2. The corpus revolution

For the corpus revolution to have taken place, corpora would have to be regularly investigated by easily accessible and usable corpus analysis software and seen as a normal everyday part of the language learning environment, in other words they would have to be ‘normalised’, the term used by Bax (Reference Bax2003, Reference Bax2011). ‘A technology has reached its fullest possible effectiveness in language education when it has arrived at the stage of ‘normalisation’, namely when it is used without our being consciously aware of its role as a technology, as a valuable element in the language learning process’ (Reference Bax2003, p. 1). Bax also emphasises that the activity involved in using technology in language learning should never been seen as of interest in itself, but rather that ‘CALL will be normalised when computers are treated as always secondary to learning itself, when the needs of learners will be carefully analysed first of all, and then the computer used to serve those needs.’ (Bax, Reference Bax2003, p. 24). There is evidence of normalisation of the use of corpora in some areas of language teaching and learning. They are increasingly used as the basis for dictionaries, course books, and grammars. Examples are the Oxford-Hachette French-English dictionary, the Touchstone series of English language course books (McCarthy, McCarten, & Sandiford, Reference McCarthy, McCarten and Sandiford2004–2006), and the Cambridge grammar of English (Carter & McCarthy, Reference Carter and McCarthy2006). A good illustration of the successful normalisation of the results of corpus analysis in these resources is that teachers and learners can use them without being aware that the definitions and examples which they contain have been obtained as a result of the compilation and analysis of large general language corpora.

The advantages of providing definitions and examples which correspond to actual language use cannot be overestimated. To give just one example, it is interesting to compare the substantial change in the definitions of the word certes in French in the non-corpus-based Harrap's standard French-English dictionary of 1934 (Example 1) and the corpus-based Oxford-Hachette French dictionary of 1997 (Example 2).

The second example, unlike the first, indicates that certes is used in contexts where the certainty is about to be qualified in some way. [My use of bold highlights the contrast in the two examples.]

The second example is a good illustration of the normalisation of the use of corpora, in that the user of the dictionary may not be aware that the examples have been selected from a corpus of actual language use. However, I can also use a brief anecdote based on the example of certes to illustrate the fact that in many ways the use of corpora is far from normalised in language education outside the case of corpus-based dictionaries, grammars and course books. Responding to a query relating to an essay from a final-year undergraduate student in a class some years ago on the difference between certes and certainement, I made some comments on differences in register between the two words, with certainement being more common in everyday language use, and said I would respond in more detail the following week. I checked the definition in the Oxford-Hachette, given in Example 2, but decided that the examples would not be very relevant in the context of essay writing, so I also loaded a one-million-word corpus of research articles in French (Chambers & Le Baron, Reference Chambers and Le Baron2007) in the concordancer AntConc (Anthony, Reference Anthony2011), and found confirmation of the prevalence of the use of certes in cases where the statement which it accompanied was about to be qualified or even negated. Of the 184 occurrences of certes, it was followed by mais in the same sentence in 79 occurrences, and followed by a sentence beginning with mais in 38 occurrences. In a small number of cases the following sentence began with alternatives to mais, such as cependant, néanmoins or toutefois. In other cases the qualification or negation was expressed by other means, such as negation, an adjective or adverb with a contrasting or negative resonance, or simply by a phrase which diminished the affirmative connotation which earlier lexicographers had clearly associated with certes. In only 15 occurrences could no qualification be observed. I was able to make a selection of these occurrences available to the students and discuss them in the following class in terms of the register in which certes is found and the specific way in which it is commonly used. Some of the occurrences are listed in Example 3.

This example serves to illustrate both the potential and the current limitations of the use of corpora in language learning. While the dictionary definitions provide a clear illustration of the normalisation of the application of corpora in an everyday language learning activity, namely consulting a dictionary, it is not practical at present to expect the consultation of a specialist corpus by the majority of teachers and learners at all levels of language education to be a normalised activity. In this case both the corpus and the concordancer are freely available, but a teacher may not be aware of their existence, or be familiar with the use of the concordancer. Moreover, this example involves only one specialised domain, namely academic writing in the arts and humanities, and only one language. The needs of language teachers and learners would of course cover many domains and languages in addition to academic writing in French.

