency of Output Modifications and L2 Learning Does the frequency of output modifications play a role in SLA? Is CO frequent enough to have an impact on language development? Some researchers doubt that CO is frequent enough to be a source of language competence. Krashen (1994, 1998), in particular, has argued that CO is too infrequent to amount to a significant source of language competence. He stated that “a problem all output hypotheses have is that output is surprisingly rare. . . . In the case of comprehensible output, the problem is especially severe” (1998, p. 175). The evidence Krashen cited in support of his argument was obtained from Pica’s (1988) study and Van den Branden’s (1997) study. Specifically, Krashen (1998, pp. 175-176) pointed out that NNSs produced an average of four instances of MO per hour in Pica’s study and an average of one instance every 5 min in Van den Branden’s study. It is not always the case, however, that MO is produced so infrequently. Other studies have observed significantly more frequent occurrences of MO than those found in Pica’s (1988) and Van den Branden’s (1997) studies (e.g., Iwashita, 2001; Shehadeh, 1999a). Iwashita and Shehadeh found that NNSs produced an average of two MO instances and one MO instance per minute, respectively. The differences between Iwashita’s and Shehadeh’s findings and Pica’s and Van den Branden’s findings may be partly accounted for in that Iwashita and Shehadeh collected data from intermediate proficiency level NNSs who were more able to respond successfully to interlocutors’ clarification requests, statements of nonunderstanding, and requests for reformulation, explanation, expansion, paraphrase, or elaboration than the low proficiency level NNSs who participated in Pica’s and Van den Branden’s studies. This gives the former considerably more opportunities to modify their IL utterances toward comprehensibility than the latter. It is possible to argue, therefore, that the relative 622 Language Learning Vol. 52, No. 3 infrequency of MO in Pica’s and Van den Branden’s studies was partly due to the low proficiency level of the NNSs who participated in these studies. In addition to the effect of level of proficiency, the frequency/occurrence of MO is determined by a host of other learner and contextual factors, including task type, gender, type of learner, and discourse variables like signal type and signal source (see earlier review of empirical research; see Nobuyoshi & Ellis, 1993, for types of learner factors). For instance, Shehadeh (1999a) found that a picture description task provided NNSs with significantly greater opportunities to modify their output toward comprehensibility than an opinion exchange task (NNSs produced an average of 2.2 instances of MO per minute on picture description and an average of 1.3 MO instances per minute on opinion exchange). In spite of the fact that MO may be infrequent in some contexts, there is no evidence, at present, to suggest that quantity (rather than quality or a number of learner and contextual factors) is what matters most. In fact, notions of “critical incidents” would suggest that although MO may be rare in some contexts (which is arguable, but nevertheless claimed by Krashen, 1994, 1998), it can be useful when it does appear. For instance, Tarone and Liu (1995, p. 118) found that the contexts in which the learner stretches his or her competence in the TL to its limits are the contexts in which the IL develops more rapidly. They argued that the data examined in their study showed that “it is precisely in those contexts where Bob [the subject of their longitudinal case study] has to produce output which his IL cannot handle that the IL develops faster, with the richest variety of IL utterances and even possibly with structures out of developmental sequence” (p. 120). Nevertheless, whether and how the frequency/occurrence of output modifications can be a source of competence in the L2 remains a venue for further research that examines such matters in more detail. We would need to know, for instance, whether the frequency of modifications affects the rate and/or the route of acquisition, causing the competence of the learner to develop differently (Tarone & Liu, 1995). A typical study would investigate, Shehadeh 623 for example, whether those tasks that provide significantly more opportunities for output modification (as in the case of picture description) are more conducive to language learning than other tasks that provide fewer opportunities (as in the case of opinion exchange) or whether it is the quality/type—rather than the number—of modification instances that really matters. (See Pica, Kanagy, & Falodun, 1993, for a typology of communication tasks and ways in which different tasks can be linked to learners’ output and IL modification.) By the same token, if, for example, future research found that learners’ output modifications leading to comprehensibility were integral to successful L2 (as suggested by Swain, 1995, 1998; Swain & Lapkin, 1995) and that the frequency of MO was an important source of language competence (as argued by Krashen, 1994, 1998), it would then not be just other-initiations that matter for L2 learning, but rather self-initiations, as these have been found to be more frequent than other-initiations (see Shehadeh, 1999a, 2001). For instance, Shehadeh (1999a), as reported earlier, found that the NNSs who participated in his study produced an average of 1 MO instance per minute in response to otherinitiation and 2.5 in response to self-initiation. In other words, instances of MO resulting from self-initiation were two-and-a-half times more frequent than those resulting from other-initiation. The implication of such research for classroom interaction is that learners need to be given both time and opportunity to achieve self-initiated, self-completed repair of their messages. This is important when we consider that some classroom studies have observed that students were not given sufficient time or opportunity to self-correct in a classroom situation (e.g., McHoul, 1978, 1990). For instance, McHoul (1990) observed that teachers initiated corrections “either (a) immediately a trouble-source is over, with usually no gap occurring or (b) immediately the repairable [i.e., the trouble-source] itself is spoken/heard” (p. 375). He goes on to say that “the latter cases of other-initiations either (i) overlap the trouble-source turn or (ii) interrupt it. In instances of (i), teacher and student can both be heard to be speaking, albeit 624 Language Learning Vol. 52, No. 3 briefly, at the same time. In instances of (ii), the student immediately yields the floor to the teacher” (p. 375). The research areas proposed so far relate to investigating the effect of MO on L2 learning. However, research has also shown that output has a triggering function in L2 learning. The research areas to be proposed below therefore relate to exploring research directions that might involve acquisitional aspects of the role of output in L2 learning as a triggering process. These include output as