by Rod Ellis
A good example of such a study is Ellis et al. (2006) referred to earlier. The results showed that the group receiving a repetition of an incorrect verb form followed by a metalinguistic comment outperformed both the control group and a group that received recasts. However, the differences only reached statistical significance in the delayed post-tests. Yilmaz (2012) carried out the same comparison in a face-to-face context and a synchronous computer-mediated context and found a clear advantage for explicit correction over recasts in both immediate and delayed tests.
Overall, explicit CF is more effective than implicit CF. However Li (2010), in another meta-analysis, found that overall implicit CF resulted in higher scores in post-tests completed a long time after the instruction. This was because its effect increased over time whereas those of explicit CF did not change. Neither Ellis et al.’s or Yilmaz’s study investigated this possibility as their post-tests were administered after only a relatively short interval.
General comments
A general finding from all the studies I have considered in this section is that corrective feedback is effective irrespective of the type of strategy involved. In this respect, these studies reinforce the conclusion I reached earlier, namely that the negotiation of meaning and of form facilitate acquisition. Li’s (2010) meta-analysis of 33 studies involving CF reached exactly this conclusion. However, Li also found that the strength of the effect was influenced by a number of factors. For example, it was much stronger in studies carried out in a laboratory or foreign-language setting than in a classroom or second-language setting perhaps—as I have already suggested—because learners pay greater attention to the feedback they receive in the former settings. Another possibility—which I consider in the next section—is that individual difference factors influence the extent to which learners benefit from CF and also which type of CF is more effective.
Interaction, working memory, and acquisition
Learner difference factors such as language aptitude, working memory, motivation, and language anxiety can be expected to influence how learners process input and output and thus the extent to which they are able to benefit from interaction. We saw evidence of this in Chapter 3. In this section, I will focus on the individual difference factor that has attracted the greatest attention in interaction research—working memory. Readers might like to refresh their memory of the construct by reading Chapter 3.
Working memory is hypothesized to mediate learners’ ability to attend to and learn from negotiation. That is, learners with higher working memory may be better able to notice the forms that have been corrected in the feedback, respond to the feedback by modifying their output, and relate information in working-term memory to that held in long-term memory.
Studies (for example, Mackey, Philp, Egi, Fujii, and Tatsumi 2002; Trofimovich, Ammar, and Gatbonton 2007) that have investigated these possibilities have produced mixed results, reflecting perhaps methodological differences in their design. The researchers used different ways of measuring working memory and the interactional treatments they provided differed in how salient recasts were made to learners. Later studies, however, have provided clearer evidence that working memory is indeed a factor influencing how recasts are processed. I will consider one of these here.
Révész (2012) obtained measures of both phonological short-term memory—i.e. the capacity to hold aural information in memory—and complex working memory—i.e. the capacity to process information held in short-term memory—and correlated these with test gain scores for two groups of learners. Both groups performed communicative tasks, but only one group received recasts. Révész found that for those learners who did not receive recasts, differences in working memory were not related to gains in the target structure. However, in the group that received recasts, there was a relationship. Phonological working memory was related to accuracy gains in an oral description task, while measures of complex working memory were related to gains in written tests. These results might seem puzzling, but they can be explained in terms of the distinction between procedural knowledge and declarative knowledge. Those learners with high phonological short-term memory may have benefited from the recasts because they were able to maintain the information in short-term memory longer, which contributed to the development of the procedural knowledge needed for oral production. Those learners with stronger complex working memory may have been able to consciously attend to the information provided by the recasts, which led to the declarative knowledge helpful for written production.
Work investigating the mediating role of working memory is still in its infancy. Révész’s results are especially interesting as they suggest that different components of working memory may be important for the development of different kinds of linguistic knowledge—i.e. procedural vs. declarative—resulting from feedback. This suggests that research investigating the effects of interaction on learning needs to pay close attention to how learning is measured. I comment on this in the next section.
Measuring the effects of input and interaction on acquisition
In Chapter 1, I pointed out that ‘acquisition’ can be considered in terms of whether learners can comprehend the meaning of a particular feature or in terms of their ability to produce it. I also noted that it is necessary to distinguish different types of knowledge: implicit/procedural and explicit/declarative. Acquisition can also be viewed as progression through developmental sequences.
By and large, the studies we have examined in this chapter have measured acquisition in terms of production rather than comprehension. In most cases, these studies have not attempted to establish whether the acquisition that results from input and interaction is of the implicit or explicit kind. It is, however, clearly important to do so.
