Self-paced listening

One of the most common criticisms of schooling is that it typically requires learners to study in lockstep, with everyone expected to use the same learning material at the same pace to achieve the same learning objectives. From everything we know about individual learner differences, this is an unreasonable and unrealisable expectation. It is only natural, therefore, that we should assume that self-paced learning is a better option. Self-paced learning is at the heart of technology-driven personalized learning. Often, it is the only meaningfully personalized aspect of technology-delivered courses.

Unfortunately, almost one hundred years of attempts to introduce elements of self-pacing into formal language instruction have failed to produce conclusive evidence of its benefits. For a more detailed look at the history of these failures, see my blog post on the topic, and for a more detailed look at Programmed Learning, a 1960s attempt to introduce self-pacing, see this post. This is not to say that self-pacing does not have a potentially important role to play. However, history should act as a warning that the simple provision of self-pacing opportunities through technology may be a necessary condition for successful self-pacing, but it is not a sufficient condition.

Of all the different areas of language learning that can be self-paced, I’ve long thought that technology might help the development of listening skills the most. Much contemporary real-world listening is, in any case, self-paced: why should the classroom not be? With online listening, we can use a variety of help options (Cross, 2017) – pause, rewind, speed control, speech-to-text, dictionary look-up, video / visual support – and we control the frequency and timing of this use. Online listening has become a ‘semi-recursive activity, less dependent on transient memory, inching its way closer to reading’ (Robin, 2007: 110). We don’t know which of these help options and which permutations of these options are most likely to lead to gains in listening skills, but it seems reasonable to believe that some of these options have strong potential. It is perhaps unlikely that research could ever provide a definitive answer to the question of optimal help options: different learners have different needs and different preferences (Cárdenas-Claros & Gruba, 2014). But what is clear is that self-pacing is necessary for these options to be used.

Moving away from whole-class lockstep listening practice towards self-paced independent listening has long been advocated by experts. John Field (2008: 47) identified a key advantage of independent listening: a learner ‘can replay the recording as often as she needs (achieving the kind of recursion that reading offers) and can focus upon specific stretches of the input which are difficult for her personally rather than for the class as a whole’. More recently, interest has also turned to the possibility of self-paced listening in assessment practices (Goodwin, 2017).

So, self-paced listening: what’s not to like? I’ve been pushing it with the teachers I work with for some time. But a recent piece of research from Kathrin Eberharter and colleagues (Eberharter et al., 2023) has given me pause for thought. The researchers wanted to know what effect self-pacing would have on the assessment of listening comprehension in a group of young teenage Austrian learners. They were particularly interested in how learners with SpLDs would be affected, and assumed that self-pacing would boost the performance of these learners. Disappointingly, they were wrong. Not only did self-pacing have, on average, no measurable impact on performance, it also seems that self-pacing may have put learners with shorter working-memory capacity and L1 literacy-related challenges at a disadvantage.

This research concerned self-paced listening in assessment (in this case the TOEFL Junior Standard test), not in learning. But might self-paced listening as part of a learning programme not be quite as beneficial as we might hope? The short answer, as ever, is probably that it depends. Eberhart et al speculate that young learners ‘might need explicit training and more practice in regulating their strategic listening behaviour in order to be able to improve their performance with the help of self-pacing’. This probably holds true for many older learners, too. In other words, it’s not the possibility of self-pacing in itself that will make a huge difference: it’s what a learner does or does not do while they are self-pacing that matters. To benefit from the technological affordances of online listening, learners need to know which strategies (and which tools) may help them. They may need ‘explicit training in exploiting the benefits of navigational freedom to enhance their metacognitive strategy use’ (Eberhart et al. 2023: 17). This shouldn’t surprise us: the role of metacognition is well established (Goh & Vandergrift, 2021).

As noted earlier, we do not really know which permutations of help options are likely to be of most help, but it is a relatively straightforward matter to encourage learners to experiment with them. We do, however, have a much clearer idea of the kinds of listening strategies that are likely to have a positive impact, and the most obvious way of providing this training is in the classroom. John Field (2008) suggested many approaches; Richard Cauldwell (2013) offers more; and Sheila Thorn’s recent ‘Integrating Authentic Listening into the Language Classroom’ (2021) adds yet more. If learners’ metacognitive knowledge, effective listening and help-option skills are going to develop, the training will need to involve ‘a cyclic approach […] throughout an entire course’ (Cross, 2017: 557).

