In 1997 Language Teaching published an article I had written entitled ‘Form-focused instruction and second language acquisition: A review of classroom and laboratory research’. It was organized around seven research questions and my goal was to review the results of studies that had investigated the questions to determine if we had any firm answers to them. The research questions were:

1. 1. Does second language instruction make a difference?

2. 2. Does type of instruction make a difference?

3. 3. Is form-focused instruction beneficial to second language acquisition (SLA)?

4. 4. Are particular types of form-focused instruction more beneficial than others?

5. 5. Is there an optimal time to provide form-focused instruction?

6. 6. Are particular language features more affected by form-focused instruction than others?

7. 7. Do particular learners benefit more from form-focused instruction?

When I was invited to give a plenary at the Canadian Association of Applied Linguistics 2009 conference it was suggested that I take a look back at that review paper in order to reflect on the research that had been done at that time, how it has evolved since and what the future holds for research in instructed SLA. Initially I had enthusiastically embraced this idea, but when I read my paper I realized that most of these questions remained unanswered twelve years later, making it seem like there had been no progress in the intervening years! Yet I knew this was not quite right – that there had in fact been quite a bit of progress. I also reminded myself that the research undertaking is a long and complex process and that it can take many years (more often decades) of systematic research to obtain answers to questions. But not only does it take time to get answers, it also takes time to ask the right questions. In fact, the subtle questions often only become apparent when you work with a problem over a long period, and eventually realize that the questions you started off asking may not have been the ones to ask at all. As Paul Meara writes in his recent book describing the process of doing research on vocabulary acquisition, ‘the longer you work in an area, and the more you worry its basic assumptions, the more likely you are to find the critical questions that are really worth answering’ (Meara Reference Meara2009: xiv).

So I would like to begin by taking a closer look at the research questions posed in my 1997 paper. Perhaps the first thing to notice is that there is tendency for them to move from the more general to the more specific. For example, question 1, ‘Does second language instruction make a difference?’, leads to question 2, ‘Does type of instruction make a difference?’, and question 3, ‘Is form-focused instruction beneficial to SLA?’, leads to question 4, ‘Are particular types of form-focused instruction more beneficial than others?’. Sometimes when I consider these questions I am reminded of Russian nesting dolls – the big doll opening up to reveal a smaller one and the smaller doll opening up to reveal an even smaller one, and so on. This image is useful when thinking about how research questions about the role of instruction in SLA have become increasingly precise and finely-tuned over the years, and how this has enabled researchers to undertake more focused research in their investigations of them. For the purpose of my presentation today, I have decided to focus only on the last five questions listed above (i.e. 3–7). I will also refer to one or two other questions that have emerged as important ones and are motivating current research in instructed SLA.

First, it is important to state how I defined form-focused instruction (FFI) in my 1997 article: ‘FFI is any pedagogical effort which is used to draw the learners’ attention to form either implicitly or explicitly . . . within meaning-based approaches to L2 instruction [and] in which a focus on language is provided in either spontaneous or predetermined ways’ (Spada Reference Spada1997: 73). This definition differs from Long's (Reference Long, de Bot, Coste, Ginsberg and Kramsch1991) conceptualization of focus on form, which is restricted to spontaneous language use. It also differs from Ellis’ (Reference Ellis2001) definition of form-focused instruction – which is broader in scope – to include instruction that focuses exclusively on forms (i.e. focus on formS). Now, let's begin with the first question.

Is form-focused instruction beneficial to SLA?

In my 1997 review of research on FFI, I identified 30 studies that had been conducted either in the classroom or the laboratory setting. Of these, 23 were relevant to the first question. Eighteen of the studies were experimental/quasi-experimental and five were descriptive in nature. In all studies, the comparison was between classes/groups in which FFI was provided and classes/groups in which it was not. That is, all classes/groups had received communicative instruction but some of them received exclusively meaning-based teaching (i.e. no FFI) and others received attention to language form(s) in their communicative instruction (i.e. FFI). In 1997 the findings related to the first question revealed mixed results. Some studies indicated benefits for FFI (White et al. Reference White, Spada, Lightbown and Ranta1991; Spada & Lightbown Reference Spada and Lightbown1993; Lyster Reference Lyster1994) while others did not (White Reference White, Doughty and Williams1998). Also, some studies revealed benefits for FFI in the short term but not in the long term (Harley Reference Harley1989; White Reference White1991). Others indicated benefits for FFI on written but not oral tasks (Day & Shapson Reference Day and Shapson1991). At that time, the best explanation for these mixed results appeared to be related to the type of instruction provided in the classes. Subsequent research (e.g. Norris & Ortega Reference Norris and Ortega2000) provided additional support for this interpretation, which leads me to the second question.