3. Research on corpora and language learning: in search of replicability

A comprehensive review of research on corpora and language learning published since the mid-1980s is well beyond the scope of this speech, but the research timeline which appeared recently in this journal (Boulton, Reference Boulton2017) provides a useful starting point. The author divides this large body of research into four categories: works on corpus linguistics providing the theoretical underpinnings; descriptive papers, such as Johns's (Reference Johns1986) early description of his experimentation with encouraging learners to explore corpus data in search of answers to their queries; empirical evaluations of DDL, the category which concerns us here; and surveys and syntheses which chart developments in the field (see, for example, Flowerdew, Reference Flowerdew, Lenko-Szymanska and Boulton2015). In relation to the empirical studies of DDL, it is important to note that they have generally provided positive results. As Boulton (Reference Boulton, Jaén, Valverde and Pérez2010, p. 140) points out in a study of learner outcomes, ‘The overall body of empirical research in DDL provides overwhelmingly favourable reactions to DDL on the part of the learners’. Moreover, two meta-analyses have shown than the research reported in empirical studies of DDL has had substantial impact (Cobb & Boulton, Reference Cobb, Boulton, Biber and Reppen2015; Boulton & Cobb, Reference Boulton and Cobb2017).

It may initially seem surprising that the use of corpora by teachers and learners has not been normalised, given the large and increasing amount of research devoted to the subject, the positive results of the studies, and the claims made by the researchers concerning the benefits, both in linguistic and educational contexts, of enabling learners to consult and analyse corpus data as part of the language learning process. In the linguistic context, giving learners access to attested examples of actual language use has become an accepted part of language learning since the communicative approach emphasised the importance of acquiring communicative competence rather than a passive knowledge of vocabulary, grammar and syntax. Using corpus data and the results of corpus analysis of features such as frequency and lexicogrammatical patterns, rather than relying on invented examples or isolated texts, was a logical development when advances in corpus linguistics began to influence linguistic descriptions of language. This has given rise to a significant body of research in applied linguistics emphasising the important role which corpus data and corpus analyses have to play in underpinning what is taught in the language classroom (see, for example, O'Keeffe, McCarthy, & Carter, Reference O'Keeffe, McCarthy and Carter2007). We can also include in this category of linguistic research, studies of the linguistic features characteristic of specific genres and how they function in constructing discourse (for example, Baker & McEnery, Reference Baker and McEnery2015).

Following pioneering work by Johns (Reference Johns1986) reporting on the active, engaged interaction of his learners with corpus data, Tribble and Jones (Reference Tribble and Jones1990) published a guide for teachers suggesting practical ways in which corpus data could be used in the language classroom, and since then, in parallel to the research on the linguistic contribution of corpora to language learning and teaching, another related yet distinct body of research developed, which explored not only the linguistic but also the pedagogic dimension of using this new resource. The continuing popularity of this type of research, namely empirical studies of providing learners with direct access to corpora and corpus analysis tools, or indirect access to concordance examples prepared in advance by the teacher, can perhaps be explained by the fact that it corresponds perfectly with a number of contemporary paradigms in research and practice in education in general, and in language teaching in particular. Constructivist approaches in particular were gaining in popularity just as research on corpora and language learning was developing, and are now generally accepted. Johns (Reference Johns, Wichmann, Fligelstone, McEnery and Knowles1997, p. 101) described the activity of learners who directly access corpus data in search of answers to their queries as ‘research’; ‘every learner a Sherlock Holmes’. This corresponds well with contemporary paradigms in education which focus on discovery learning (Bernardini, Reference Bernardini, Burnard and McEnery2000), and could also be seen as integrating well with developments in second language acquisition research which stress the importance of language awareness (Farr, Reference Farr2008) and of ‘noticing’ (Schmidt, Reference Schmidt1990). The learner who has mastered the use of corpus analysis software and who has access to relevant corpora could decide independently, outside the classroom, what aspects of language use to investigate, thus developing autonomy, another influential concept in language learning research. Finally, at a time when languages for specific purposes was becoming increasingly the focus of attention in teaching and research, the potential of accessing genre-specific corpora seemed to provide the perfect environment. For an early example of concordancing in this context, see Tribble (Reference Tribble1990), and, more recently, Anthony (Reference Anthony, Flowerdew and Costley2016). The spread of the corpus revolution from the realm of dictionaries, course books and grammars to the everyday practice of teachers and active, autonomous learners of general and specialised language use thus seemed a realistic prospect.

It is important to note that the vast majority of the empirical studies of learners using corpora are small-scale studies by one or a small number of researchers, often using corpora which they have created and which are not publicly available. In other words, they involve teacher-researchers with a high level of skills in corpus creation and consultation, who are prepared to devote a lot of time and effort to an activity which will both produce what they believe are worthwhile learning experiences for their learners and at the same time constitute a valid research activity for them, resulting in a research publication. They are also, understandably, mostly situated in the context of higher education, which also limits their replicability across the broad spectrum of language learning and teaching. This situation may well explain to a very large extent the persistence of the research–practice gap.