If, on the other hand, our approach to listening in the classroom continues to be (as it is in so many coursebooks) one of testing listening through comprehension questions, we should not be too surprised when learners have little idea what strategy to approach when technology allows self-pacing. Self-paced self-testing of listening comprehension is likely to be of limited value.

References

Cárdenas-Claros, M. S. & Gruba, P. A. (2014) Listeners’ interactions with help options in CALL. Computer Assisted Language Learning, 27 (3): 228 – 245

Cauldwell, R. (2013) Phonology for Listening: Teaching the Stream of Speech. Speech in Action

Cross, J. (2017) Help options for L2 listening in CALL: A research agenda. Language Teaching, 50 (4), 544–560. https://doi.org/10.1017/S0261444817000209

Eberharter,K., Kormos, J.,  Guggenbichler, E.,  Ebner, V. S., Suzuki, S.,  Moser-Frötscher, D., Konrad, E. & Kremmel, B. (2023) Investigating the impact of self-pacing on the L2 listening performance of young learner candidates with differing L1 literacy skills. Language Testing 0 10.1177/02655322221149642 https://journals.sagepub.com/doi/epub/10.1177/02655322221149642

Field, J. (2008) Listening in the Language Classroom. Cambridge: Cambridge University Press

Goh, C. C. M. & Vandergrift, L. (2021) Teaching and learning second language listening: Metacognition in action (2nd ed.). Routledge. https://doi.org/10.4324/9780429287749

Goodwin, S. J. (2017) Locus of control in L2 English listening assessment [Doctoral dissertation]. Georgia State University. https://scholarworks.gsu.edu/cgi/viewcontent.cgi?article=1037&context=alesl_diss

Robin, R. (2007) Commentary: Learner-based listening and technological authenticity. Language Learning & Technology, 11 (1): 109-115. https://www.lltjournal.org/item/461/

Thorn, S. (2021) Integrating Authentic Listening into the Language Classroom. Shoreham-by-Sea: Pavilion

Computer-assisted native-speakerism (CAPTs)

Recent years have seen a proliferation of computer-assisted pronunciations trainers (CAPTs), both as stand-alone apps and as a part of broader language courses. The typical CAPT records the learner’s voice, compares this to a model of some kind, detects differences between the learner and the model, and suggests ways that the learner may more closely approximate to the model (Agarwal & Chakraborty, 2019). Most commonly, the focus is on individual phonemes, rather than, as in Richard Cauldwell’s ‘Cool Speech’ (2012), on the features of fluent natural speech (Rogerson-Revell, 2021).

The fact that CAPTs are increasingly available and attractive ‘does not of course ensure their pedagogic value or effectiveness’ … ‘many are technology-driven rather than pedagogy-led’ (Rogerson-Revell, 2021). Rogerson-Revell (2021) points to two common criticisms of CAPTs. Firstly, their pedagogic accuracy sometimes falls woefully short. He gives the example of a unit on intonation in one app, where users are told that ‘when asking questions in English, our voice goes up in pitch’ and ‘we lower the pitch of our voice at the end of questions’. Secondly, he observes that CAPTs often adopt a one-size-fits-all approach, despite the fact that we know that issues of pronunciation are extremely context-sensitive: ‘a set of learners in one context will need certain features that learners in another context do not’ (Levis, 2018: 239).

There are, in addition, technical challenges that are not easy to resolve. Many CAPTs rely on automatic speech recognition (ASR), which can be very accurate with some accents, but much less so with other accents (including many non-native-speaker accents) (Korzekwa et al., 2022). Anyone using a CAPT will experience instances of the software identifying pronunciation problems that are not problems, and failing to identify potentially more problematic issues (Agarwal & Chakraborty, 2019).

We should not, therefore, be too surprised if these apps don’t always work terribly well. Some apps, like the English File Pronunciation app, have been shown to be effective in helping the perception and production of certain phonemes by a very unrepresentative group of Spanish learners of English (Fouz-González, 2020), but this tells us next to nothing about the overall effectiveness of the app. Most CAPTs have not been independently reviewed, and, according to a recent meta-analysis of CAPTs (Mahdi & Al Khateeb, 2019), the small number of studies are ‘all of very low quality’. This, unfortunately, renders their meta-analysis useless.