a tool for (a) metalinguistic talk; (b) noticing of, and focusing learners’ attention on, subsequent input; (c) hypothesis testing; and (d) syntactic processing. Output as a Tool for Metalinguistic Talk and L2 Learning Immersion program studies have shown that learners notice problems in their output, prompting them to use their internal resources by themselves or in collaboration to solve these problems (e.g., Kowal & Swain, 1997; LaPierre, 1994; Swain, 1997; Swain & Lapkin, 1998). Specifically, these studies have shown that verbalization helps learners to solve linguistic problems through reflection on them, or metalinguistic talk. For example, in Swain’s (1997) study, production enabled learners to notice problems in their IL system, prompting them to reflect consciously on the language they were producing and to negotiate collaboratively about TL forms and structures until a satisfactory resolution was reached. The next step is to investigate whether the solutions reached during collaborative dialogues are retained in students’ IL. Some studies have actually begun to address this issue (Swain, 1998, 2000; Swain & Lapkin, 1998). Swain (1998) cited evidence from previous studies that used language-related episodes (LREs)2 as an analytic tool (Kowal & Swain, 1994, 1997; LaPierre, 1994) to argue that the solutions reached during the dialogues were actually retained in the students’ IL. In particular, Swain discussed the study conducted by LaPierre (1994), who collected data from 48 students from two grade 8 classes of an early French immersion Shehadeh 625 program. The 48 students were divided into two groups: a metalinguistic (M) group that consisted of 26 students and a comparison (C) group that consisted of 22 students. A dictogloss passage was read aloud twice to the students. In the first reading, students only listened to the passage. In the second reading, they were encouraged to take notes of familiar words and phrases to help them reconstruct the passage. Following this, students worked in pairs for about 25 min to reconstruct the passage as closely to the original as possible. Data were collected in session 3, sessions 1 and 2 being the modeling and practicing sessions, respectively. The difference in conditions between the two groups involved what the teacher and the researcher said to each other as they reconstructed the text during the modeling session. The metalinguistic talk, or “metatalk” in Swain’s (1998, p. 68) words, that was modeled for the M group included the provision of rules and metalinguistic terminology. The goal was to give the students a way of seeing how to deploy explicit linguistic knowledge to solve a linguistic problem caused by a “hole” in their IL. The metatalk that was modeled for the C group, on the other hand, did not make use of rules or metalinguistic terminology. The goal here was just to draw students’ attention to grammatical form without invoking explicit rules. In this sense, the students were not provided with a demonstration of how to solve an encountered linguistic problem. The results showed that there was a total of 256 LREs in the two groups combined: 140 (54.7%) were of type 1 (problem solved correctly); 50 (19.5%) were of type 2 (problem not solved or disagreement about problem solution); 21 (8.2%) were of type 3 (problem solved incorrectly or disagreement about problem solution); and 45 (17.6%) were of type 4 (other). The results also showed that the M group produced an average of 14.8 LREs and the C group an average of 5.8 LREs. In other words, the M group produced approximately two-and-a-half times as many LREs as the C group. Swain (1998) argued that these differences suggest that “metatalk that included the explicit statement of rules and the use of metalinguistic terminology succeeded to a greater 626 Language Learning Vol. 52, No. 3 extent in capturing students’ attention and focusing it on their own language use” (p. 77). Example 12 is an LRE in which the problem is solved correctly (type 1). 12. S1: S2: S1: S2: S1: S2: S1: S2: j’ai fait un rêve effrayant la nuit dernière (I had a frightening dream last night) la nuit dernière (last night) puis je sais le debut de la seconde phrase (and I know the beginning of the second sentence) attends. Attends. Attends. Il y a quelque chose de mal avec cette phrase. Est-ce que c’est “une rêve” ou “un rêve”? (wait. Wait. Wait. There is something wrong with this sentence. Is it “dream” [feminine] or “dream” [masculine]?) je pense que c’est “un rêve” (I think it’s “dream” [masculine]) le rêve, la rêve, le rêve? [testing whether dream is masculine or feminine; seeing which sounds better] on va le laisser comma _a (we’re going to leave it like that) j’ai fait un rêve .... OK (I had a dream [masculine] ....OK) (Swain, 1998, p. 76) To determine if the solutions reached during the dialogue were retained in the students’ IL, a dyad-specific posttest was administered 1 week later. The results revealed that there was a strong tendency for students to “stick with” the knowledge they had constructed collaboratively the previous week. Students’ responses on the posttest showed a 70–80% correspondence with the solutions—right or wrong—that they arrived at in their dialogues. Swain (1998, 2000) interpreted these results as a strong indicator that the students’ dialogues mediated the Shehadeh 627 construction of linguistic knowledge. She argued that these results show that there is a relationship between metalinguistic talk and L2 learning such that metalinguistic talk supports L2 learning and that it is evidence of learning at work. Swain (1998) stated that “these results suggest rather forcefully that these LREs, during which students reflect consciously on the language they are producing, may be a source of language learning” (p. 79). In another study, Swain and Lapkin (1998) found that a preand posttest design as a research tool did not work very well with LREs. Specifically, they found that it was very difficult to predict what dyads would talk about, because each pair in their study focused on different aspects of the language and did so in a different way, even though all pairs were given the same task and belonged to the same class level and the same school. As a result, Swain and Lapkin attempted an acquisition measure via individualized posttest items based on previous LREs. In a relatively small number of instances in which an LRE related a pretest to a posttest item, Swain and Lapkin were able to demonstrate that the LRE was an occasion for L2 learning. More finely tuned designs must therefore be followed that may reveal to what degree dialogue can be an enactment of mental processes and an occasion for L2 learning. Building on Swain and Lapkin’s (1998) study, which revealed that the same task does not provide similar occasions for L2 learning to all student dyads, one way of doing this is to attempt to trace language learning specific to the dialogue of individual pairs and measure acquisition by tailoring items on the posttest. Another way suggested by Swain and Lapkin is “to combine an analysis of students’ collaborative dialogue with follow-up interviews in order to derive a more fine-grained understanding of the mental processes” enacted in these dialogues (p. 333). Such interviews also enable us to find out what the individual students found appealing or unappealing, conducive or unconducive to learning. Researchers may also contemplate constructing methodologically focused tasks (Nobuyoshi & Ellis, 1993) that examine specific grammatical knowledge or discourse aspects and, at the same time, combining multiple test 628 Language Learning Vol. 52, No. 3 measurements (e.g., pre-/posttests as well as retrospective interviews or introspective analyses). These may better trace the longterm impact of learners’ reflection on their own language use, or metalinguistic talk, on L2 learning. Output as a Tool for Noticing of, and Focusing Learners’ Attention on, Subsequent Input and L2 Learning Does output trigger cognitive processes that affect noticing of, and focusing learners’ attention on, subsequent input and SLA? In a series of studies, Izumi and his colleagues (Izumi, 2000, in press; Izumi & Bigelow, 2000; Izumi, Bigelow, Fujiwara, & Fearnow, 1999) attempted to answer this question by investigating whether output would alter learners’ subsequent input processing and promote their IL development. Specifically, these researchers sought to use output to enhance the noticing and learning of specific grammatical forms if input containing these forms was subsequently provided to learners. Focusing on the English past hypothetical conditional, Izumi and Bigelow (2000) and Izumi et al. (1999) explored this issue by comparing the performance of two groups of learners: One group was given output opportunities and subsequent exposure to relevant input, and the other group received the same input for the sole purpose of comprehension. Both studies employed the same treatment procedure and types of tasks—a text reconstruction task and a guided essay-writing task—which were delivered in reverse orders in the two studies. The results showed a significant improvement on the target form only after the second phase of the treatment for both types of tasks in both studies, which suggests the need for extended opportunities for producing output if it is to have a real effect on L2 learning. For example, in their 1999 study, Izumi and his colleagues addressed the question of whether learners’ recognition of linguistic problems in their output during production would prompt them to seek subsequent input with more focused attention and Shehadeh 629 whether this would lead to the noticing and learning of specific grammatical forms. Izumi et al. stated that the question remains: Does learners’ recognition of linguistic problems prompt them to notice relevant features if input is subsequently provided to them? The current study attempts to investigate this question by providing learners with opportunities for output, which are then followed by opportunities to receive relevant input, to see whether they would notice and learn the targeted feature in the input. (p. 425) The researchers collected data from 22 participants: 11 were assigned as an experimental group and 11 as a comparison group. The researchers compared these two groups with regard to their learning of the past hypothetical conditional in English. They followed two treatment phases for data collection. In treatment phase 1, the experimental-group participants reconstructed a reading passage as accurately as possible after being exposed to it. This was followed by a second exposure to the same input material and a second reconstruction opportunity. In treatment phase 2, the experimental-group participants wrote on specific topics followed by the presentation of a model essay written by an NS. Following this, participants wrote a second time on the same topic. The comparison-group participants were exposed to the same input, but instead of reconstructing a text or writing an essay, they answered comprehension questions only. Izumi et al. (1999) found that although phase 1 tasks resulted in noticing and immediate incorporation of the target form, performance on the posttest failed to reveal any effects. The phase 2 tasks, in contrast, resulted in improvement on the posttest. In a more controlled study, Izumi (2000, in press) explored how output may interact or contrast with other factors in promoting SLA. Specifically, Izumi investigated whether output and input enhancement, in isolation or in combination, promote noticing and L2 learning. He collected data from five groups with a pretest-posttest design: four treatment groups and one control group. The treatment groups differed with respect to their 630 Language Learning Vol. 52, No. 3 requirement of output (denoted as ±O) and exposure to enhanced input (denoted as ±IE). The first group was required to produce output and received enhanced input (+O+IE). The second group was required to produce output and was exposed to regular, unenhanced input (+O–IE). The third group received enhanced input without output (–O+IE), and the fourth group received unenhanced input without any output requirement (–O–IE). The control group participated in the pre- and posttests only. Sixty-one intermediate proficiency level students participated in the study, divided into the five groups as follows: 11 for the +O+IE group, 12 for the +O–IE group, 12 for the –O+IE group, 12 for the –O–IE group, and 14 for the control group. The study was conducted over a period of approximately 25 days, including the pretest and the posttest, divided as follows: 7.13 days between the pretest and the commencement of the treatment, 2 weeks for the treatment sessions, and 3.43 days between the last day of the treatment and the posttest. The target form the researcher examined was English relative clauses as acquired by adult L2 learners of English. The task employed for collecting data was a modified version of the text reconstruction task used in previous studies (Izumi & Bigelow, 2000; Izumi et al., 1999). The task was specially designed to provide learners with more extended opportunities for producing output and receiving relevant input to ensure maximal benefit obtainable from the output-input treatment. The task required that students (a) read and understand a text; (b) reconstruct the text as accurately as possible (for the output groups) or answer questions about the text (for the nonoutput groups); and (c) demonstrate comprehension by writing a recall summary in their first language. Izumi (2000, in press) found that those engaged in the output-input treatment outperformed those exposed to the same input for the sole purpose of comprehension (without any output) in noticing and learning the target form. On the other hand, visual input enhancement failed to show any measurable Shehadeh 631 effect on learning in spite of the positive effect it had on noticing the target form items in the input. Izumi (in press) interpreted these findings as demonstrating that output, as an internally driven mechanism in which attention arises through production