Before addressing this gap, we will look at two features of the empirical studies of corpora and language learning where the researchers appear to be aware of the gap and to be making an effort to overcome it. Firstly, we will briefly examine the concept of the pedagogic corpus, a corpus created not for linguistic research but specifically for language learning purposes. Secondly, we will briefly examine three studies which investigate the use of such corpora with learners in different contexts, namely academic writing by early career stage researchers (Lee & Swales, Reference Lee and Swales2006); a rare example of corpus use in secondary education, involving general spoken language use (Braun, Reference Braun2007); and more specialised spoken language use by student teachers (Amador, O'Riordan, & Chambers, Reference Amador Moreno, O'Riordan and Chambers2006). Of course this cannot claim to provide a comprehensive evaluation of this body of work, but I have chosen these examples as studies which illustrate in several ways the strengths and the success of DDL. They appear to be making efforts to bridge the gap, while at the same time giving us some hints as to why it persists, despite the excellent fit with current linguistic and educational paradigms which I have outlined above. We will therefore focus in particular on the potential for replicability not only in the sense of being able to reproduce the study by other researchers, but, more importantly here, in the sense that teachers could replicate it or adapt it in their everyday educational practice. As we shall see, a number of factors are involved, including the corpus resources necessary for the activity, the knowledge and skills which the teacher needs to have to prepare the classes and guide the learners, and access to the means of dissemination. Indeed, one of the reasons for the research–practice gap may well be that the means of dissemination of the research, namely the research articles or book chapters, are very often published in collections, such as special issues of journals or edited volumes, which are aimed primarily at other specialists and not at non-specialist practitioners.

In the majority of teaching situations, it is clearly not practicable to expect teachers to create all of their own custom-made resources, and it is therefore not surprising that a number of teacher-researchers have created, either individually or as part of a collaborative project, what are known as pedagogic corpora, or as Braun (Reference Braun2007) terms it ‘pedagogically relevant corpora’. On the basis of the body of research using pedagogic corpora, it is very difficult to provide a definition. Meunier and Gouverneur (Reference Meunier, Gouverneur and Aijmer2009, p. 186) propose: ‘A large enough and representative example of the language, spoken and written, a learner has been or is likely to be exposed to via teaching material, either in the classroom or during self-study activities’, also known as a textbook corpus. The concept of the pedagogic corpus, as a corpus created to meet the needs of learners rather than researchers in linguistics, has been present in research on corpora and language learning from the 1990s (for example, Tribble, Reference Tribble1990). An early example is Tim Johns's three-million-word corpus, which he created ‘to give a rough reflection of an overseas student's life in Birmingham, both on and off the campus’ (Johns, Reference Johns, Wichmann, Fligelstone, McEnery and Knowles1997, p. 103). The corpus includes a scientific research journal, a popular scientific journal, lecture transcripts, newspapers, childcare and household management texts, and ‘one or two “interloper” texts …, for example Frederick Forsyth's The Day of the Jackal to give an occasional glimpse of sex-and-violence’ (Johns, Reference Johns, Wichmann, Fligelstone, McEnery and Knowles1997, p. 104). A contemporary equivalent of this broad definition of the pedagogic corpus is Lextutor,Footnote ¹ the excellent collection of corpora and accompanying language learning activities compiled by Tom Cobb and accompanied by a built-in concordancer, which is extremely easy to use.

This combination of several genres is, however, rare in this body of research, and most pedagogic corpora which have been created specifically for use by language learners concentrate on one genre to meet a specific need which the creator of the corpus has identified, either for the learners of the teacher in question or, in the case of projects involving the creation of freely available online corpora, of a much larger community of learners. Examples of pedagogic corpora which are freely available online include the SACODEYL and Backbone corpora,Footnote ² which consist of interviews with adolescents and adults respectively talking about their lives. These corpora are clearly intended to give learners examples of native-speaker language use to influence their own language learning. It is important to note that not all of the research on corpora and language learning involves small pedagogic corpora. Bernardini (Reference Bernardini, Burnard and McEnery2000) uses the British National Corpus with her learners, what she calls large corpus concordancing. It will be interesting to see if the increasing availability of larger corpora, mostly but not exclusively in English, through sites such as the Brigham Young corpora,Footnote ³ will attract teacher-researchers to experiment with the use of such corpora with their learners. The 2017 special issue of Language Learning & Technology, which focuses on corpora and language learning, includes both studies involving large reference corpora and articles where the authors have chosen to create their own corpora to match the learners’ needs. Using the web as a corpus is also an understandable attempt to bring the activity of corpus consultation close to the everyday practice of the learners. (See, for example, Renouf, Kehoe, & Banerjee, Reference Renouf, Kehoe, Banerjee, Nesselhauf, Hundt and Biewer2007.)Footnote ⁴ However, as pedagogic corpora as defined above are used by the majority of researchers, we will concentrate on them here.

