by Rod Ellis
There is another reason why negotiated interaction is not always beneficial for grammar learning. As we saw in Chapter 4, learners do not acquire new grammatical features immediately, but pass through transitional stages on route to the target structure. If learners are not developmentally ready, the input they receive from negotiation may not connect with their internal processing mechanisms so acquisition does not take place. However—even if they do not master the target structure—negotiation may still help them progress to a more advanced transitional stage. Mackey’s (1999) study set out to examine if this was the case. She found that only those learners who received interactionally modified input demonstrated clear evidence of having advanced along the developmental sequence for questions. That is, even if they did not reach the final target stage, the interaction helped them developmentally.
Another important question is whether the positive effects for negotiated interaction are durable or just short-lived. This can be addressed by comparing the results of post-tests administered shortly after the interactive activity has been completed with those of post-tests completed later. In fact, a number of studies have reported stronger effects in a delayed post-test. Ellis, Loewen, and Erlam (2006), for example, investigated the effects of two feedback strategies—recasts and metalinguistic comments—on learners’ acquisition of regular English past tense. The effect was more clearly evident in the delayed test administered two weeks later. This might seem surprising, but can be explained by the fact that sometimes information is not immediately integrated into the learner’s interlanguage, but is put into storage and only fully activated later—as proposed by Gass (1997).
Contextual factors have also been found to influence the effect that negotiated interaction has on acquisition. Many of the studies have been laboratory based. Mackey and Goo’s (2007) meta-analysis included 18 laboratory-based studies and ten classroom-based studies. Also, the majority of the studies in this meta-analysis involved foreign as opposed to second language contexts. These contextual differences can affect the extent to which learners are predisposed to attend to the linguistic features that are negotiated. Many of the laboratory studies involved one-on-one interactions where the target features were likely to have been more salient to learners than in classrooms, where the teacher is interacting with the whole class. Also, foreign language learners are likely to be more form-conscious than second language learners. Negotiated interaction has been found to have a larger effect on acquisition in laboratory and foreign language settings than in classroom and second language settings.
Another contextual factor is whether the interaction takes place face to face or synchronously in a chat room. The fact that the online interaction is written allows more time for noticing. However, the absence of non-verbal cues may make the online input less comprehensible. Researchers have addressed two key questions: ‘Does negotiation of meaning occur in synchronous computer-based interaction?’ and, if so, ‘Does it facilitate acquisition?’ A number of studies (for example, de la Fuente 2003) found that negotiation is common when learners perform communicative tasks online and that—as in face-to-face negotiation—much of it is directed at lexical rather than grammatical problems. Ziegler (2015) reported a meta-analysis of studies that had compared the effects of negotiation on acquisition in face-to-face and synchronous computer-mediated communication. The main finding was that negotiation was beneficial in both modes and that overall there were no significant differences.
So far, we have considered the overall benefits of negotiated interaction. However, we also need to consider more precisely how negotiated interaction assists acquisition. In the next section, I focus on the role of output and then move on to examine the effects of different negotiation/feedback strategies.
Modified output and acquisition
Long’s revised Interaction Hypothesis drew in part on Swain’s (1985, 1995) Output Hypothesis which proposed that language acquisition is not just driven by input but also by output, a view that is now widely accepted.
The Output HypothesisNOTE 1
Swain (1985) proposed the Output Hypothesis as a complement to the Input Hypothesis. Her conviction that output plays an important role in L2 acquisition grew out of her experience of investigating immersion programmes in Canada. These programmes cater to incidental acquisition by teaching subject content in the L2—for example, maths or science. The Input Hypothesis provided a theoretical basis for such programmes as they afforded plentiful comprehensible input. However, Swain’s evaluation of the immersion programmes showed that although learners developed considerable fluency in the L2, they did not achieve high levels of grammatical and sociolinguistic competence. This led her to conclude that comprehensible input was insufficient and that what was missing was the opportunity for students to engage in pushed output—i.e. output consisting of messages that were concise and socially appropriate. As Swain (1985) put it, ‘learners … can fake it, so to speak, in comprehension, but they cannot do so in the same way in production’ (p. 127). As we have already seen, learners can comprehend input using top-down strategies. Production, however, requires learners to process bottom-up. Swain (1995) went on to suggest that production was important in three main ways: (1) it can trigger noticing when learners realize that something they said does not effectively communicate what they wanted to say; (2) it serves a hypothesis-testing function when learners try out a way of saying something and then receive feedback; and (3) it provides a basis for metalinguistic reflection when learners consciously think about what they have said or written.