Even if the studies in the meta-analysis had not been of very low quality, we would need to pause before digesting any findings about CAPTs’ effectiveness. Before anything else, we need to develop a good understanding of what they might be effective at. It’s here that we run headlong into the problem of native-speakerism (Holliday, 2006; Kiczkowiak, 2018).

The pronunciation model that CAPTs attempt to push learners towards is a native-speaker model. In the case of ELSA Speak, for example, this is a particular kind of American accent, although ‘British and other accents’ will apparently soon be added. Xavier Anguera, co-founder and CTO of ELSA Speak, in a fascinating interview with Paul Raine of TILTAL, happily describes his product as ‘an app that is for accent reduction’. Accent reduction is certainly a more accurate way of describing CAPTs than accent promotion.

Accent reduction, or the attempt to mimic an imagined native-speaker pronunciation, is now ‘rarely put forward by teachers or researchers as a worthwhile goal’ (Levis, 2018: 33) because it is only rarely achievable and, in many contexts, inappropriate. In addition, accent reduction cannot easily be separated from accent prejudice. Accent reduction courses and products ‘operate on the assumption that some accents are more legitimate than others’ (Ennser-Kananen, et al., 2021) and there is evidence that they can ‘reinscribe racial inequalities’ (Ramjattan, 2019). Accent reduction is quintessentially native-speakerist.

Rather than striving towards a native-speaker accentedness, there is a growing recognition among teachers, methodologists and researchers that intelligibility may be a more appropriate learning goal (Levis, 2018) than accentedness. It has been over 20 years since Jennifer Jenkins (2000) developed her Lingua Franca Core (LFC), a relatively short list of pronunciation features that she considered central to intelligibility in English as a Lingua Franca contexts (i.e. the majority of contexts in which English is used). Intelligibility as the guiding principle of pronunciation teaching continues to grow in influence, spurred on by the work of Walker (2010), Kiczkowiak & Lowe (2018), Patsko & Simpson (2019) and Hancock (2020), among others.

Unfortunately, intelligibility is a deceptively simple concept. What exactly it is, is ‘not an easy question to answer’ writes John Levis (2018) before attempting his own answer in the next 250 pages. As admirable as the LFC may be as an attempt to offer a digestible and actionable list of key pronunciation features, it ‘remains controversial in many of its recommendations. It lacks robust empirical support, assumes that all NNS contexts are similar, and does not take into account the importance of stigma associated with otherwise intelligible pronunciations’ (Levis, 2018: 47). Other attempts to list features of intelligibility fare no better in Levis’s view: they are ‘a mishmash of incomplete and contradictory recommendations’ (Levis, 2018: 49).

Intelligibility is also complex because of the relationship between intelligibility and comprehensibility, or the listener’s willingness to understand – their attitude or stance towards the speaker. Comprehensibility is a mediation concept (Ennser-Kananen, et al., 2021). It is a two-way street, and intelligibility-driven approaches need to take this into account (unlike the accent-reduction approach which places all the responsibility for comprehensibility on the shoulders of the othered speaker).

The problem of intelligibility becomes even more thorny when it comes to designing a pronunciation app. Intelligibility and comprehensibility cannot easily be measured (if at all!), and an app’s algorithms need a concrete numerically-represented benchmark towards which a user / learner can be nudged. Accentedness can be measured (even if the app has to reify a ‘native-speaker accent’ to do so). Intelligibility / Comprehensibility is simply not something, as Xavier Anguera acknowledges, that technology can deal with. In this sense, CAPTs cannot avoid being native-speakerist.

At this point, we might ride off indignantly into the sunset, but a couple of further observations are in order. First of all, accentedness and comprehensibility are not mutually exclusive categories. Anguera notes that intelligibility can be partly improved by reducing accentedness, and some of the research cited by Levis (2018) backs him up on this. But precisely how much and what kind of accent reduction improves intelligibility is not knowable, so the use of CAPTs is something of an optimistic stab in the dark. Like all stabs in the dark, there are dangers. Secondly, individual language learners may be forgiven for not wanting to wait for accent prejudice to become a thing of the past: if they feel that they will suffer less from prejudice by attempting here and now to reduce their ‘foreign’ accent, it is not for me, I think, to pass judgement. The trouble, of course, is that CAPTs contribute to the perpetuation of the prejudices.