Before examining our three studies, let us look briefly at an example of how one pedagogic resource, SACODEYL can be used with learners. Created as part of a European-funded project led by the University of Tübingen, SACODEYL includes seven corpora consisting of interviews with teenagers in seven European languages. Given the age range of the interviewees, it is clearly intended to be of use at secondary level. The topics, which focus on their studies, interests and future plans, are aligned with the Common European Framework of Reference for Languages (CEFR). In addition to the videos and transcripts, the site includes a simple built-in concordancer, and pedagogic enhancement in the form of exercises. To give a simple illustration of the use of this resource, I consulted the French corpus and searched for the word bon (58 occurrences, 7 by the interviewer and 51 by pupils). I have deliberately avoided using the term ‘adjective’ to describe bon, for a language learner studying the 51 occurrences by the French-speaking teenagers will quickly discover that only in 17 cases is the word used as an adjective, while in the other 34 cases it is used as a discourse marker. Some occurrences of use as a discourse marker are listed in Example 4, with the keyword and one recurring pattern, namely mais bon (18 occurrences) highlighted.

As the example stands it seems devoid of context, but if learners encountered mais bon while listening to part of one interview in class, it could be interesting for the teacher to have prepared a concordance of the other occurrences of mais bon to emphasise that this is a common use of the word alongside its use as an adjective, and perhaps to discuss its use in qualifying what has just been said, and particularly the common practice of using it in sentence-final position. It would be even better if the learners could access the site directly to explore the use of this aspect of the teenagers’ language use. In other words, the site provides not only recorded interviews but also instant access to examples to complement one occurrence in a text, what Cobb (Reference Cobb1997) terms multicontextual learning.

Lee and Swales's (Reference Lee and Swales2006) study is aimed at a much higher level of language learner and more specialised language use. It involves a corpus-informed course in English for academic purposes (EAP) for Ph.D. students and has several positive features in the context of replicability. Firstly, the paper provides a clear account of the structure of the course which includes the corpus consultation, thus corresponding to Bax's comments on the importance of the learners’ needs rather than the use of technology being the primary interest. The very clear description of the structure of the course is summarised below.

1. 1. Induction: specialised corpora of academic writing and speaking, concordancers, ‘inducted into the skills needed to best exploit the data and the tools for directed learning as well as self-learning’.

2. 2. Participants compile two written corpora: their own writing (learner corpus); ‘expert’ writing.

3. 3. Participants are assessed by either a study of expert writing or by comparing their own writing with that of established writers in the field.

4. 4. Assessment consists of either a study of expert writing or a comparison of the novice writer's own writing with the writing of experts.

Although the numbers involved in this study were very small, just four in all, it is interesting to note that the results were very positive. The authors’ reports on the activities carried out by the participants produced interesting and relevant findings about language use in their disciplines. More particularly, in the post-course interviews, participants stated that they found the use of corpora confidence-building and empowering, in that they could check aspects of language use without the need to consult native speakers (Lee & Swales, Reference Lee and Swales2006, p. 71) and that the examples which they found were ‘closer to their contextual and textual circumstances’ than those found in reference books or grammars (Lee & Swales, Reference Lee and Swales2006, p. 71). Finally, the course participants stated that they would consider using corpora in the future.

In terms of replicability by teachers, this study is particularly interesting. Lee and Swales's study initially appears particularly demanding in terms of corpus creation, as it involves the creation of not one but two corpora for each learner, an expert corpus and a learner corpus, made up of the novice researcher's own writing. This problem has been solved by the fact that in this case the learners create their own corpora, and are likely to be highly motivated to do so by the fact that the aim is to enable them to study expert writing and to directly compare it with their own writing. However, Lee and Swales had access to a large corpus of academic writing in several disciplines which was made available to their students and from which they compiled their own corpus. Unfortunately, this is not publicly available, which means that a teacher attempting to replicate the study would have to advise their learners on the creation of discipline-specific corpora. An extensive range of corpora are catalogued in CLARIN, for example, and it is possible for teachers and learners to generate corpora from the web using a tool such as BootCaT (Baroni & Bernardini, Reference Baroni, Bernardini, Lino, Xavier, Ferreira, Costa, Silva, Pereira, Carvalho, Lopes, Catarino and Barros2004) or from classified text collections using software such as AntCorGen,Footnote ⁵ both freely available. This is thus an example of a study which would not require an enormous amount of preparation for a teacher who had the skills necessary to guide the learners in their exploration of the corpus. In other words, it is possible to replicate, but it applies only to a very small, highly qualified and highly motivated group of learners.