Are particular types of form-focused instruction more beneficial than others?

A closer analysis of the instructional input provided in the studies revealed that the FFI was more explicit in some studies, that is, teachers provided overt instruction and corrective feedback, including the use of meta-language and clear signals to the learners that there was a right and a wrong way to say/write something. In other studies, however, the FFI was provided in more subtle ways: for example, by providing learners with high frequency exposure to a particular language feature but not drawing their attention to it in any explicit way (Trahey & White Reference Trahey and White1993; Trahey Reference Trahey1996) or by drawing learners’ attention to a target form through typographical enhancement but not providing any explicit instruction or corrective feedback (White Reference White, Doughty and Williams1998). When I examined these studies more closely for type of instruction and compared those that had provided more or less explicit FFI, I concluded that explicit FFI was more effective than implicit FFI in communicative and content-based classrooms.

While I was fairly confident with the conclusion I had drawn based on my review of the studies at that time, I was acutely aware of the pitfalls of the methodology that I had used to arrive at this conclusion. My 1997 article is an example of what is referred to as a narrative synthesis – a review which is carried out by examining the findings of the existing literature focused on the same research question and comparing them to see if the findings are similar or different. There are several problems associated with narrative syntheses, however. First, there is a tendency to select the studies under consideration (i.e. primary studies) somewhat unsystematically and idiosyncratically. The process I used in 1997 was to select papers that I was aware of and those that my students were reading about in FFI and L2 learning. I also consulted the references in those papers and tracked down additional studies. This process is considered to be incomplete and biased, however, because of the observed tendency for researchers to give more attention to studies that report results that are similar to their own and to ignore or understate the findings from articles reporting different results. This is related to the third problem associated with narrative syntheses: that conclusions are often made on the basis of theoretical orientation rather than inspection of the evidence that is compiled systematically across studies (Norris & Ortega Reference Norris, Ortega, Norris and Ortega2006a).

While I was aware of these problems in 1997, I was not familiar with other procedures that could be used to carry out a more complete, systematic and unbiased review until the publication of the Norris & Ortega (Reference Norris and Ortega2000) meta-analysis of the effects of instruction on L2 learning. As most of you know, a meta-analysis allows for cross-study comparisons of quantitative research even when the methodologies differ widely. This is accomplished by using a statistic referred to as an effect size (e.g. Cohen's d): a standard statistical measurement of the magnitude of a treatment effect. Effect sizes are calculated for individual studies and because they are a standardized measurement, they can be combined to find the average effect size for many studies investigating the same question. Therefore, the results are more precise and, most importantly, replicable and verifiable. In the Norris & Ortega (Reference Norris and Ortega2000) meta-analysis, 49 studies of instructed SLA provided enough data for effect sizes to be calculated and compared. This more than doubled the number of studies that I compared in my 1997 review. The overall results of their meta-analysis indicated that explicit instruction is more effective than implicit instruction and that the effects of instruction are durable. Thus, their results were compatible with my results, albeit obtained in a much more systematic and rigorous manner. Later in my presentation I will refer to studies that have been carried out since the Norris & Ortega meta-analysis and discuss how type of instruction has been examined in interaction with other variables including learner factors and type of language features.

There were also other questions that emerged from the Norris & Ortega meta-analysis which are critical, and which have helped to lay the groundwork for current FFI research. These questions are related to the observation that the advantages of explicit instruction may be due to the fact that the majority of tests to measure learners’ progress in instructed SLA research are tests of explicit knowledge. In fact, 90% of the outcome measures in the primary studies included in the Norris & Ortega meta-analysis used highly constrained, discrete-focus linguistic tasks, whereas only 10% required extended communicative use of the L2. This creates an obvious bias and has opened up a long-standing question: ‘What type of knowledge results from L2 instruction?’. Some argue that instruction contributes only to explicit knowledge – the kind of knowledge that serves L2 learners well on grammar tests – but not to implicit knowledge – the kind of spontaneous fluent ability that serves L2 learners well on communicative language tasks (Schwartz Reference Schwartz1993; Krashen Reference Krashen and Ellis1994; Ellis Reference Ellis2005). As it happens, this question was not included in the long list that I posed in my 1997 article, but it has since become one of the central questions occupying the efforts of many instructed SLA researchers, and I will return to it later on in my presentation. But let's move now to the next question addressed in my 1997 paper.