processes, works better for promoting noticing and L2 learning than input enhancement, a mechanism in which attention is induced by external means. Further, unlike enhanced input, output, by virtue of producing utterances, can create a favorable condition for learners to make a cognitive comparison between their IL form and the TL form, leading them to expunge the non-target-like form from their developing IL in favor of the TL form. Izumi commented that the contrasting results of output and input enhancement in this study may be explained by positing that input enhancement was not sufficient to induce noticing of the IL-TL mismatches. Output, in contrast, promoted both the processes of noticing the form and noticing the mismatches, which enabled the output learners to attain successful and superior learning of the form. With regard to the specific role of learner output in noticing and L2 learning, Izumi (in press) argued that his study revealed a threefold facilitative effect of pushed output on SLA: (a) it promotes detection of formal elements in the input, (b) it promotes integrative processing (Graf, 1994) of the target structure, and (c) it promotes noticing of the mismatches between the learner’s IL form and the TL input. Izumi concluded that the findings of his study showed that “pushed output can induce learners to process the input effectively for their greater IL development.” Because the time span of these studies was rather short, it is not possible to argue that Izumi and his colleagues have completely tapped the long-term impact of output on noticing and L2 learning. For example, Izumi acknowledged that his study may not be considered to have revealed the long-term impact of output on L2 learning, suggesting that “long-term effects of the output-input treatment need to be examined.” More long-term studies are therefore needed into the role of output in L2 learning as a means 632 Language Learning Vol. 52, No. 3 of enhancing learners’ noticing of, and focusing learners’ attention on, subsequent input. Future research must also examine other grammatical aspects (e.g., modality, tense). Will the same results be obtained if another grammatical aspect is examined? Further, do we get the same result if another input enhancement procedure or instructional technique is used? Finally, Izumi used the written mode for his text reconstruction task. Do we get the same result if an oral/aural mode is used? The dominance of oral language use in naturalistic and classroom contexts makes tackling this issue a priority (see also Izumi, in press). The findings of Izumi and his colleagues’ studies, however, and the research methods they employed remain a baseline for future research that attempts to trace the long-term impact of learner output on noticing and L2 learning.3 Output as a Tool for Hypothesis Testing and Internalizing Linguistic Knowledge The triggering process of output must be explored not just with respect to the role of hypothesis formation and testing as a joint activity in collaborative dialogues (see Swain, 1997, 1998; Swain & Lapkin, 1998), but also with respect to the role of hypothesis testing in internalizing linguistic knowledge without partners’ help or any other external feedback. In other words, the question “What role does hypothesis testing by the learner play in internalizing linguistic knowledge in the absence of external feedback?” is still unanswered. This question is quite important in the light of wellestablished observations by IL studies to suggest that learners form hypotheses about the structural properties of the TL on the basis of the input data to which they are exposed (e.g., Corder, 1981; McLaughlin, 1987, 1990; Schachter, 1984; Selinker, 1972). Ellis (1994, p. 352) argued that these hypotheses enable learners to form a “hypothetical grammar” that can then be tested receptively and productively. Hypotheses are tested receptively by means of listening and reading and productively by means of Shehadeh 633 speaking and writing. Listening and reading provide more input against which existing hypotheses are tested. The new input provided enables the learner to confirm, reject, or modify his or her existing hypotheses about the TL accordingly. Other hypotheses are tested productively. Production involves a series of learner-generated hypotheses and assumptions about the L2 (Swain, 1995, 1997). It represents the learner’s best guess about how the TL works. When the learner’s best guesses (hypotheses) are tested, they are either confirmed or rejected/disconfirmed. Hypotheses are confirmed if the learner successfully transmits the intended meaning of the message successfully. Hypotheses are disconfirmed (rejected) and revised if the learner’s output fails to communicate the intended meaning of the learner’s message successfully (Ellis, 1994, p. 352). Incomprehensible or incorrect/inaccurate output generates external feedback (e.g., explicit corrections, clarification requests, recasts). In some cases (e.g., explicit corrections and recasts), external feedback enables the learner to replace the incorrect hypotheses about the TL structures and rules with the correct ones (e.g., Lyster, 1998a, 1998b; Lyster & Ranta, 1997; Oliver, 1995; White, 1991). White, for instance, argued that negative evidence triggers the resetting of parameters to their L2 values. In such cases, hypotheses are revised in the light of the new input provided. In other cases (e.g., clarification requests), external feedback forces the learner to modify his or her IL utterances toward comprehensibility or accuracy. In these cases, learner hypotheses generate requests for clarification, reformulation, expansion, etc., that force the learner to restructure and modify his or her IL performance toward more comprehensible or more accurate output (Swain, 1998; Swain & Lapkin, 1995; see also the empirical research on CO and IL modification reviewed earlier). Unlike these contexts in which feedback is available, very little is known about the effect of hypothesis testing by the learner on L2 learning in the absence of partners’ or external feedback. We do not know yet whether and how learner-generated 634 Language Learning Vol. 52, No. 3 hypotheses play a role in internalizing linguistic knowledge in the absence of external feedback. In a preliminary investigation, Shehadeh (1999b) sought to explore how output and hypothesis testing by the learner might be implicated in internalizing linguistic knowledge in the absence of partners’ or external feedback. Shehadeh collected data from 16 participants, 8 NSs and 8 NNSs of English, forming 8 NS-NNS dyads. A picture description task was used to collect data. In carrying out the task, an NNS had to describe a picture to an NS partner, who had to reproduce the picture as precisely as possible solely on the basis of the NNS’s description. Successful completion of the task depended in the first place on the NNS’s ability to supply a clear and accurate description of the picture he or she was holding. Dyads sat back to back to prevent any paralinguistic or nonlinguistic