There is, however, one area where the digital evaluation of accentedness is, I think, unambiguously unacceptable. According to Rogerson-Revell (2021), ‘Australia’s immigration department uses the Pearson Test of English (PTE) Academic as one of five tests. The PTE tests speaking ability using voice recognition technology and computer scoring of test-takers’ audio recordings. However, L1 English speakers and highly proficient L2 English speakers have failed the oral fluency section of the English test, and in some cases it appears that L1 speakers achieve much higher scores if they speak unnaturally slowly and carefully’. Human evaluations are not necessarily any better.

References

Agarwal, C. & Chakraborty, P. (2019) A review of tools and techniques for computer aided pronunciation training (CAPT) in English. Education and Information Technologies, 24: 3731–3743. https://doi.org/10.1007/s10639-019-09955-7

Cauldwell, R (2012) Cool Speech app. Available at: http://www.speechinaction.org/cool-speech-2

Fouz-González, J (2020) Using apps for pronunciation training: An empirical evaluation of the English File Pronunciation app. Language Learning & Technology, 24(1): 62–85.

Ennser-Kananen, J., Halonen, M. & Saarinen, T. (2021) “Come Join Us and Lose Your Accent!” Accent Modification Courses as Hierarchization of International Students. Journal of International Students 11 (2): 322 – 340

Holliday, A. (2006) Native-speakerism. ELT Journal, 60 (4): 385 – 387

Jenkins. J. (2000) The Phonology of English as a Lingua Franca. Oxford: Oxford University Press

Hancock, M. (2020) 50 Tips for Teaching Pronunciation. Cambridge: Cambridge University Press

Kiczkowiak, M. (2018) Native Speakerism in English Language Teaching: Voices From Poland. Doctoral dissertation.

Kiczkowiak, M & Lowe, R. J. (2018) Teaching English as a Lingua Franca. Stuttgart: DELTA Publishing

Korzekwa, D., Lorenzo-Trueba, J., Thomas Drugman, T. & Kostek, B. (2022) Computer-assisted pronunciation training—Speech synthesis is almost all you need. Speech Communication, 142: 22 – 33

Levis, J. M. (2018) Intelligibility, Oral Communication, and the Teaching of Pronunciation. Cambridge: Cambridge University Press

Mahdi, H. S. & Al Khateeb, A. A. (2019) The effectiveness of computer-assisted pronunciation training: A meta-analysis. Review of Education, 7 (3): 733 – 753

Patsko, L. & Simpson, K. (2019) How to Write Pronunciation Activities. ELT Teacher 2 Writer https://eltteacher2writer.co.uk/our-books/how-to-write-pronunciation-activities/

Ramjattan, V. A. (2019) Racializing the problem of and solution to foreign accent in business. Applied Linguistics Review, 13 (4). https://doi.org/10.1515/applirev2019-0058

Rogerson-Revell, P. M. (2021) Computer-Assisted Pronunciation Training (CAPT): Current Issues and Future Directions. RELC Journal, 52(1), 189–205. https://doi.org/10.1177/0033688220977406

Walker, R. (2010) Teaching the Pronunciation of English as a Lingua Franca. Oxford: Oxford University Press

AI and teacher training

Who can tell where a blog post might lead? Over six years ago I wrote about adaptive professional development for teachers. I imagined the possibility of bite-sized, personalized CPD material. Now my vision is becoming real.

For the last two years, I have been working with a start-up that has been using AI to generate text using GPT-3 large language models. GPT-3 has recently been in the news because of the phenomenal success of the newly released ChatGPT. The technology certainly has a wow factor, but it has been around for a while now. ChatGPT can generate texts of various genres on any topic (with a few exceptions like current affairs) and the results are impressive. Imagine, then, how much more impressive the results can be when the kind of text is limited by genre and topic, allowing the software to be trained much more reliably.

This is what we have been working on. We took as our training corpus a huge collection of English language teaching teacher development texts that we could access online: blogs from all the major publishers, personal blogs, transcriptions from recorded conference presentations and webinars, magazine articles directed at teachers, along with books from publishers such as DELTA and Pavilion ELT, etc. We identified topics that seemed to be of current interest and asked our AI to generate blog posts. Later, we were able to get suggestions of topics from the software itself.