Modified output creates opportunities for learning
Interaction is not the only way in which learners can be ‘pushed’ to modify their output, but it is certainly one of the main waysNOTE 2. Interaction involving feedback on learners’ attempts to express their meaning often results in modified output, which constitutes one kind of pushed output. Example (5) below illustrates this. The problem arises when Learner 2 does not hear (or understand) the word ‘nuclear’ and signals a problem by means of a confirmation check (turn 2). Learner 1 responds by explicitly disconfirming Learner 2’s hypothesis and repeating the key word (‘nuclear’) three times (turn 3). The modified output occurs in turn 4, where Learner 2 produces the correct word and also indicates her understanding (‘Ah, I see.’). We do not know if this learner acquired ‘nuclear’, but it is clear that the negotiation helped the learner to both understand and pronounce the word correctly. Thus, there is potential for acquisition to have taken place.
Example (5)NOTE 3
1 L1 Einstein’s scientific work helped Americans make the nuclear bomb.
2 L2 Clear bomb?
3 L1 No nuclear, nuclear, nuclear bomb.
4 L2 Nuclear bomb. Ah, I see.
(Aubrey: unpublished data)
Example (5) shows how modified output can arise in negotiation, triggered by a learner’s failure to comprehend another speaker’s utterance. Example (6) illustrates how learners modify their own utterances when these cause a problem. The problem arises in turn 3. The NS responds with a confirmation request that corrects the learner’s utterance (turn 4). The learner then modifies her initial utterance (turn 5). This example illustrates how negotiation can help learners to notice the gap between their own ill-formed utterance and the target-language version. Again, though, we cannot be sure that this has helped the learner to acquire the irregular verb ‘came’. It is possible that the learner is just parroting the native speaker’s correction. But it is also possible that genuine intake has occurred, establishing the potential for acquisition.
Example (6)
1 L After that he was waiting for her every day.
2 NS Waiting and waiting mhm.
3 L But she was never come back.
4 NS She never came back?
5 L Never came back, he was very sad.
(McDonough 2007: 332)
In Example (6), the learner is provided with the correct form. In Example (7), however, the learner i
s pushed to reformulate the initial utterance, but without any positive evidence. This occurs when the negotiation sequence involves a prompt (in this case a clarification request). In turn 2, the teacher requests clarification which pushes the learner to produce the correct past tense form in turn 3. In this kind of negotiation sequence, however, a learner will only be able to modify her output if she already knows the correct form. Such sequences may still assist acquisition by enabling the learner to achieve fuller control over a partially acquired feature.
Example (7)
1 L He pass his house.
2 T Sorry?
3 L He passed, he passed, ah, his sign.
(Nobuyoshi and Ellis 1993: 204)
In these examples, the learners succeeded in correcting their errors when they modified their output. In fact, this does not always happen. Sometimes, learners modify their output, but fail to produce the correct target form. However, Mackey (2007) argued that benefits can accrue even when the modified output is not target-like.
Finally, modified output does not only occur as a result of negotiation. Swain (1998) chose to investigate output in language-related episodes (LREs), which she 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’ (p. 70). LREs include both negotiation of meaning and negotiation of form sequences, but they also include sequences where learners elect to focus pre-emptively on some linguistic object that they wish to use in production, as illustrated in Example (8). Pre-emptive attention to form often involves metalinguistic reflection—the third of the functions of output that Swain proposed in her Output Hypothesis.
Example (8)
1 L1 How can I say that? Ay. I don’t remember, but I know the name.
It’s go whoo and you can make choice.
2 L2 Ah, blender?
(Swain 1998)
Experimental studies of modified output and acquisition
As already noted, the fact that learners modify their output—often correcting their errors in the process—is not in itself evidence of acquisition. At best, it demonstrates that intake has occurred (i.e. learners have noticed their error and may have carried out a cognitive comparison). Experimental studies are needed to demonstrate that output modification leads to acquisition. In such studies, scores in a pre-test administered before the interaction took place are compared with scores in a post-test administered after the interaction is over. Recent years have seen a plethora of experimental studies investigating the effects of modified output on acquisition.
One of the earliest experimental studies was Nobuyoshi and Ellis (1993). Three learners performed an oral narrative and were pushed to modify their output through clarification requests whenever they made an error in the use of a regular past tense. Nobuyoshi and Ellis found that two of these learners responded by correcting their errors, but the other learner did not. Only the learners who self-corrected demonstrated gains in accuracy in the use of the past tense when they performed another oral narrative. Another group of learners who performed the same tasks—but without any negotiation—also failed to improve in accuracy. This study was small scale but it suggests that pushing learners to produce more target-like output has an effect on acquisition for some learners.
There is also some evidence that modified output works more effectively for acquisition than either pre-modified or interactionally-modified input. Both Ellis and He (1999) and de la Fuente (2002) reported that negotiated interaction that led to pushed output promoted productive acquisition of new words to a greater extent than either pre-modified input or negotiated interaction without pushed output. Izumi (2002) also found that groups of learners who performed a task where they had to produce the target structure (English relative clauses) outperformed those groups that just received pre-modified input. In line with Swain’s Output Hypothesis, Izumi suggested that output was more effective because it induced a cognitive comparison.