interference. Dyads took 6–9 min to complete the task. The data were specifically analyzed and examined for hypothesis-testing episodes (HTEs) by NNSs. An HTE was defined as any utterance or part of an utterance in which the learner externalizes and explicitly experiments with his or her hypotheses about the TL by (a) verbalizing these hypotheses to test which sounds better or (b) explicitly testing hypotheses against the competence of the (NS) interlocutor by means of (1) requesting confirmation or (2) appealing for help. The study found that there was a total of 39 HTEs by NNSs across all eight dyadic interactions, with an average of 4.875 HTEs per NNS. More importantly, the study found that the overwhelming majority (34 cases or 87%) of the learners’ externalized and explicitly tested hypotheses about the TL went completely unchallenged, regardless of whether these HTEs resulted in grammatically well-formed or ill-formed utterances. This means that in only five cases, or 13% of the learners’ HTEs, did NS interlocutors provide feedback on the NNSs’ output. Furthermore, these were the occasions on which the NNSs themselves requested confirmation or appealed for help. In other words, the study revealed that unless the learner requested confirmation or appealed for help, Shehadeh 635 the NS interlocutor provided no feedback on the outcome of the learner’s HTE, whether it was grammatically well-formed or illformed. Example 13 is an instance of the outcome of an HTE that went unchallenged by the NS interlocutor, even though the utterance was ungrammatical. 13. NNS: NS: you have two chairs (0.8) one near of the bed (0.8) near of the bed (0.9) near to the bed (1.0) and the other near to the bottom left corner (2.0) and behind the chair (1.0) you have a door (3.0) now near to the door and go to the left corner yes (0.9) can you see the door full view? [Note: ( . . . ) indicates pauses in seconds/fractions of a second.] (Shehadeh, 1999b, p. 5) Shehadeh (1999b) commented that this example illustrates the NNS’s experimentation with and testing out of hypotheses to see which sounds better (near of the bed . . . near of the bed . . . near to the bed). Shehadeh argued that the example is suggestive because it reveals the learner’s satisfaction with the outcome of his hypothesis-testing attempts as demonstrated by the resurfacing of that same utterance subsequently in the turn (near to the bottom left corner . . . near to the door) since the outcome of his experimentation attempts was not challenged by the NS interlocutor. Shehadeh (1999b) speculated that in those instances like Example 13 in which the outcome of HTEs is not challenged by partners’ or external feedback,4 hypotheses are likely to be confirmed (not revised or rejected) and given the “go-ahead” signal from the perspective of the internal processing systems of the learner to be internalized and integrated into the learner’s linguistic knowledge base. Stated differently, the absence of partners’ or external feedback may constitute a signal for the learner to confirm his or her self-generated hypotheses about the TL, which, in 636 Language Learning Vol. 52, No. 3 turn, constitutes a step in the direction of internalizing linguistic knowledge.5 Because no posttests were administered, however, we do not know if these learner-generated hypotheses were actually confirmed, as suggested here, or had any long-term impact on SLA. It is difficult to speculate to what degree the absence of partner feedback to this hypothesis-testing process actually affected L2 learning. Nevertheless, Shehadeh’s (1999b) study may be considered a basis for broader investigation into the role of hypothesis testing in internalizing linguistic knowledge. Future research must first conduct a detailed analysis of hypothesis testing by the learner (for instance, by using self-reports and conducting introspection analyses and retrospective interviews) to show to what extent learner-generated hypotheses about the TL can actually be a factor in internalizing linguistic knowledge. After that, researchers must explore whether confirmed hypotheses exhibiting non-target-like utterances or rules would lead to permanent internalization of linguistic knowledge, or permanent “premature stabilization” in Long’s (1996, p. 423) words, or whether and to what degree these hypotheses are still open to further revisions in the light of new evidence. Output as a Tool for Syntactic Processing and L2 Learning Finally, and importantly, future research must examine the function of output as a tool for syntactic processing and L2 learning. According to the output hypothesis, speaking or writing enables learners to move from the open-ended semantic analysis of the TL to a more syntactic analysis of it (Swain, 1985, 1993). As proposed by Swain (1985) and specified by Kowal and Swain (1997), syntactic processing is primarily concerned with the relationships between words (such as word order, inflection, and agreement) and operates mainly at the phrasal, clausal, and sentential levels. Kowal and Swain, for example, wrote Syntactic processing . . . include[s] instances where learners move beyond processing words as independently Shehadeh 637 functioning lexemes and come to consider them in their relationship to other words in the sentence. It thus includes phenomena such as adjectival agreements, verb-subject agreements, [and] word order. (p. 287) As stated earlier, after a period of over 8 years of comprehensible input, the productive skills, speaking and writing, of immersion students “remain far from native-like” (Swain, 1991, p. 98). As far as grammar is concerned, these students continue to make errors in, for example, verb tenses, prepositional usage, and gender markings. The central question to be asked here is therefore: Does output as a triggering tool promote learners’ syntactic processing skills and L2 learning? Kowal and Swain (1997) sought to answer this question by exploring how output enables learners to move from semantic to syntactic processing through collaborative dialogue. The researchers examined data from a larger project (Kowal, 1999) to argue that collaborative dialogue enables learners to move from semantic processing of the TL to a more syntactic processing of it. The data examined were collected from 19 grade 8 immersion students in Ontario, Canada, employing a dictogloss task in which students jointly reconstructed a dictogloss passage. Example 14 is an episode in which two students, Rachel and Sofie, are working together to reconstruct the dictogloss passage as closely to the original as possible. As with other dictogloss tasks (see above), a short, dense text was read to the students at normal speed. While it was being read, the students jotted down familiar words and phrases. After that, the students worked in pairs to reconstruct the text from their shared resources. In the example below, the two students were jointly reconstructing the sentence “M_me les solutions _cologiques causent quelquefois de nouvelles menaces” [Even ecological solutions sometimes cause new threats]. 14. 