We then contacted a number of teachers and trainers who contribute to the publishers’ blogs and contracted them, first, to act as human trainers for the software, and, second, to agree to their names being used as the ‘authors’ of the blog posts we generated. In one or two cases, the authors thought that they had actually written the posts themselves! Next we submitted these posts to the marketing departments of the publishers (who run the blogs). Over twenty were submitted in this way, including:

• What do teachers need to know about teaching 21^st century skills in the English classroom?
• 5 top ways of improving the well-being of English teachers
• Teaching leadership skills in the primary English classroom
• How can we promote eco-literacy in the English classroom?
• My 10 favourite apps for English language learners

We couldn’t, of course, tell the companies that AI had been used to write the copy, but once we were sure that nobody had ever spotted the true authorship of this material, we were ready to move to the next stage of the project. We approached the marketing executives of two publishers and showed how we could generate teacher development material at a fraction of the current cost and in a fraction of the time. Partnerships were quickly signed.

Blog posts were just the beginning. We knew that we could use the same technology to produce webinar scripts, using learning design insights to optimise the webinars. The challenge we faced was that webinars need a presenter. We experimented with using animations, but feedback indicated that participants like to see a face. This is eminently doable, using our contracted authors and deep fake technology, but costs are still prohibitive. It remains cheaper and easier to use our authors delivering the scripts we have generated. This will no doubt change before too long.

The next obvious step was to personalize the development material. Large publishers collect huge amounts of data about visitors to their sites using embedded pixels. It is also relatively cheap and easy to triangulate this data with information from the customer databases and from activity on social media (especially Facebook). We know what kinds of classes people teach, and we know which aspects of teacher development they are interested in.

Publishers have long been interested in personalizing marketing material, and the possibility of extending this to the delivery of real development content is clearly exciting. (See below an email I received this week from the good folks at OUP marketing.)

Earlier this year one of our publishing partners began sending links to personalized materials of the kind we were able to produce with AI. The experiment was such a success that we have already taken it one stage further.

One of the most important clients of our main publishing partner employs hundreds of teachers to deliver online English classes using courseware that has been tailored to the needs of the institution. With so many freelance teachers working for them, along with high turnover of staff, there is inevitably a pressing need for teacher training to ensure optimal delivery. Since the classes are all online, it is possible to capture precisely what is going on. Using an AI-driven tool that was inspired by the Visible Classroom app (informed by the work of John Hattie), we can identify the developmental needs of the teachers. What kinds of activities are they using? How well do they exploit the functionalities of the platform? What can be said about the quality of their teacher talk? We combine this data with everything else and our proprietary algorithms determine what kinds of training materials each teacher receives. It doesn’t stop there. We can also now evaluate the effectiveness of these materials by analysing the learning outcomes of the students.

Teaching efficacy can by massively increased, whilst the training budget of the institution can be slashed. If all goes well, there will be no further need for teacher trainers at all. We won’t be stopping there. If results such as these can be achieved in teacher training, there’s no reason why the same technology cannot be leveraged for the teaching itself. Most of our partner’s teaching and testing materials are now quickly and very cheaply generated using GPT-3.5. If you want to see how this is done, check out the work of edugo.AI (a free trial is available) which can generate gapfills and comprehension test questions in a flash. As for replacing the teachers, we’re getting there. For the time being, though, it’s more cost-effective to use freelancers and to train them up.

Learning and earning

In the campaign for leadership of the British Conservative party, prime ministerial wannabe, Rishi Sunak, announced that he wanted to phase out all university degrees with low ‘earning potential’. This would mean the end of undergraduate courses in fashion, film, philosophy, English language and media studies. And linguistics. More of an attention-grabbing soundbite than anything else, it reflects a view of education that is shared by his competitor, Liz Truss, who ‘is passionate about giving every child basic maths and science skills’ as a way of driving the contribution of education to the economy.

It’s a view that is shared these days by practically everyone with any power and influence, from national governments to organisations like the EU and the OECD (Schuller, 2000). It is rooted in the belief that what matters most in education are the teachable knowledges, skills and competences that are relevant to economic activity (as the OECD puts it). These competences are seen to be essential to economic growth and competitivity, and essential to individuals to enhance their employment potential. Learning equals earning. The way for societies to push this orientation to education is to allow market forces to respond to the presumed demands of the consumers of education (students and their sponsors), as they seek to obtain the best possible return on their investment in education. Market forces are given more power when education is privatized and uncoupled from the state. For this to happen, the market may need a little help in the form of policies from the likes of Sunak and Truss.