Is there an optimal time to provide FFI?

According to Lightbown (Reference Lightbown, Doughty and Williams1998) there are at least two ways in which we can think about the timing of form-focused instruction. One is psycholinguistic – the right time in a learner's development – and the other is pedagogic – the right time in the instructional sequence. In 1997 I was able to compare only six studies that had investigated the psycholinguistic timing question, and the findings were mixed. While there was some support for the claim that targeting the next stage in a learner's development was effective (Pienemann Reference Pienemann1989; Spada & Lightbown Reference Spada and Lightbown1993; Mackey & Philp Reference Mackey and Philp1998), this research also revealed that learners’ first language and L2 proficiency level interacted with developmental readiness to produce varying results. Furthermore, other research revealed just the opposite: that it is not necessary to target instruction to the next stage in a learners’ development. Instead, instruction can lead to learner progress even when it is targeted to language that is several stages beyond the learner's current level of development (Gass Reference Gass, Hines and Rutherford1982; Eckman, Bell & Nelson Reference Eckman, Bell and Nelson1988; Hamilton Reference Hamilton1994). Again, the L1 background of learners has proved to be an influential variable in this research (Ammar & Lightbown Reference Ammar, Lightbown, Housen and Pierrard2005). Since the late 1990s there have been, to my knowledge, no published studies that have directly investigated the psycholinguistic timing issue. This may be related to some of the challenges associated with this line of research, particularly with reference to questions as to how one might usefully apply these findings in classrooms with learners who are typically at different stages of L2 development.

Let's now turn our attention to the pedagogical aspect of the timing question – whether there is a better time in the instructional sequence to draw learners’ attention to form. No research had directly investigated this question when I wrote my 1997 review, despite lots of lively debate about this question in the pedagogical literature. Since then a few studies have been done that are related to this issue (e.g. Gass & Alvarez Torres Reference Gass and Torres2005; Hondo Reference Hondo2009), but to my knowledge no research has been carried out in which the timing of FFI has been manipulated in instructional materials and its effects examined in relation to L2 learning. Recently, Patsy Lightbown and I conceptualized two types of FFI that are directly related to the pedagogical timing issue: integrated and isolated FFI (Spada & Lightbown Reference Spada and Lightbown2008). Both types of instruction assume a primary focus on meaning with the inclusion of attention to form, but they differ in terms of when attention to form is provided. In integrated FFI, the learner's attention is always drawn to form within communicative practice and activities. In isolated FFI, the learner's attention is always drawn to form separately from communicative practice and activities.Footnote ¹ Over the past two years I have carried out research on the effects of integrated and isolated FFI on L2 learning with my research group at the University of Toronto. One of the central questions motivating our research is whether isolated and integrated FFI contribute to different types of L2 knowledge. For example, if teachers/textbooks draw learners’ attention to a particular language form separately from communicative practice, does this mean that learners will be more likely to accurately recognize and use the target form on isolated grammar tests but not on communicative tasks? Similarly, will learners whose teachers/textbooks draw their attention to a specific form within meaning-based practice be more likely to use that form accurately and spontaneously on communicative tasks, but not do as well on discrete-point grammar tests? These questions are related to a cognitive theory of learning referred to as transfer appropriate processing, which claims that we retrieve knowledge best in contexts that are similar to those in which we originally acquired it. The greater the similarity between the processing types that were activated when we learned something and those engaged in our later attempts to retrieve it, the greater the likelihood of quick and accurate retrieval (Morris, Bransford & Franks Reference Morris, Bransford and Franks1977; Blaxton Reference Blaxton1989).

We investigated the effects of isolated and integrated FFI in a quasi-experimental study with adult ESL learners (Spada et al. Reference Spada, Tomita, Shiu and Yalcin2010). The target structure was the passive construction, and learners were pre-tested on their knowledge of the structure before instruction, post-tested immediately after instruction and again three weeks later. The tests included two written measures (error correction task and cloze task) that were considered to be measures of learners’ conscious, analyzed L2 knowledge, and one oral measure (picture-cued description task) that was considered to be a measure of learners’ spontaneous, non-analyzed L2 ability. Consistent with transfer appropriate processing theory, our predictions were that learners who received integrated FFI would perform better on the oral task and that learners who received isolated FFI would perform better on the error correction task. Using multi-leveling modeling (Raudenbush & Bryk Reference Raudenbush and Bryk2002), the analyses indicated that learners in both groups improved on all three language measures over time. Furthermore, there appeared to be no advantage for the learners who received isolated FFI on the error correction test. However, there were marginal advantages for learners who received integrated FFI on the oral production task. This latter finding is consistent with our hypothesis, but because the differences are not great, no firm conclusions can be drawn. One of the questions that emerged from this research is whether the target feature itself may play a role in whether one type of instruction is more effective than another. This was the next question that I addressed in my 1997 review.