A larger and more recent study (Charles, Reference Charles2014) involving graduate students creating a personal corpus of discipline-specific academic writing produced overwhelmingly positive results. Charles examined 40 graduate students’ perceptions of the usefulness of having a personally created corpus one year after they had created the corpus as part of a course in academic writing in English. She found that 35 of the 40 students (86%) either had used or intended to use the corpus (p. 34). Using a five-point scale to evaluate if they thought the corpus consultation had improved their writing, the vast majority responded ‘yes, definitely’ (50%) or ‘yes, probably’ (43%) (p. 35). Unfortunately, however, eight years after Lee and Swales's study, as Charles notes (p. 1), within the body of research on corpus consultation in EAP, the number of studies devoted to personal corpus creation is still very small. In the wider context of research involving corpora and the teaching of academic writing the number of studies is of course much larger, although there is a dearth of studies reporting on corpus use in classes other than those taught by the researchers in corpora and language learning who are writing the articles. It is important to note here that there is also a substantial body of research in the use of corpora in translator education (see, for example, Frankenberg-Garcia, Reference Frankenberg-Garcia2014, Reference Frankenberg-Garcia2015). A detailed study of this specialised area of language teaching is beyond the scope of this study. As the use of corpora in translation software has been common for several decades, it is probable that knowledge of corpora may be more widespread within this specialist community.

Returning to the broader context of language education, Braun's (Reference Braun2007) study of the use of corpus data to enhance general spoken language use is a rare example of an empirical study based in secondary education. It is interesting to note here that there has also been some experimentation at primary level, for example, Sealey and Thompson (Reference Sealey and Thompson2007). Braun's project involves the learning and teaching of English in Germany and uses what was at the time of the writing and publication of her article a freely available corpus, accompanied by transcripts and by supplementary material. ‘I developed a method for the design and exploitation of a pedagogically relevant corpus (Braun, Reference Braun2005) and created a small experimental corpus – ELISA (English Language Interview Corpus as a Second-Language Application) – to exemplify this method. ELISA currently consists of 25 video-based narrative interviews with native speakers of English from different countries and different walks of life.’ (Braun, Reference Braun2007, p. 309). The author begins by emphasising that the corpora and corpus analysis software created for the purpose of linguistic research are not necessarily suitable for teaching purposes.

She explains that ELISA is a pedagogically enriched corpus, intended to be freely accessible and easy to use, as it ‘relies on query methods which do not require much corpus linguistic knowledge. The corpus was pedagogically enriched by ready-made wordlists and concordances, and a set of learning activities for selected interviews.’ (Braun, Reference Braun2007, p. 309). Having created this pedagogic corpus and the accompanying user-friendly resources, she explains that it is important to realise the corpus cannot become the centre of attention in the teaching, but must fit in with the needs of the teaching curriculum and with its time constraints.

The author's account of the integration of corpus work and the textbook in four sessions per week over a four-week period is a rare example of the integration of corpus-related work with an existing textbook. I have added the titles of the interviews in square brackets below.

The article provides a detailed analysis of the learners’ performance and their reactions to the ‘explorative’ (p. 311) corpus-based activities based on the second interview, comparing them with a control group in which the exercises were based on more traditional computer-based activities related to the same interview. Both performance (p. 316) and perceptions (p. 319) of the success of the activity were significantly higher in the corpus group. Braun concludes, however, that the methodological challenge of integrating corpus work in secondary education should not be underestimated. It is interesting to note that, despite the positive results described above this study shares one of the limitations of many other publications reporting experiments involving the use of corpora and language learning in that the targeted journal was aimed specifically at researchers on the subject, as this was published in a special issue. It is, however, as in the case Lee and Swales's (Reference Lee and Swales2006) study, possible for a teacher to replicate in the classroom. Although the original ELISA project website, which included the corpus and the supplementary material, is no longer available, the 25 videos and transcripts of the ELISA corpus are still available. They are now integrated in the Backbone website in a collection of 45 videos and transcripts of international varieties of English, the IVY English corpus. (This can be downloaded freely by clicking on Corpus Search on the Backbone website and following the link to the IVY English corpus).