This set of beliefs has a name: human capital theory (Becker, 1993). Human capital refers both to the skills that individuals ‘bring to bear in the economy and the need for capital investment in these’ (Holborow, 2012). It is impossible to overstate just how pervasive this theory in contemporary approaches to education is. See, for example, this selection of articles from Science Direct. It is also very easy to forget how recently the lens of human capital has become practically the only lens through which education is viewed.

Contemporary language teaching is perhaps best understood as a series of initiatives that have been driven by human capital theory. First and foremost, there is the global ‘frenzied rush towards acquiring English’ (Holborow, 2018), driven both by governments and by individuals who see that foreign language competence (especially English) ‘might […]open up new opportunities for students [and] assist them in breaking social barriers’ (Kormos & Kiddle, 2013). Children, at ever younger ages (even pre-school), are pushed towards getting a headstart in the race to acquire human capital, whilst there has been an explosive growth in EMI courses (Lasagabaster, 2022). At the same time, there has been mushrooming interest in so-called 21^st century skills (or ‘life skills’ / ‘global skills’) in the English language curriculum. These skills have been identified by asking employers what skills matter most to them when recruiting staff. Critical and creative thinking skills may be seen as having pre-Human Capital, intrinsic educational worth, but it is their potential contribution to economic productivity that explains their general current acceptance.

Investments in human capital need to be measured and measurable. Language teaching needs to be made accountable. Our preoccupation with learning outcomes is seen in the endless number of competency frameworks, and with new tools for quantifying language proficiency. Technology facilitates this evaluation, promises to deliver language teaching more efficiently, and technological skills are, after English language skills themselves, seen to be the most bankable of 21^st century skills. Current interest in social-emotional learning – growth mindsets, grit, resilience and so on – is also driven by a concern to make learning more efficient.

In all of these aspects of language teaching / learning, the private sector (often in private-public partnerships) is very visible. This is by design. Supported by the state, the market economy of education grows in tandem with the rising influence of the private sector on national educational policy. When education ministers lose their job, they can easily find well-paid consultancies in the private sector (as in the case of Sunak and Truss’s colleague, Gavin Williamson).

One of the powers of market-economy ideologies is that it often seems that ‘there is no alternative’ (TINA). There are, however, good reasons to try to think in alternative terms. To begin with, and limiting ourselves for the moment to language teaching, there is a desperate lack of evidence that starting English language learning at very young ages (in the way that is most typically done) will lead to any appreciable gains in the human capital race. It is generally recognised that EMI is highly problematic in a variety of ways (Lasagabaster, 2022). The focus on 21^st century skills has not led to any significant growth in learning outcomes when these skills are measured. There is a worrying lack of evidence that interventions in schools to promote improvements in critical or creative thinking have had much, if any, impact at all. Similarly, there is a worrying lack of evidence that attention to growth mindsets or grit has led to very much at all. Personalized learning, facilitated by technology, likewise has a dismal track record. At the same time, there is no evidence that the interest in measuring learning outcomes has led to any improvement in those outcomes. For all the millions and millions that have been invested in all these trends, the returns have been very slim. Perhaps we would have done better to look for solutions to those aspects of language teaching which we know to be problematic. The obsession with synthetic syllabuses delivered by coursebooks (or their online equivalents) comes to mind.

But beyond the failure of all these things to deliver on their promises, there are broader issues. Although language skills (usually English) have the potential to enhance employment prospects, Holborow (2018) has noted that they do not necessarily do so (see, for example, Yeung & Gray, 2022). Precisely how important language skills are is very hard to determine. A 2016 survey by Cambridge English found that ‘approximately half of all employers offer a better starting package to applicants with good English language skills’ and a similar number indicate that these skills result in faster career progression. But these numbers need to be treated with caution, not least because Cambridge English is in the business of selling English. More importantly, it seems highly unlikely that the figures that are reported reflect the reality of job markets around the world. The survey observes that banking, finance and law are the sectors with the greatest need for such skills, but these are all usually graduate posts. An average of 39% of the population in OECD countries has tertiary education; the percentage is much lower elsewhere. How many students of a given age cohort will actually work in these sectors? Even in rich countries, like Germany and the Netherlands, between 40 and 60% of workers are employed in what is termed ‘nonstandard forms of work’ (OECD, 2015) where language skills will count for little or nothing. These numbers are growing. Language skills are of most value to those students who are already relatively advantaged. That is not to say that there are no potential benefits to everyone in learning English, but these benefits will not be found in better jobs and wages for the majority. One interesting case study describes how a Swiss airport company exploits the language skills of migrant workers, without any benefits (salary or mobility) accruing to the workers themselves (Duchêne, 2011).