Are particular language features more affected by FFI than others?

This question is of considerable interest to L2 classroom researchers and teachers, yet little research has been done to compare the effects of type of instruction on type of language form. Some researchers have argued that simple features can be taught but that complex ones are probably best learned implicitly (Krashen Reference Krashen1982; Reber Reference Reber1989), via exposure to language. Others have claimed the opposite, arguing that complex features are difficult to notice in naturally occurring input and therefore explicit instruction is necessary (Hulstijn & de Graaff Reference Hulstijn and de Graaff1994). In 1997, four studies had directly examined this question – three in the laboratory setting (DeKeyser Reference DeKeyser1995; Robinson Reference Robinson1996; de Graaff Reference de Graaff1997) and one in the classroom (Williams & Evans Reference Williams, Evans, Doughty and Williams1998). Since then, only a few studies have investigated this question (e.g. Housen, Pierrard & Van Daele Reference Housen, Pierrard, Van Daele, Housen and Pierrard2005) and the overall results have been mixed. One explanation is that complex/simple and easy/hard have been defined differently across studies and often refer to different things, for example, characteristics of the language feature versus aspects of the learning process (see DeKeyser Reference DeKeyser2005 and Spada & Tomita Reference Spada and Tomita2010 for more details). In my research group there has been quite a bit of interest in this question, particularly as we try to make decisions about what language features to focus on in our instructed SLA studies. This led to a meta-analysis in which we used the same criteria to define simple and complex features across a large number of studies and compared the effects of type of instruction on type of language feature (Spada & Tomita Reference Spada and Tomita2010). We were able to include 41 studies in our meta-analysis, and while many of these were the same ones used in the Norris & Ortega (Reference Norris and Ortega2000) meta-analysis, several of the studies were different because our meta-analysis was conducted eight years later. In conducting our meta-analysis the most challenging task was deciding on a definition for ‘simple’ and ‘complex’ to categorize the target features investigated in the primary studies. After much reading and discussion, we decided to go with the theoretical definition provided by Hulstijn and de Graaff (Reference Hulstijn and de Graaff1994), which is based on the number of transformations applied to arrive at the correct form. Table 1 presents two forms: ‘Wh-questions as object of preposition’ and the ‘Past tense’ to illustrate why one is considered complex and the other simple. To produce the example sentence Who did you talk to? one must apply seven criteria, whereas the formation of the past tense of regular verbs requires only one transformation. To distinguish between the features in the primary studies we decided that at least four transformations were required for a feature to be considered complex; anything below that was coded as simple. Our categorization of explicit and implicit teaching followed Norris & Ortega (Reference Norris and Ortega2000), in that instruction was considered to be explicit if it included linguistic rule explanation, and the learners’ attention was mainly on forms. Instruction was coded as implicit if there was no rule explanation or if students were not asked to directly draw their attention to forms (see Spada & Tomita Reference Spada and Tomita2010 for details about the coding and analyses).

Table 1

Number of transformations – complex and simple rules.

