Making a mess of the evidence on error correction


Earlier this year, Russ Mayne invited Chris Smith to share his thoughts on error correction. On Friday, Russ advised teachers to use this weekend to “do their homework” and read Smith’s post, as if it were an important part of their education. Being a firm believer in taking the weekend off,  my advice is: “Don’t bother!”

Smith reviewed some of the literature on error correction and concluded:

So, when it comes to evidence based EFL, we can conclude that the evidence shows that error correction works. I would also assert that if people want to argue that it does not work, they cannot merely cherry pick one or two articles that did not find a link. They would need to show why all the clear evidence mentioned above (and more) is wrong. So going back to Krashen and Terrell, they asserted that EC is useless, and this idea has been dogmatically perpetuated. However, this is demonstrably wrong. The evidence shows that EC clearly is effective.

I’d like to make three points:

  1. Unless Chris Smith tells us what kind of error correction works with what kind of error, his claim that “error correction works” has little force.
  2. Smith states that anyone who disagrees with his claims for error correction has to show why all the evidence mentioned is wrong. This is based on a misunderstanding of the role of evidence in SLA research.
  3. Smith fails to distinguish between different views on error correction, which stem from different views of second language learning. By presenting the argument as a binary choice (for or against error correction), Jones misrepresents the scholarly discussion of error correction that’s taken place in the last twenty years, and misinterprets the evidence.


As to the first point, in order to investigate the question of whether or not error correction for speaking works, we need to define “error correction” and agree on what counts as evidence that “it works”. Smith claims that evidence from 6 different studies supports his poorly-articulated assertion, but there’s good reason to question this claim.

Study 1

The first study cited by Smith is by Lightbown and Spada (1990). This study gave no indication of what type of error correction leads to what kind of improvement in the participants’ speaking, and furthermore, as Lyster and Ranta (1997) point out, it

examined the effect of a combination of …. both form-focused instructional materials and feedback on error, and thus shed no light on the effectiveness of error correction on its own. (Emphasis added.)

Studies 2 & 3

The second study Smith mentions, by Carroll, Roberge and Swain (1992) and the third, by Carroll and Swain (1993) have methodological problems, as the authors admit, that make generaslisation questionable. In the case of the 1993 study, for example, the time between initial and final testing was 1 week, far too short to know if learning was retained. Apart from these shortcomings, the studies only look at learning a particular kind of grammatical generalisation. Lyster and Ranta (1997) comment:

In Carroll, Roberge, and Swain (1992), adult subjects were trained and given feedback (or not) on two rules of suffixation in French, whereas Carroll and Swain (1993) investigated the effect of different types of feedback on the learning of the dative alternation rule in English, also by adults. It is difficult to know just what relevance the findings of these studies have for the treatment of learner errors during communicative interaction in school settings, particularly with younger learners.

The 1993 article itself ends

Despite these limitations, we tentatively conclude that our study lends support to Schacter’s claim that indirect and direct corrective feedback can help adult second language learners learn abstract linguistic generalisations.

This modest claim for the study falls far short of the claim Smith makes for it.

Study 4

The fourth study by Lyster and Ranta (1997) looks at four different types of error correction and concludes that

 “the feedback-uptake sequence engages students more actively when there is negotiation of form, that is, when the correct form is not provided to the students—as it is in recasts and explicit correction—and when signals are provided to the learner that assist in the reformulation of the erroneous utterance” (emphasis added).

Thus Chris Smith’s summary, which says that the data suggests that “explicit EC was more effective than implicit EC” is inaccurate. Furthermore, the article concludes

Providing feedback as part of a negotiated sequence in this way, however, is of course feasible in L2 classrooms only where learners already possess an adequate level of proficiency. If this condition is met, then the four feedback types that serve to actively engage learners in the negotiation of form remain non-threatening and potentially useful.

Again, the claim that, providing a certain very constraining condition is met, then 4 types of feedback are “potentially useful”, falls far short of the claim Jones makes.

Studies 5 & 6

Loewen’s (2005) analysis of 17 hours of classroom interaction concludes, as Jones says, that “incidental focus on form does have some effect on L2 learning” (p381). Once again, this is a long way from concluding that “error correction clearly works”.

The Ellis, Loewen and Erlam (2006) study did, as Jones says, find that explicit feedback was more effective than implicit feedback and that the benefits became more evident over time. But Jones neglects to mention works which question these findings, including Li’s (2010) meta-analysis of 33 studies, which found that the longer-term effect of recasts was larger and more effective than explicit feedback, and Long’s (2015) summary of recasts, which cites Li (2010) and concludes that recasts are a better form of oral error correction for classroom based SLA than explicit error correction.