Our final example (Amador, O'Riordan, & Chambers, Reference Amador Moreno, O'Riordan and Chambers2006) also focusses on teachers in secondary education, but in a different way. In this case the learners are student teachers, some preparing to teach French in secondary education in Ireland, others Spanish, and the corpus is used to provide them with examples of native-speaker teachers teaching through the target language. In other words, the corpus use is integrated into a course which is not aimed at teaching them about corpora. The corpus resources are very small, consisting of just two one-hour classes in each language, audio-recorded and transcribed. For each language, one class involved a native speaker of French/Spanish teaching the language to non-native-speaker pupils in Ireland, while the other involved a native speaker of French/Spanish teaching the language to native-speaker pupils in a country where the language is spoken. The French classes were recorded as part of a larger corpus of 15 classes for both teaching and research purposes, and the Spanish classes were recorded solely to form a resource for teaching purposes. Neither of these is publicly available. Despite this major limitation, the study is included here as it illustrates how a corpus of classroom discourse, if it were publicly available and if used in teacher education as in this case, could not only provide a valuable resource for student teachers preparing to teach through the target language but could also provide the student teachers with a very powerful example of the benefits of having easy access to attested examples of language use as a way of mastering that particular type of language.

In the teacher education course three aspects of the corpus were studied, namely discourse markers, corrective feedback, and learner discourse. In this article (Amador et al., Reference Amador Moreno, O'Riordan and Chambers2006) we focussed only on discourse markers. As these are used to organise language, and as the teacher is the person in charge of the organisation of the class, it was important to give the student teachers the opportunity to observe how the native-speaker teachers used the discourse markers and to reflect on how this could influence their own practice. The analysis of the classes in both languages revealed that the teachers used a limited number of discourse markers, and that they fulfilled a limited number of functions. The Spanish discourse markers, their functions and examples of their use are listed below.

The reactions of the student teachers to their use of the recordings and transcriptions were also examined. They appreciated the opportunity to study how native-speaker teachers taught through the target language, and, interestingly, in relation to pupil talk, they noted the contrast between what they observed in the recordings and the set phrases from a text book which they had been introduced to as part of their course in the previous semester. As one student teacher observed, ‘The stuff we were given for the students to use in the class did not correspond at all to what was in the recordings.’ (Amador et al., Reference Amador Moreno, O'Riordan and Chambers2006, p. 97).

The authors of the two articles examining the use of spoken corpora with learners have shown how simple audio or video texts and transcripts can be helpful for learners. More sophisticated consultation of spoken texts is also possible, for example enabling the learner to hear the audio recording corresponding to a given concordance line, using the concordance WordSmith Tools 5.0 or later versions (Scott, Reference Scott2008). This makes it possible for a prosodic form to be associated with a lexicogrammatical one. Aston (Reference Aston, Lenko-Szymanska and Boulton2015) argues that technology such as this has an important role to play in the context of learning about phraseology. It is tempting to imagine a future scenario where recordings of relevance to language learners at various levels and student teachers can be accessed by users who do not have the expertise of an experienced researcher such as Aston, and where consulting a spoken corpus in this way can be a normalised activity.

Amador et al.’s study shares a number of features with the two previous ones which help to explain the research–practice gap. Firstly, the journal to which it was submitted and where it was published, ReCALL, was aimed at researchers in computer-assisted language learning, and not primarily at language teachers who are not researchers. Secondly, a language teacher educator wishing to replicate the study would have needed corpus consultation and analysis skills, although as the corpus resources used here (as in the case of Braun's study) are very limited in size, the task of corpus analysis would not have presented the same challenge as a larger corpus. In addition, unlike the resources used in Braun's (Reference Braun2007) study, the corpus used in Amador et al.’s (Reference Amador Moreno, O'Riordan and Chambers2006) experiment is not publicly available, with the result that a language teacher educator would have had to record and transcribe a number of classes to be able to replicate the teaching situation. In summary these three studies, in common with many others in this body of research, reveal the enormous potential of corpora in the language teaching context while at the same time helping to explain the research–practice gap. It is important to note, of course, that one cannot in any way blame researchers for publishing in research journals. It is also easy to understand how teachers who are also researchers in corpora and language learning can decide to incorporate corpus-based work in their classes without having the possibility of depositing the corpora they create so that they would be available to teachers wishing to follow their example. However, it seems clear that maintaining the current paradigm in research in corpora and language learning will not on its own solve the research–practice gap. It is tempting to conclude at this point that this body of research in corpora and language learning is at an important turning point. It has laid the groundwork for the elusive corpus revolution, and the research timeline and meta-analyses cited above have provided an overview of what has been achieved. However, to propose a possible solution to the research–practice gap one must at this point look beyond this body of research and these researchers, while in no way excluding them.