04 Sofie: 05 Rachel: 06 Sofie: Numéro deux, OK. (Number two, OK.) You had a phrase. OK. [ Students look back over Sofie’s notes.] 638 Language Learning 07 Rachel: 08 Sofie: Même les, les (Even the, the) OK, oui, même le so-lu-tions é-co-logiques . . . um . . . um, posent quelquefois . . . des (OK, yes, even the ecological solutions . . . um . . . sometimes present . . . the), there’s an ‘s’ on ‘Éécologiques’, I remember that. Oooh, look at that! [congratulating her friend on her knowledge] Stud-ley [i.e., Great!], OK, causent, e-n-t. [spelling out the silent third person plural verb ending] Caus-ah-ent [pronouncing the normally unheard third person plural verb ending] Studs [i.e., Great!] again! Quelquefois des (Sometimes) Causent des nouvelles problèmes, pour nos (Cause new problems for our) Des nouveaux (new). Cher[chez] nou[veaux], des nouveaux menaces (Look up new [as in] new threats) Good one! [Congratulating her friend on finding a synonym for “problem”] Yeah, nouveaux, des nouveaux. (new, ‘des’ new, ‘de’ new) [checking which partitive form to use]. Is it ‘des nouveaux’ or ‘de nouveaux’? Des nouveaux or de nouveaux? [Masculine plural form of the adjective or feminine plural form] Nou[veaux], des nou[veaux], de nou[veaux] It’s menace, un menace, une menace, un menace, menace ay ay ay! [exasperated] (it’s threat [then checking if ‘threat’ is masculine or feminine] Je vais le pauser (I’m going to put it on pause.) [i.e., the tape recorder]. [They look it up in the dictionary.] [triumphantly] C’est des nouvelles! (it’s ‘des nouvelles’!) [i.e., the feminine form.] 09 Rachel: 10 Sofie: 11 Rachel: 12 Sofie: 13 Rachel: 14 Sofie: 15 Rachel: 16 Sofie: 17 Rachel: 18 Sofie: 19 Rachel: 20 Sofie: 21 Rachel: 22 Sofie: Vol. 52, No. 3 Shehadeh 23 Rachel: 24 Sofie: 639 C’est féminin; des nouvelles menaces. (It’s feminine ‘des nouvelles menace’.) Menaces. (Threats). (Kowal & Swain, 1997, pp. 298-299) In their joint effort to reconstruct the dictogloss passage, Rachel and Sofie discussed the targeted grammatical form, the present-tense verb endings (as in causent, e-n-t and caus-ah-ent in turns 10 and 11, respectively), as well as other grammatical forms, namely, agreement (turns 13–19) and gender (turns 20–23). This means, Kowal and Swain (1997) commented, that the dictogloss procedure was a tool for focusing students’ attention on syntactic relationships between the words in a sentence; it made the students more aware of specific grammatical problems and how these arise in the process of production. Kowal and Swain argued that the dictogloss procedure is thus a successful vehicle for promoting students’ syntactic processing skills, concluding that it “can be used to encourage students to process language syntactically and to make them aware of the role that . . . syntax plays in conveying meaning” (p. 305). Later on in the dialogue, the researchers also found that the students discussed other grammatical points, including how the present participle is formed in French and how the past participle can also be used as an adjective. In other words, although the dictogloss task was designed with one grammatical domain in mind, in practice, the students’ discussion covered other domains. Kowal and Swain (1997) argued that “this suggests that the dictogloss approach might be better suited to promoting syntactic processing skills in general than as a means of drawing attention to a particular grammatical point” (p. 300). Kowal and Swain’s (1997) study showed that output encourages learners’ syntactic processing skills, thus answering the first part of the question posed above, but the question of whether or how syntactic processing has a facilitative effect on SLA has yet to be answered. In a recent study, Izumi (2000, in press) invoked the issue of syntactic processing, or using Graf ’s (1994) words 640 Language Learning Vol. 52, No. 3 “integrative processing,” to account for the results of his study with regard to the acquisition of English relative clauses by adult L2 learners of English (see Izumi, 2000, in press). Specifically, Izumi argued that pushed output enables learners both to attend to individual elements that make up a structure and to perceive the relationship among these elements in order to facilitate grammar learning. Izumi (2000) stated that in order to learn how relative clauses are formed in English, one needs not only to pay attention to key form elements in the input, such as a head noun, a relative pronoun, and a preposition, but also to focus one’s attention on how these form elements are related to one another. . . . Attending to and noticing the individual items like “which” or “who,” no matter how intensely one does so, will not by itself lead to the acquisition of the relative clause structure unless these items are grasped in relation to the other related items within the same clause or the sentence. (pp. 365–366) Izumi (2000, p. 367) further argued that output facilitates detection of form elements in the input through internal priming caused by such production mechanisms as grammatical encoding and monitoring mechanisms and promotes integrative processing that enables the learner to conceive a coherent structure among the detected elements. He identified three related processes important to the learning of grammatical structures that are facilitated by output: (1) detection of formal elements in the input via priming induced by internal feedback and monitoring mechanisms engaged during production processing; (2) integrative processing of the target structure, prompted by the grammatical encoding operations engaged during production processing; and (3) noticing of the mismatches between one’s IL form and the TL model, which is aided by the engagement of the above processes plus the encounter to the juxtaposition of the two versions of the form uses highlighting the difference between the two. (p. 370) Shehadeh 641 Izumi (2000) concluded that output facilitates syntactic processing, which, in turn, pushes learners further in their cognitive processing and prompts them to perceive the targeted structure as a single unit and that this occurs by virtue of grammatical encoding operations during the production process, leading to improvement in the use of the targeted grammatical form. Hence, output serves “both as a stimulator of integrative processing and as the glue to connect individual form elements” (p. 366). However, Izumi (2000) acknowledged that his study addressed only the short-term impact of syntactic processing on L2 learning, pointing to the need for investigation that explores the long-term effect of syntactic processing on SLA: “since the present study assessed the impact of the treatment only in the short term, investigation on the long-term effects will be necessary” (p. 377). On the basis of existing research (Izumi, 2000, in press; Kowal & Swain, 1997), one can safely say that production facilitates syntactic processing, but whether or not the latter has a facilitative effect or a long-term impact on SLA remains an open question that awaits further investigation. Concluding Remarks The