How, then, can we tell whether input and interaction lead to ‘development’ or ‘storage’? One way is by administering tests designed to provide separate measures of explicit and implicit knowledge. Here I, will point to two studies that have attempted it. Ellis et al. (2006), in the study discussed above, measured the effects of the two types of corrective feedback by means of an elicited oral imitation test—intended to measure implicit knowledge—and an untimed grammaticality judgement test—designed to measure explicit knowledge. Overall, they found that the explicit correction proved to be a more effective corrective treatment than the implicit in both tests. One interpretation of their study, therefore, is that explicit correction assisted the development of both types of knowledge. Interestingly, however, while there was no evidence that the recasts contributed to explicit knowledge, there was some evidence that they led to gains in implicit knowledge. While no firm conclusions can be drawn from this study, it does suggest that different types of corrective feedback may affect learning in different ways.
Révész (2012), in the study referred to above, also used different tests, designed to tap different types of knowledge. She included an untimed grammaticality judgment test, a written picture description test, and an oral description task. She suggested that the grammaticality judgment test tapped the learners’ declarative knowledge; the oral test tapped their procedural knowledge; and the written test potentially both types of knowledge. The results of her study showed that the effectiveness of the feedback strategy she investigated (recasts) varied according to how acquisition was measured. The recasts were found to have the greatest effect on scores in the oral production test; less effect on the written production test scores; and the least effect on the grammatically judgment test scores. As in Ellis et al (2006), then, the recasts appear to have impacted on learners’ implicit knowledge but not on their explicit knowledge.
Some studies have measured acquisition in terms of learners’ progress along developmental sequences. Mackey (1999) showed that negotiation enabled learners to advance along the stages for the acquisition for questions. Doughty and Varela (1998) showed that corrective recasts helped learners progress from one interlanguage stage to the next for past tense. Progress through a sequence of acqu
isition can be seen as indicative of the development of learners’ implicit knowledge systems (see Chapter 4). Thus, arguably, this constitutes a better measure of the effect of interaction than tests that simply measure accurate use of specific features.
The assumption of the theories we have examined in this chapter—the Interaction Hypothesis, the Noticing Hypothesis, and the Output Hypothesis—is that input and interaction involve incidental acquisition, which is manifest in gradual changes in learners’ implicit knowledge. To demonstrate this convincingly, however, it is necessary to employ methods of assessment capable of providing valid measures of this type of knowledge.
Conclusion
Input and interaction have been major focuses of research in SLA. Why is this? I think there are several reasons. First, the most obvious case of successful language acquisition—acquisition of our first language—is interaction-driven. Children learn their first language through exposure to input and by interacting with their caretakers. Thus, it is likely that a second language—to some extent at least—is learned in the same way. Second, developments in language teaching have led to greater emphasis being placed on communication and this has served as an incentive for investigating how learners learn from communicating. Third, technological developments have made it easier to collect and transcribe samples of interactions involving learners. Finally, developments in discourse analysis have provided the descriptive tools needed to analyse these samples. There is every sign that input and interaction will continue to figure as a major line of research in SLA.
Below is a summary of the main findings that have emerged from the interaction approach:
The interaction approach assumes that input and interaction provide learners with L2 data that are processed internally as intake in working memory and then potentially incorporated into their developing interlanguages. However, not all intake enters the learner’s L2 system.
Researchers in this paradigm have identified a number of key constructs about the nature of input and interaction that are theorized to influence intake and incidental L2 acquisition—pre-modified input, interactionally modified input, and modified output.
A number of different theories have informed the interaction approach. These are summarized in Table 7.2.
Whereas the Input Hypothesis assumes that acquisition is a subconscious process, both the Interaction Hypothesis and the Output Hypothesis claim that noticing facilitates acquisition—i.e. learners benefit from consciously focusing on form while they are communicating. Current accounts of the role of input and interaction accept that noticing is important for acquisition to take place.
Researchers have investigated whether enhanced input results in noticing. In general, the results have been somewhat disappointing. Enhanced input does induce noticing, but only to a limited extent. Ultimately, it is what a learner chooses to selectively attend to in the input that matters.
Researchers have also investigated whether interactionally-modified input leads to noticing. Negotiation of meaning and form has been shown to induce noticing of phonological, lexical, semantic, and some grammatical elements (for example, question forms). However, it is much less effective in helping learners to notice morphological features, such as verb tense inflections.
Other research has focused on the relationship between pre-modified input and acquisition. When learners are exposed to massive amounts of simplified input that they can comprehend, acquisition has been shown to take place incidentally. Input-enhancement, however, has been found to have only a small effect on acquisition.