The important thing to note here is that there is a major disagreement among scholars of instructed SLA about the relative merits of implicit and explicit oral correction. Smith rightly notes that more studies these days prefer explicit error correction, but he doesn’t acknowledge that these studies represent work done by a group of scholars who all share the same view of instructed SLA; namely that there is a “strong interface” between explicit and implicit knowledge. Likewise, Smith does not acknowledge that studies which give evidence for the superiority of implicit correction are carried out by those taking the “weak interface”, and “no inteface” position. Instead of dealing with the fact that there are important disagreements among scholars about the quality of the evidence regarding error correction and the conclusions we can draw from it, Smith presents the argument as being between those, like Krashen, who argue that error correction “doesn’t work” and those like him, who think it “clearly works”. This allows Smith to lump all the evidence showing some effect for error correction together; which amounts to deliberately misinterpreting the evidence, a far worse academic sin than cherry picking.


The Role of Evidence

Which brings us to Smith’s assertion that if people want to argue that error correction does not work, “they cannot merely cherry pick one or two articles that did not find a link. They would need to show why all the clear evidence mentioned above (and more) is wrong”. Well, no, they wouldn’t, because evidence doesn’t work like that. Empirical evidence is used to support or to challenge a theory or hypothesis, which in turn tries to answer a question or explain a problem that we’ve articulated.

In the case of this issue, the question is “What is the effect of error correction on instructed SLA?” How you answer this question depends on how you see the process of instructed SLA, which in turn involves how you see the interface (the connection or overlap) between explicit and implicit knowledge. As Jones indicates, the non-interface position is the one taken by Krashen (although he has modified this position a bit in the last 20 years), but this leaves importance differences between the weak-interface the strong- interface positions.

Strong Interface

The strong-interface position is the one that Smith seems to take, and is based on skill acquisition theory as applied to SLA. As Han and Finneran (2014, p.317) say:

It suggests that language learning, consists of, and proceeds through, a series of stages: a declarative stage, where learners first accumulate a factual understanding (developing ‘knowledge that’); then, a procedural stage, where learners act on the declarative knowledge (developing ‘knowledge how’); and finally, a stage of automatization, where the procedural knowledge becomes fluent, spontaneous, and effortless.

Han and Finneran (2014) give the example of Schmidt’s Noticing Hypothesis (Schmidt 1990; 1995; 2001) and the claim that ‘you can’t learn a foreign language (or anything else, for that matter) through subliminal perception,’ implying that conscious attention is the only viable pathway to learning, everything else ensuing as its spinoffs.

Weak Interface

Han and Finneran (2014) describe two variants of the weak-interface position. First, R. Ellis (1994; 2005; 2006) argues that

“explicit knowledge can turn into implicit knowledge, contingent on the nature of grammatical elements – whether they are developmentally constrained or not. Explicit knowledge of developmental elements can become implicit only when learners are developmentally ready, while explicit knowledge of non-developmental or so-called variational elements can turn into implicit knowledge at any time” (Han and Finneran 2014, p. 318).

Nick Ellis (2005; 2006; 2007), on the other hand, holds a much “weaker” position and sees learning as a largely implicit, associative and rational process,

whereby learners intuitively identify and organize constructions or form-function mappings based on their probabilistic encounters with relevant exemplars in the communicative environment.

However, the L1 often interferes with the learner’s processing of L2 input, and explicit instruction can help fix the problems, if the instruction

involves the learner in a conscious tension between the conflicting forces of their current interlanguage (IL) productions and the evidence of feedback, either linguistic, pragmatic, or metalinguistic, that allows socially scaffolded development’ (N. Ellis 2007: 84, cited in Han and Finneran, 2014, p. 321).

Nick Ellis’ view is one attempt to articulate an emegentist, or ‘associative- cognitive’, or ‘connectionist’ theory of SLA, based on a usage-based view of language and language development.

We may easily note that the two weak-interface positions are qualitatively different and rest on opposing theories of SLA. One position considers the learning process largely explicit and the other largely implicit; in one, explicit knowledge is necessary, in the other it is ancillary.


As for the non-interface position, Krashen (1982) differentiates between consciously learned knowledge (explicit knowledge) and subconsciously acquired knowledge (implicit knowledge), and argues learned knowledge and acquired knowledge are dissimilar, separate, and mutually irreplaceable. This view rests on Chomsky’s generative view of language development, where competence is enabled by Universal Grammar interacting with L2 natural input; instruction, by implication, may help performance only under monitored conditions, but it does not help competence or implicit knowledge, which drives spontaneous performance.