interest in Swain’s (1985, 1995) CO hypothesis is both theoretical and pedagogical. From a theoretical perspective, researchers want to know to what extent learners’ output plays a role in SLA. From a pedagogical perspective, knowing the extent to which learners’ production of CO affects SLA may provide insights that help educators and language teachers make language learning more effective. In this article, I reviewed research on CO and demonstrated that most of existing research has been descriptive in nature, focusing more on occurrence than acquisition. I argued that it is necessary to move beyond occurrence to examine the impact, long or short term, that learner output may have on language development. Specifically, I outlined a research agenda consisting of two sets of research directions. The first set relates to MO and includes 642 Language Learning Vol. 52, No. 3 examining the effect of MO on L2 learning, examining the direction toward which modifications are made, the specific type of linguistic modification NNSs produce, and the effect of frequency of MO on L2 learning. The second set relates to the function of output in L2 learning as a triggering process and includes output as a tool for metalinguistic talk, for noticing and focusing learners’ attention on subsequent input, for hypothesis testing, and for syntactic processing. This research agenda makes acquisitional research central to the study of CO. It suggests research directions and methods that take the field beyond descriptive studies to investigate whether and how learner output affects language learning, enabling researchers, consequently, to make stronger claims about the role of CO in SLA. Revised version accepted 15 February 2002 Notes 1 It must be noted that supremacy of self-initiated over other-initiated MOs depends to some extent on the level of the learner, because self-initiation is feasible only when learners already possess an adequate level of proficiency (Lyster & Ranta, 1997, p. 58). 2 A language-related episode (LRE) is defined as “any part of a dialogue in which students talk about the language they are producing, question their language use, or other- or self-correct” (Swain, 1998, p. 70). 3 The noticing function of output can include not just focusing learners’ attention on subsequent input or a search in further incoming input, but also a search of one’s own existing knowledge base and “playing” with it. Typically, this either (a) pushes learners to reprocess and modify their output toward greater message comprehensibility (see “MO and L2 Learning”) or (b) allows learners to formulate and test hypotheses about comprehensibility and linguistic well-formedness of TL rules and structures (see “Output as a Tool for Hypothesis Testing and Internalizing Linguistic Knowledge”). 4 Shehadeh (1999b, p. 12) pointed out that there are occasions when learner hypotheses go unchallenged by partners’ or external feedback, namely, (a) when these hypotheses are compatible with the TL norms (i.e., when they are correct), (b) when the interlocutor, for some reason, does not want to provide negative or corrective feedback, and (c) when the interlocutor is unable to provide negative or corrective feedback (e.g., as in the case of a fellow NNS interlocutor). Shehadeh 643 5 Shehadeh (1999b, pp. 12-13) also pointed out that whether the hypothesistesting process suggested here leads to the internalization of the correct TL structures/rules or the incorrect ones, this does not negate the possibility that this process may actually lead to the internalization of linguistic knowledge (see also Swain, 1998, pp. 78–80, for examples; Pica, 2001, for discussion). In fact, the present explanation also accounts in part for the phenomenon of fossilization, or what Long (1996, p. 423) calls “premature stabilization,” in the L2. That is, the absence of negative evidence to the NNS’s output which exhibits non-target-like utterances or rules can also lead to the confirmation of these utterances or rules, internalizing and integrating them into the linguistic knowledge base of the learner, thus resulting in premature stabilization. For a similar argument, see Ellis (1994), which, on the basis of previous evidence (Vigil & Oller, 1976), argued that “positive cognitive feedback (signaling ‘I understand you’) results in fossilization” (p. 354). References Aston, G. (1986). Trouble-shooting in interaction with learners: The more the merrier? Applied Linguistics, 7(2), 128–143. Corder, P. (1981). Error analysis and interlanguage. Oxford, England: Oxford University Press. Ellis, R. (1994). The study of second language acquisition. Oxford, England: Oxford University Press. Gass, S., & Varonis, E. (1986). Sex differences in nonnative speaker–nonnative speaker interactions. In R. Day (Ed.), Talking to learn (pp. 327–352). Rowley, MA: Newbury House. Graf, P. (1994). Explicit and implicit memory: A decade of research. Attention and Performance, 15, 681–696. Hatch, E., Flashner V., & Hunt, L. (1986). The experience model and language teaching. In R. Day (Ed.), Talking to learn (pp. 5–22). Rowley, MA: Newbury House. Hawkins, B. (1985). Is an “appropriate response” always so appropriate? In S. Gass & C. Madden (Eds.), Input in second language acquisition (pp. 162–178). Rowley, MA: Newbury House. Iwashita, N. (1993). Comprehensible output in NNS-NNS interaction in Japanese as a foreign language. Unpublished master’s thesis, University of Melbourne, Melbourne, Australia. Iwashita, N. (1999). Tasks and learners’ output in nonnative-nonnative interaction. In K. Kanno (Ed.), The acquisition of Japanese as a second language (pp. 31–52). Amsterdam: John Benjamin. 644 Language Learning Vol. 52, No. 3 Iwashita, N. (2001). The effect of learner proficiency on interactional moves and modified output in nonnative-nonnative interaction in Japanese as a foreign language. System, 29, 267–287. Izumi, S. (2000). Promoting noticing and SLA: An empirical study of the effects of output and input enhancement on ESL relativization. Unpublished doctoral dissertation, Georgetown University, Washington, DC. Izumi, S. (in press). Output, input enhancement, and the noticing hypothesis: An experimental study on ESL relativization. Studies in Second Language Acquisition, 24. Izumi, S., & Bigelow, M. (2000). Does output promote noticing and second language acquisition? TESOL Quarterly, 34(2), 239–278. Izumi, S., Bigelow, M., Fujiwara, M., & Fearnow, S. (1999). Testing the output hypothesis: Effects of output on noticing and second language acquisition. Studies in Second Language Acquisition, 21, 421–452. Kowal, M. (1999). French immersion students’ language growth in French: Perceptions, patterns and programming. Unpublished doctoral dissertation, Ontario Institute for Studies in Education, University of Toronto, Toronto, Ontario, Canada. Kowal, M., & Swain, M. (1994). Using collaborative language production tasks to promote students’ language awareness. Language Awareness, 3(1), 73–93. Kowal, M., & Swain, M. (1997). From semantic to syntactic processing: How can we promote it in the immersion