The research questions explored in our meta-analysis were: (i) Do the effects of explicit and implicit instruction vary with simple and complex features? (ii) Do the effects of explicit and implicit instruction lead to similar types of language knowledge for complex and simple features? The findings revealed that explicit FFI is more effective than implicit FFI on simple and complex features in both the short and long term. But some of you may be thinking: could the same bias be at play in these results as with previous research syntheses? That is, was there an advantage for explicit instruction because the tests used to measure learners’ progress were explicit measures of L2 knowledge (i.e. grammar tests)? Interestingly, this does not appear to be the case. As I mentioned earlier, even though many of the studies examined in our meta-analysis were the same as those used in Norris & Ortega (Reference Norris and Ortega2000), we included twelve new studies that were published after their cut-off point (i.e.1998). Of these, eight included free outcome measures to examine learners’ progress, such as picture-cued story-telling and description tasks. Furthermore, the results from our meta-analysis indicated that explicit FFI contributed not only to learners’ conscious knowledge of the target forms but to their ability to use the forms in unanalyzed spontaneous ways. But to return to the main finding of this meta-analysis: type of instruction does not interact with type of language feature. However, an important caveat is in order: if we had used a different definition of ‘simple’ and ‘complex’, would the results have been the same? For example, we could have used a psycholinguistic definition and categorized easy forms as ones learned early in the learners’ development and difficult ones as learned late. Alternatively, we could have used salience as a criterion, and described forms that are difficult to perceive in the input (e.g. third person singular -s in English) as complex, compared with forms that are easier to perceive in the stream of speech (i.e. progressive -ing, which we would describe as simple). Another approach could have included L1 influence as a distinguishing characteristic, categorizing forms that are more similar to learners’ L1 as simple and those that operate differently from learners’ L1 as complex. While the use of other criteria might lead to different results, this would not take care of another issue raised by DeKeyser (Reference DeKeyser2005), which is that objective measures of difficulty do not take into consideration characteristics of the learner, and the fact that what may be difficult for one learner may not be difficult for another learner. This is related to such factors as language learning aptitude and language learning experience, which lead me to the final question I addressed in my 1997 paper.

Do particular learners benefit more from FFI?

Of all the questions I posed in my 1997 paper, this was the most under-investigated, and it remains so today. While there has been extensive research on individual differences and SLA on the one hand, and considerable research on the effects of instruction on SLA on the other, there has been little research on the interaction between individual and instructional variables and their combined effects on learning outcomes. In 1997 there were just a few studies that had examined individual learner factors in relation to different learning/teaching conditions, all focused on aptitude and taking place in laboratory settings. They include two studies that investigated learner aptitude under different learning conditions (e.g. explicit and implicit), with both reporting benefits for aptitude in different learning contexts (Robinson Reference Robinson and Schmidt1995; de Graaff Reference de Graaff1997). More recently, learner aptitude has been investigated within the context of communicative and content-based instruction (Harley & Hart Reference Harley and Hart1997; Ranta Reference Ranta and Robinson2002) and found to be as relevant a factor in these contexts as in more traditional approaches to L2 teaching. Some of this research has revealed that different components of aptitude play a varying role in L2 learning success, depending on the learners’ age. For example, in the Harley & Hart (Reference Harley and Hart1997) study, learners in early French immersion programs depended more on memory components of aptitude to be successful learners, whereas learners in late French immersion programs depended more on analytic ability.

One of the earliest classroom studies to investigate the interaction between individual differences and L2 teaching approaches was carried out by Wesche (Reference Wesche and Diller1981). She investigated whether matching learners’ individual learning styles (including aptitude) with compatible instructional approaches would enhance L2 learning. She found that it did. More recent studies include Erlam (Reference Erlam2005) and Sheen (Reference Sheen2007), which also report that aptitude is a predictor of L2 learning in most (not all) instructional contexts. However, the limited number of studies that have investigated aptitude/instruction interactions prevent instructed SLA researchers from drawing any firm conclusions. There is a clear need for more research exploring relationships between aptitude (and other individual differences), type of instructional approach and SLA. Theoretical and empirical work in this area has the potential to provide L2 teachers with useful information about designing effective L2 curricula and instructional materials.

As I stated at the beginning of my presentation, the definition of FFI used in my 1997 article included instruction and corrective feedback, and most of the research reviewed in that paper included studies that examined the combined effects of both. I did not examine in detail studies that had investigated the sole effects of corrective feedback on L2 learning. Since then there has been a proliferation of studies, most of them in the laboratory setting, inquiring whether corrective feedback benefits L2 learning and if different types of corrective feedback are more effective than others. Space limitations do not permit me to review that research in any detail, but the overall findings indicate that corrective feedback (CF) is beneficial for L2 learning (Russell & Spada Reference Russell, Spada, Norris and Ortega2006), that explicit CF is more beneficial than implicit CF, particularly in communicative and content-based L2 classrooms (Lyster & Saito Reference Lyster and Saito2010), and that the usefulness of different types of CF depends on contextual, linguistic, and learner variables (Nicholas, Lightbown & Spada Reference Nicholas, Lightbown and Spada2001; Sheen Reference Sheen2007)Footnote ². I will now conclude by responding to the questions posed in my 1997 paper, after re-examining them twelve years later.