The relationship between learning English and earning more is a lot more complex than is usually presented. The same holds true for learning more generally. In the US, ‘nearly two-thirds of job openings in 2020 required no more than a high school diploma’ (Brown et al., 2022: 222). Earnings for graduates in real terms are in decline, except for those at the very top. For the rest, over $1.3 trillion in student loan debt remains unpaid. Elsewhere in the world, the picture is more mixed, but it is clear that learning does not equal earning in the global gig economy.

This evident uncoupling of learning from earning has led some to conclude that education is ‘a waste of time and money’ (Caplan, 2018), a view that has been gaining traction in the US. It’s not an entirely unreasonable view, if the only reason for education is seen to be its contribution to the economy. More commonly, the reaction has been to double-down on human capital theory. In Spain, for example, with its high levels of youth unemployment, there are calls for closer links between educational institutions, and graduates themselves are blamed for failing to take ‘advantage of the upgrading in the demand for skills’ (Bentolilla et al., 2022). This seems almost wilfully cruel, especially since the authors note that there is global trend in falling economic returns in tertiary education (ILO, 2020).

But, rather than doubling-down on human capital theory (e.g. more vocational training, more efficient delivery of the training), it might be a good idea to question human capital theory itself. Both early and more recent critics have tended to accept without hesitation that education can enhance worker productivity, but argue that, as a theory, it is too simplistic to have much explanatory power, and that the supporting evidence is weak, vague or untestable (Bowles & Gintis, 1975; Fix, 2018). Language skills, like education more generally, do not always lead to better employment prospects and salaries, because ‘wider, systemic social inequalities come into play’ (Holborow, 2018). It is not because black women need to brush up on their 21^st century skills that they earn less than white men.

Until recently, critics of human capital theory have been a minority, and largely unheard, voice. But this appear to be changing. The World Bank, more guilty than anyone for pushing human capital theory on the global stage (see here), has recognised that hoped-for job outcomes do not always materialize after massive investments in training systems (World Bank, 2013). Mainstream critics include the Nobel prize winners Joseph Stiglitz and Amartya Sen, and the recent OUP title, ‘The Death of Human Capital?’ (Brown et al., 2020) is likely to spur debate further. The assumption that human capital theory holds water no longer holds water.

When we turn back to English language teaching, we might benefit from some new thinking. For sure, there will be large numbers of English language learners whose only purpose in studying is utilitarian, whose primary desire is to enhance their human capital. But there are also millions, especially children studying in public schools, for whom things are rather different. A major change in thinking involves a reconceptualization of the point of all this English study. If learning English is not, for the majority, seen primarily as a preparation for the workplace, but as compensation for the realities of (un)employment (Brown et al., 2020: 13), most of the recent innovations in ELT would become highly suspect. We would have a much less impoverished view of ‘the complex and multifaceted nature of language’ (Holborow, 2018) and we would find more space for plurilingual practices. A brake on relentless Englishization might be no bad thing (Wilkinson & Gabriëls, 2021). We might be able to explore more fully artistic and creative uses of language. Who knows? We might finally get round to wider implementation of language teaching approaches that we know have a decent chance of success.

References

Becker, G. S. (1993). Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education (3^rd ed.). University of Chicago Press.

Bentolila, S., Felgueroso, F., Jansen, M. et al. (2022). Lost in recessions: youth employment and earnings in Spain. SERIEs 13: 11–49. https://doi.org/10.1007/s13209-021-00244-6

Bowles, S. & Gintis, H. (1975). The Problem with Human Capital Theory – a Marxian critique. The American Economic Review, 65 (2): 74 – 83

Brown, S., Lauder, H. & Cheung, S. Y. (2020). The Death of Human Capital? New York: Oxford University Press

Caplan, B. (2018). The Case against Education: Why the Education System is a Waste of Time and Money. Princeton, NJ: Princeton University Press

Duchêne, A. (2011). Neoliberalism, Social Inequalities, and Multilingualism: The Exploitation of Linguistic Resources and Speakers. Langage et Société, 136 (2): 81 – 108

Fix, B. (2018). The Trouble with Human Capital Theory. Working Papers on Capital as Power, No. 2018/7