classroom? In R. K. Johnson & M. Swain (Eds.), Immersion education: International perspectives (pp. 284–309). Cambridge, UK: Cambridge University Press. Krashen, S. (1985). The input hypothesis: Issues and implications. London: Longman. Krashen, S. (1994). The input hypothesis and its rivals. In N. Ellis (Ed.), Implicit and explicit learning of languages (pp. 45–77). London: Academic Press. Krashen, S. (1998). Comprehensible output? System, 26, 175–182. LaPierre, D. (1994). Language output in a cooperative learning setting: Determining its effects on second language learning. Unpublished master’s thesis, University of Toronto, Toronto, Ontario, Canada. Long, M. (1996). The role of the linguistic environment in second language acquisition. In W. Ritchie & T. K. Bhatia (Eds.), Handbook of second language acquisition (pp. 413–468). Orlando, FL: Academic Press. Loschky, L., & Bley-Vroman, R. (1990) Creating structure-based communication tasks for second language development. University of Hawaii Working Papers in ESL, 9, 161–209. Lyster, R. (1998a). Recasts, repetition, and ambiguity in L2 classroom discourse. Studies in Second Language Acquisition, 20, 51–81. Shehadeh 645 Lyster, R. (1998b). Negotiation of form, recasts, and explicit correction in relation to error types and learner repair in immersion classrooms. Language Learning, 48(2), 183–218. Lyster, R., & Ranta, L. (1997). Corrective feedback and learner uptake: Negotiation of form in communicative classrooms. Studies in Second language Acquisition, 19, 37–66. McHoul, A. (1978). The organization of turns at formal talk in the classroom. Language in Society, 7, 182–213. McHoul, A. (1990). The organization of repair in classroom talk. Language in Society, 19, 349–377. McLaughlin, B. (1987). Theories of second language learning. London: Edward Arnold. McLaughlin, B. (1990). Restructuring. Applied Linguistics, 11(2), 113–128. Nobuyoshi, J., & Ellis, R. (1993). Focused communication tasks and second language acquisition. ELT Journal, 47, 203–210. Oliver, R. (1995). Negative feedback in child NS-NNS conversation. Studies in Second Language Acquisition, 17, 459–481. Pica, T. (1988). Interlanguage adjustments as an outcome of NS-NNS negotiated interaction. Language Learning, 28(1), 45–73. Pica, T. (1994). Research on negotiation: What does it reveal about secondlanguage learning conditions, processes and outcomes? Language Learning, 44(3), 493–527. Pica, T. (2001). The content-based curriculum: Does it provide an optimal or optional approach to second language learning? In W. Renandya & N. Sunga (Eds.), Language curriculum and instruction in multicultural societies (pp. 145–174). Singapore: SEAMEO Regional Language Center. Pica, T., Holliday, L., Lewis, N., Berducci, D., & Newman, J. (1991). Language learning through interaction: What role does gender play? Studies in Second Language Acquisition, 13(3), 343–376. Pica, T., Holliday, L., Lewis, N., & Morgenthaler, L. (1989). Comprehensible output as an outcome of linguistic demands on the learner. Studies in Second Language Acquisition, 11(1), 63–90. Pica, T., Kanagy, R., & Falodun, J. (1993). Choosing and using communication tasks for second language research and instruction. In G. Crookes & S. Gass (Eds.), Tasks and language learning: Integrating theory and practice (pp. 9–34). Clevedon, England: Multilingual Matters. Pica, T., Lincoln-Porter, F., Paninos, D., & Linnell, J. (1996). Language learners’ interaction: How does it address the input, output, and feedback needs of L2 learners? TESOL Quarterly, 30(1), 59–84. Sato, C. (1986). Conversation and interlanguage development: Rethinking the connection. In R. Day (Ed.), Talking to learn (pp. 23–45). Rowley, MA: Newbury House. 646 Language Learning Vol. 52, No. 3 Schachter, J. (1984). A universal input condition. In W. Rutherford (Ed.), Universals and second language acquisition (pp. 167–183). Amsterdam: John Benjamin. Selinker, L. (1972). Interlanguage. International Review of Applied Linguistics, 10, 209–231. Shehadeh, A. (1994). Gender differences and second language acquisition. Research Journal of Aleppo University (Syria), 26, 73–98. Shehadeh, A. (1999a). Non-native speakers’ production of modified comprehensible output and second language learning. Language Learning, 49(4), 627–675. Shehadeh, A. (1999b). Learner hypothesis testing and internalizing linguistic knowledge. In Selected Papers from the 12th World Congress of the International Association of Applied Linguistics (AILA ’99), Tokyo, Japan (pp. 1–14) [CD-ROM publication, CD # 3]. Shehadeh, A. (2001). Self- and other-initiated modified output during taskbased interaction. TESOL Quarterly, 35(3), 433–457. Swain, M. (1984). A review of immersion education in Canada: Research and evaluation studies. ELT Documents, 119, 35–51. Swain, M. (1985). Communicative competence: Some roles of comprehensible input and 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. (1991). French immersion and its offshoots: Getting two for one. In B. F. Freed (Ed.), Foreign language acquisition research and the classroom (pp. 91–103). Lexington, MA: D.C. Heath. Swain, M. (1993). The output hypothesis: Just speaking and writing aren’t enough. Canadian Modern Language Review, 50, 158–164. Swain, M. (1995). Three functions of output in second language learning. In G. Cook & B. Seidlhofer (Eds.), Principle and practice in applied linguistics: Studies in honor of H.G. Widdowson (pp. 125–144). Oxford, England: Oxford University Press. Swain, M. (1997). Collaborative dialogue: Its contribution to second language learning. Revista Canaria de Estudios Ingleses, 34, 115–132. 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, England: Cambridge University Press. Swain, M. (2000). The output hypothesis and beyond: Mediating acquisition through collaborative dialogue. In C. Lantolf (Ed.), Sociocultural theory and second language learning. Oxford, England: Oxford University Press. Swain, M., & Lapkin, S. (1995). Problems in output and the cognitive processes they generate: A step towards second language learning. Applied Linguistics, 16(3), 371–391. Shehadeh 647 Swain, M., & Lapkin, S. (1998). Interaction and second language learning: Two adolescent French immersion students working together. Modern Language Journal, 82(3), 320–337. Tarone, E., & Liu, G. Q. (1995). Situational context, variation, and second language acquisition theory. In G. Cook & B. Seidlhofer (Eds.), Principle and practice in applied linguistics: Studies in honor of H. G. Widdowson (pp. 107–124). Oxford, England: Oxford University Press. Van den Branden, K. (1997). Effects of negotiation on language learners’ output. Language Learning, 47(4), 589–636. Varonis, E., & Gass, S. (1985). Non-native/non-native conversations: A model for negotiation of meaning. Applied Linguistics, 6(1), 71–90. Vigil, F., & Oller, J. (1976). Rule fossilization: A tentative model. Language Learning, 26, 281–295. White, L. (1991). Adverb placement in second language acquisition: Some effects of positive and negative evidence in the classroom. Second Language Research, 7, 133–161.