4. Bridging the research–practice gap: in search of a new disciplinary framework

The various groups of specialists who could contribute to this development are those who have been mentioned in this paper so far, namely:

1. 1. corpus linguists;

2. 2. applied linguistics researchers who specialise in second language acquisition or computer-assisted language learning, and who provide the theoretical underpinning outlined above (normalisation, authenticity, autonomy etc.);

3. 3. the teacher-researchers who have been carrying out empirical studies in this area for some decades;

4. 4. language teacher educators;

5. 5. materials developers and publishers;

6. 6. and last but certainly not least, teachers and learners.

The work undertaken by the first two categories above is certainly relevant for the theoretical underpinning of developments in this area. Their work is often cited in the publications of the third category, as claims are made that exploring corpus resources promotes studying actual language use, noticing, a constructivist approach to learning, or autonomy. However, they are not in the front line of experimenting with the use of corpus data with learners, so their role in the actual implementation of corpora in the language classroom is of necessity indirect. We have also seen how the third category, the teacher-researchers, have laid the groundwork for more generalised use of corpus data, but cannot, as a group of individuals, be directly responsible for the provision of resources, for the dissemination of information and training directly to teacher educators and teachers. As we shall see, those who have addressed the research–practice gap have seen the fourth and fifth categories named above as having the potential to influence developments more effectively.

Language teacher education has been proposed by a number of researchers as the most suitable context for effecting change in the practice of teachers. Römer (Reference Römer2006, pp. 128–129) describes this using a religious metaphor, ‘missionary work’, ‘spreading the word’. In addition to targeting syllabus designers and materials writers, she suggests that initial teacher training at universities, practising teachers and advanced learners should be targeted. Throughout the development of research in corpora and language learning the point has been made by a number of researchers (for example, Conrad, Reference Conrad2000; Farr, Reference Farr, O'Keeffe and McCarthy2010) that the introduction of corpus analysis in teacher education programmes could play an important role in popularising corpus consultation by language teachers and learners. There have even been a number of studies of the reactions of student teachers and practising teachers to the use of corpora with their learners (for example, Farr, Reference Farr2008; Breyer, Reference Breyer2009; Aljohani, Reference Aljohani2017). Positive reactions included an appreciation of access to actual language use, the ability to focus on areas of difficulty for learners, and the potential to promote autonomy. On the negative side, the student teachers and teachers were still faithful to the primacy of grammatical rules rather than evidence from practice; some even worried that the corpus might contain mistakes. In addition, the variation and messiness of spoken language presented a challenge. Finally, a number of factors inhibiting the integration of corpora in language learning which have already emerged from our brief overview of three studies above, have also been cited in these studies focussing on teachers and student teachers, namely the difficulty of finding suitable corpora (Breyer, Reference Breyer2009, p. 165), lack of expertise in corpus consultation and analysis (Farr, Reference Farr2008, p. 38), and also problems relating to time, technical issues and expense (Breyer, Reference Breyer2009, p. 166; Farr, Reference Farr2008, p. 38). It is important to note that relying on language teacher education programmes to bridge the research–practice gap will not solve all of these issues, such as the lack of suitable corpora and the issue of the time teachers have available. In addition, the availability of colleagues in universities with the expertise necessary to teach the relevant courses in teacher education programmes could be a limiting factor, and produce a situation where only institutions which had both language teacher education programmes and researchers in corpora and language learning could contribute to ‘spreading the word’. Despite these limitations, language teacher education programmes will have an essential role to play if the practice of corpus consultation and analysis is to be normalised in secondary education.

At the beginning of this exploration of the research–practice gap in relation to corpora and language learning and teaching, we saw that language resources such as course books, dictionaries and grammars had enjoyed considerable success in making the transition from a rule-based to an evidence-based approach to the language which is taught. Indeed, in the article cited at the start of this paper, Conrad (Reference Conrad2000) names researchers, materials writers and language teacher educators as the three groups who could bring about the corpus revolution which is the subject of her article. As course books greatly influence what is taught in language classes in secondary education and also to a certain extent in higher education, the question arises if they can also contribute to extending the role of corpora beyond serving to inform the language which is presented to the learners. It has been common practice for course books to include a CD with additional textual material and exercises, and it is now usual for websites to accompany course books. The major publishers who have invested considerable funding in the creation of vast corpora may not want to make them freely available to their competitors, but it is easy to envisage that a small part of these corpora could provide teachers with a valuable teaching resource, accompanied by guidance in the exploration of the corpus in the teaching and learning context. The body of research we have been considering has laid the groundwork for this. This could form a very positive response to Breyer's (Reference Breyer2009, p. 156) comments, in the context of corpora, on ‘the need for more teaching materials with concrete teaching suggestions, and, perhaps more importantly, teaching materials which are integrated into existing textbooks’. While such resources could not be expected to include corpora which would correspond to the needs of learners in all possible specialist genres, the resources provided could also suggest how teachers, or as in the case of Lee and Swales's (Reference Lee and Swales2006) study in higher education, their learners, could create small corpora to meet their needs, which they could then explore using the guidelines provided by the authors of the resources accompanying the course book.