A number of studies have looked at the effect of modified output in learner– learner interactions. Adams, Nuevo, and Egi (2011) reviewed the relevant studies. They noted that modified output occurs in negotiation sequences between learners but that ‘learner–learner feedback may not always prompt modified output that can lead to learning gains’ (p. 47). Their own study of adult ESL learners found that the production of modified output following recasts and explicit corrections was related to learning, but only in a delayed grammaticality judgement test, which they suggested afforded a measure of explicit knowledge. They reported no relationship when learning was measured by tests that were more likely to tap implicit knowledge.
Overall, modified output is more likely to occur in interactions with competent speakers of the language than with other learners. However, not all studies involving native speaker interlocutors have found that modified output following feedback aids acquisition. Mackey and Goo (2007) felt unable to come to any clear conclusion about the contribution that modified output makes to acquisition. One reason for this is that, in many instances, modified output may simply involve repeating or mimicking a recast. McDonough and Mackey (2006) reported a laboratory-based study, which showed that repeating or mimicking a recast was not beneficial for learning, whereas productive use of a form a short time after hearing it—suggestive of deeper processing—was.
Corrective feedback and L2 acquisition
The studies we will now consider all involved corrective feedback strategiesNOTE 4. Earlier, I pointed out that these strategies can be input-providing or output-prompting and implicit or explicit (see Table 7.1). The question that researchers have addressed is ‘Are some types of corrective feedback strategies more effective than others?’
Input-providing vs. output-prompting strategies
The relative effectiveness of input-providing strategies—such as recasts or confirmation checks (CF) which reformulate the learner’s erroneous utterances—and prompts—such as requests for clarification which push learners to modify their output—has been a matter of debate. In part, this debate hangs on the importance that is attached to modified output. If acquisition is viewed as primarily input driven, then, clearly, whether or not learners repair their errors is of little importance. If, on the other hand, modified output is seen as assisting acquisition, as claimed by the Output Hypothesis, then negotiation that results in repair of errors is desirable. Long (2006) argued that it is recasts that are important for acquisition on the grounds that learners cannot acquire new target features unless they receive input. In contrast, Lyster (2004) has argued that output-prompting CF is more likely to facilitate acquisition because it helps learners to achieve greater control over those grammatical structures they have partially acquired. Lyster and Saito’s (2010) meta-analysis of studies that had compared input-providing and output-prompting CF reported in favour of the latter.
However, a number of caveats are in order. First, recasts constitute a single corrective strategy, whereas the prompts include four different strategies—i.e. clarification requests, repetition of error, elicitation, and metalinguistic clues. It is possible that the greater effect found for prompts is simply because many strategies are more effective than one. Also—whereas recasts are often implicit—prompts include a mixture of implicit and explicit strategies, so it is possible that they are more effective because they are more salient. Third, the effects of the two types of CF are likely to be mediated by a number of factors, such as the instructional tenor of the classroom (i.e. whether it is primarily meaning or form focused); the proficiency level of the learners; and the nature of the target feature. For example, Lyster and Mori (2006) reported that recasts proved more effective than prompts in an instructional context where the learners were more inclined to pay attention to form.
Recasts are clearly effective in promoting acquisition, especially when they are made salient to the learner. Doughty and Varela (1998) investigated the effect of ‘corrective recasts’ on ESL learners’ use of past verb forms. As t
hey defined them, these consist of an initial prompt—in this case, a repetition of the learner’s erroneous utterance—followed by a recast if the learner fails to self-correct. See Example (9). Doughty and Varela reported that the recasts resulted in progress through the acquisition sequence for past tense and also in more target-like use in an immediate and delayed oral post-test.
Example (9)
1 L I think that the worm will go under the soil.
2 T I think that the worm will go under the soil?
3 L (no response)
4 T I thought that the worm would go under the soil.
(Doughty and Varela 1998: 124)
The debate about the relative effectiveness of input-providing and output-prompting types of CF is ongoing. See for example, Goo and Mackey (2013) and Lyster and Ranta (2013). Perhaps this debate is somewhat unnecessary. Recasts provide learners with positive evidence and thus make it possible for them to acquire new linguistic features. Prompts—on the other hand—can help learners consolidate their knowledge of those features they have prior knowledge of, so both are important. Arguably, learners benefit most when they receive a mixture of these two types of CF.
Implicit vs explicit corrective strategies
Studies that have investigated the relative effects of implicit and explicit CF have typically compared recasts (an implicit strategy) with one or more types of explicit strategies (for example, metalinguistic comments or explicit correction). It should be noted, however, that recasts (the implicit CF strategy most commonly investigated) differ considerably in how implicit or explicit they are. It is the more explicit types of recasts that have proved more likely to promote learning (Loewen and Philp 2006).