When input is modified interactionally through negotiation, both comprehension and acquisition benefit. However—where acquisition is concerned—the effect appears to be much larger for vocabulary than for grammar, reflecting once again the role of noticing as learners are more likely to pay attention to lexical than to grammatical problems. Another reason why interaction has less effect on grammar is that structures are acquired gradually in stages. Interaction may help learners to advance through a sequence of acquisition without enabling them to achieve target-like accuracy. How learners orientate to interaction—for example, whether they focus exclusively on meaning or whether they also take opportunities to focus on form—influences its effect on acquisition.
Modifying output through negotiation provides opportunities for learning as it can ‘push’ learners to express themselves more clearly and more accurately. Modified output is more likely to occur when learners interact with competent L2 speakers than with other learners. The relative contribution of interactionally modified input and output to acquisition remains a controversial issue.
Recent research has focused on the effect of different types of corrective feedback—in particular, input-providing strategies, such as recasts, and output-prompting strategies, such as elicitation or metalinguistic comments. Both types have been found to lead to gains in grammatical accuracy. However, what may be crucial is the extent to which the feedback is salient to learners. Reflecting this, explicit types of feedback have been found to be more facilitative than implicit types.
Working memory is believed to play a crucial role in intake. Recent research (for example, Révész 2012) indicates that different components of working memory—i.e. phonological short-term memory and the central executive—may be involved in the processing needed to develop different kinds of knowledge—i.e. procedural and declarative.
Researchers have begun to pay careful attention to how acquisition is measured. There is increasing recognition that input and interaction can contribute to the development of both procedural/implicit and declarative/explicit knowledge and that it is important to ascertain precisely what their contributions are by designing suitable tests.
Theory Description
Input Hypothesis (Krashen 1985) The Input Hypothesis claims that acquisition takes place automatically and without consciousness when learners are exposed to input made comprehensible through context and simplification. It draws on research into simplified registers (i.e. foreigner talk and teacher talk).
Interaction Hypothesis (Long 1983b, 1996) The Interaction Hypothesis proposes that input and output modified through negotiation provide the best data for acquisition. Long (1996) argued that it was the negotiation of meaning that is important. However, other researchers (e.g. Lyster and Ranta 1997) have argued that negotiation of form (i.e. negotiation triggered by a purely linguistic problem) also assists acquisition. Both the negotiation of meaning and form involve the same set of negotiation strategies which differ in terms of whether they are (1) input-providing or output-prompting and (2) implicit or explicit.
Output Hypothesis (Swain 1985, 1995) The Output Hypothesis claims that acquisition is not just dependent on input but that output—especially pushed output—also plays a role. One way in which output can be ‘pushed’ is when a learner modifies an initially non-target like utterance during negotiation. The later version of the Interaction Hypothesis also allows for this.
Noticing Hypothesis (Schmidt 2001) The Noticing Hypothesis claims that input works best for acquisition if learners pay conscious attention to linguistic forms and the meanings they convey and, in particular, if noticing-the-gap occurs (i.e. learners compare what they have noticed in the input with their own output).
Table 7.2 Summary of key theories in the interaction approach
As this chapter has shown, interaction researchers have drawn on cognitive theories of L2 acquisition. They see input and interaction as influencing the selective attention that learners pay to linguistic features. They also see cognitive systems as influencing how input is processed. In the next chapter we take a close look at cognitive theories in SLA.
Notes
1 Swain’s hypothesis is also sometimes referred to as the Comprehensible Output Hypothesis.
2 Pushed output can be other-initiated, through interaction, or self-initiated when learners monitor their own output and self-correct.
3 I am grateful to Scott Aubrey for allowing
me to use his data.
4 Corrective feedback can occur in both meaning-centred communication when learners perform tasks and in form-centred language exercises. I am only concerned with the former in this chapter.
8
Cognitive aspects of second language acquisition
Introduction
Cognitive SLA draws extensively on cognitive psychology to investigate the internal mechanisms and processes involved in the representation of L2 knowledge and the way in which the knowledge develops over time—i.e. acquisition. Eysenck (2001) described the subject matter of cognitive psychology in this way:
… the subject matter of cognitive psychology consists of the main internal psychological processes that are involved in making sense of the environment and deciding what action might be appropriate. These processes include attention, perception, learning, memory, language, problem solving, reasoning, and thinking. (p. 1).
Cognitive SLA is of course concerned primarily with two of the processes Eysenck mentions—learning and language—but, as we will see, these processes also involve other internal psychological processes. Cognitive SLA, then, is that branch of SLA that examines the mental processes involved in the acquisition of a second language.
Paradigms in cognitive SLA
A paradigm is the set of practices that define a particular approach to investigating a given phenomenon (Kuhn 1962). These practices concern the kinds of questions that are asked, what kind of data needs to be collected to answer them, and how these data will be interpreted. Within cognitive SLA it is possible to distinguish two distinct paradigms: symbolism and connectionism (see Hulstijn 2002).