I’m not going to argue for any particular view here, suffice it to say that Chris Smith fails to recognise the complexity of the issue of error correction, fails to make his own position clear, and fails to appreciate that arguments about the effectiveness of different kinds of error correction stem from arguments about different theories of SLA. Thus, he is wrong to demand that people who want to argue that error correction does not work need to show “why all the evidence mentioned above (and more) is wrong”.  Unless the evidence is faulty (and not all of it is), we can’t show that it’s wrong, but that doesn’t mean we are bound to accept the claims that Jones makes for it. In general, when we look at evidence, we don’t ask “Is it right or wrong?”, we ask “Does it support or challenge a particular theory or hypothesis?”

All of which suggests that Chris Smith’s claim for error correction is both simplistic and misleading. The evidence does not in fact show that “error correction works”, rather conflicting evidence from different studies gives different amounts of support to different, often contradictory claims, which flow from different, often contradictory hypotheses and theories of SLA. The important question is “What kind and degree of error correction works best with what kinds of errors?”; and there is little likelihood that those who take different positions on the interface between explicit and implicit knowledge will agree, because they are informed by conflicting explanations of SLA.


Carroll, S. and Swain, M. (1993) Explicit and Implicit Negative Feedback: An Empirical Study of the Learning of Linguistic Generalizations. Studies in Second Language Acquisition 15, 357-386.

Carroll, S. and Swain, M., and Roberge, Y. (1992). The role of feedback in adult second language acquisition: Error correction and morphological generalizations. Applied Psycholinguistics 13, no. 2 173-198.

Ellis, R. Loewen, S. and Erlam, R. (2006). Implicit and Explicit Corrective Feedback and the Acquisition of L2 Grammar. Studies in SLA, 28,2.

Han, Z. and Finneran, R. (2014) Re-engaging the interface debate:strong, weak,none, or all?International Journal of Applied Linguistics,24. 3.

Krashen, S. and Terrell, T. (1988) The Natural Approach: Language acquisition in the classroom. Hemel Hempstead, Prentice Hall.

Li, S. (2010) The effectiveness of corrective feedback in SLA: A meta-analysis. Language Learner, 60, 2.

Lightbown P. and Spada, N. (1990) Focus on Form and Corrective Feedback in Communicative Language Teaching. Studies in Second Language Acquisition 12, 429-448.

Loewen, S. (2005). Incidental focus on form and second language learning. Studies in Second Language Acquisition, 27(03), 361-386.

Long, M. (2015) TBLT and SLA.  Oxford, Wiley.

Lyster, R. and Ranta. L. (1997). Corrective Feedback and Learner Uptake: Negotiation of form in communicative classrooms. Studies in Second Language Acquisition, 20, 37-66.



8 thoughts on “Making a mess of the evidence on error correction

  1. hi Geoff
    slightly off the main topic of error correction there is a nice recent article by VanPatten from the non-interface position Why Explicit Knowledge Cannot Become Implicit Knowledge []

    Liked by 1 person

      1. it’s good, makes a great point about explicit/implicit discourse often missing talking about content of such knowledge rather than process of such knowledge


  2. Hi Geoff

    Interesting piece. On your first criticism of my article I would largely agree with you. It is an overgeneralisation to talk about ‘error correction’ in general without specifying the type of error (grammar/ pron/ lexis) level or context, although many other writers, for example Krashen and Ellis, have done the same. But yes, a less generalised approach could be taken, exploring different contexts and types of error. That would require a much longer article than the short blog post I wrote.

    Second, you say “Smith states that anyone who disagrees with his claims for error correction has to show why all the evidence mentioned is wrong.” No Geoff – you misunderstand my point. What I mean is that people who want to claim error correction is ineffective cannot merely use one or 2 confirmatory articles to suport their position. They would need to engage with the contradictory evidence rather than ignoring it.

    You also accuse me of “deliberately misinterpreting the evidence”. I think you have completely missed the point of my original blog post. It was only intended to be a quick snapshot of some empirical research that suggests error corrective can be effective. But what you have done in your article is to suggest the brevity is representative of failure and deliberate misrepresentation. This is firstly grossly unfair and secondly, since logical fallacies seem to be a theme, is an example of a straw man argument.


    1. On Point 1, I think we agree, except that surely it doesn’t take a “longer article” to avoid the error of overgeneralisation.

      On Point 2, You said “anyone who disagrees with his claims for error correction has to show why all the evidence mentioned is wrong.” That’s what you said. I didn’t “misunderstand”, you made an absurd claim which I objected to. What you now say you meant to say is inoffensive and obvious – hardly worth saying.

      On Point 3, I didn’t miss the point: I argued that your “quick snapshot of some empirical research” was a lazy, inaccurate and misleading summary of the evidence.

      Nothing you’ve said addresses the weaknesses of the post you wrote: nothing!


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s