The combined effects of language teacher education programmes and initiatives by major publishers to make part of their corpora available and provide pedagogic enhancement could make a very significant contribution to bridging the research–practice gap. However, it seems to me that these can only be really successful if they are accompanied by another development which could exist in parallel to the current empirical research in corpora and language learning, namely studies of language teachers using corpora in their classes who are not the authors of the research articles, not the specialists in corpus creation and analysis. It is clear that the real motivation in this body of research is the desire to enable teachers and learners to benefit from the resources and methods of corpus linguistics. The starting point in many studies is the learners’ needs, which the experimentation with corpus data strives to meet, but the corpus is usually the dominant, indeed often the only aspect of the course which is presented. This is of course a necessary part of a research project in which the presence or absence of corpus-based work is the variable which is being assessed. Moreover, the teacher-researchers have a high level of expertise in corpus creation and analysis, and are very motivated in their combined roles of teachers and researchers. In other words their situation is very different from that of a practising teacher in secondary or higher education, who will only have very limited time available to integrate corpus-based work with existing activities. We have already seen that a number of researchers have experimented with providing training to teachers or student teachers at secondary level in corpus consultation and analysis, and examined their reactions. Studies such as these, also involving language teachers in higher education who do not specialise in corpus linguistics, with a clear focus on the integration of corpus-based work in the existing curricula, could make a more direct impact on classroom practice than has been the case with research to date. Such integration will be necessary if corpus consultation and analysis are to be normalised.

Major innovations in language learning and teaching have generally not replaced earlier methods, but existed alongside them and have been considered as enhancements. This is particularly true in the case of technology and language learning, where technological resources, course books and face-to-face teaching all co-exist. In other words, the exploration of corpus data would not form a separate activity, but would be an integral part of a lesson. Boulton (Reference Boulton, Jaén, Valverde and Pérez2010, p. 144) underlines the need ‘to take DDL out of the hands of expert teacher-researchers and make it accessible to ordinary learners’. The combination of teacher-researchers and ‘ordinary teachers’ at all levels of education who do not have such expertise, could give a new focus and impetus to research in corpora and language learning, and have a much greater chance of leading to normalisation than experimentation by experts alone. The results of such research would be directly replicable, and thus would easily form the basis of publications other than research articles aimed at teachers and student teachers at secondary level, and language teachers in higher education. They would also serve to inform the content of the supplementary resources which major publishers could provide to encourage language teachers to actively explore their corpus resources with their learners.

As in the case of research on innovations in technology and language learning, evaluation criteria based on the suitability of the corpus work would be required. Indeed, Chapelle's (Reference Chapelle2001, p. 59) ‘Qualities and questions for judgmental analysis of CALL appropriateness’ could serve as a useful starting point. Her criteria for evaluating computer-assisted learning resources include language learning potential, learner fit, meaning focus, authenticity, impact and practicality. In a study I carried out some years ago (Chambers, Reference Chambers, Granger, Gilquin and Meunier2015), these criteria provided a framework for comparing the appropriateness of native-speaker and learner corpora as language learning resources. The alignment with the CEFR used in the pedagogical enhancement of the SACODEYL corpora could also make an important contribution to the integration of corpus data alongside other language resources in the teaching environment. In addition, the nature of the training of learners for corpus consultation and analysis would be an important consideration. Kennedy and Miceli (Reference Kennedy and Miceli2017), in their work with university undergraduate learners of Italian, emphasise the importance of giving the learners useful skills in corpus consultation and analysis without trying to make them into corpus linguists. The strategies they propose could have an important role to play in the development of evaluation criteria. It is beyond the scope of this paper to propose a framework for the integration of corpus resources, consultation and analysis in the everyday teaching environment of the various levels of language education. Indeed it is not a task for one researcher or one paper. It is, however, an undertaking which could perhaps help to build a bridge between the very substantial research carried out by experts in corpus consultation with their own learners and the everyday practice of language teachers and learners in secondary and higher education.