Holborow, M. (2012). Neoliberal keywords and the contradictions of an ideology. In Block, D., Gray, J. & Holborow, M. Neoliberalism and Applied Linguistics. Abingdon: Routledge: 33 – 55

Holborow, M. (2018). Language skills as human capital? Challenging the neoliberal frame. Language and Intercultural Communication, 18: (5): 520-532

ILO (2020). Global employment trends for youth, 2020. Geneva: International Labour Organization

Kormos, J., & Kiddle, T. (2013). The role of socio-economic factors in motivation to learn English as a foreign language: the case of Chile. System, 41(2): 399-412

Lasagabaster, D. (2022). English-Medium Instruction in Higher Education. Cambridge: Cambridge University Press

OECD (2015). In It Together, Why Less Inequality Benefits All. Paris: OECD

Schuller, T. (2000). Social and Human Capital: The Search for Appropriate Technomethodology. Policy Studies, 21 (1): 25 – 35

Wilkinson, R., & Gabriëls, R. (Eds.) (2021). The Englishization of Higher Education in Europe. Amsterdam: Amsterdam University Press.

World Bank (2012). World Development Report 2013: Jobs. Washington, DC: World Bank

Yeung, S. & Gray, J. (2022). Neoliberalism, English, and spoiled identity: The case of a high-achieving university graduate in Hong Kong. Language in Society, First View, pp. 1 – 22

How to write about AI in language learning

The paragraph above was written by an AI-powered text generator called neuroflash https://app.neuro-flash.com/home which I told to produce a text on the topic ‘AI and education’. As texts on this topic go, it is both remarkable (in that it was not written by a human) and entirely unremarkable (in that it is practically indistinguishable from hundreds of human-written texts on the same subject). Neuroflash uses a neural network technology called GPT-3 – ‘a large language model’ – and ‘one of the most interesting and important AI systems ever produced’ (Chalmers, 2020). Basically, it generates text by predicting sequences of words based on huge databases. The nature of the paragraph above tells you all you need to know about the kinds of content that are usually found in texts about AI and education.

Not dissimilar from the neuroflash paragraph, educational commentary on uses of AI is characterised by (1) descriptions of AI tools already in use (e.g. speech recognition and machine translation) and (2) vague predictions which invariably refer to ‘the promise of personalised learning, adjusting what we give learners according to what they need to learn and keeping them motivated by giving them content that is of interest to them’ (Hughes, 2022). The question of what precisely will be personalised is unanswered: providing learners with optimal sets of resources (but which ones?), providing counselling services, recommendations or feedback for learners and teachers (but of what kind?) (Luckin, 2022). Nearly four years ago, I wrote https://adaptivelearninginelt.wordpress.com/2018/08/13/ai-and-language-teaching/ about the reasons why these questions remain unanswered. The short answer is that AI in language learning requires a ‘domain knowledge model’. This specifies what is to be learnt and includes an analysis of the steps that must be taken to reach that learning goal. This is lacking in SLA, or, at least, there is no general agreement on what it is. Worse, the models that are most commonly adopted in AI-driven programs (e.g. the deliberate learning of discrete items of grammar and vocabulary) are not supported by either current theory or research (see, for example, VanPatten & Smith, 2022).

In 2021, the IATEFL Learning Technologies SIG organised an event dedicated to AI in education. Unsurprisingly, there was a fair amount of input on AI in assessment, but my interest is in how AI might revolutionize how we learn and teach, not how we assess. What concrete examples did speakers provide?

Rose Luckin, the most well-known British expert on AI in education, kicked things off by mentioning three tools. One of these, Carnegie Learning, is a digital language course that looks very much like any of the ELT courses on offer from the big publishers – a fully blendable, multimedia (e.g. flashcards and videos) synthetic syllabus. This ‘blended learning solution’ is personalizable, since ‘no two students learn alike’, and, it claims, will develop a ‘lifelong love of language’. It appears to be premised on the idea of language learning as optimizing the delivery of ‘content’, of this content consisting primarily of discrete items, and of equating input with uptake. Been there, done that.

A second was Alelo Enskill https://www.alelo.com/about-us/ a chatbot / avatar roleplay program, first developed by the US military to teach Iraqi Arabic and aspects of Iraqi culture to Marines. I looked at the limitations of chatbot technology for language learning here https://adaptivelearninginelt.wordpress.com/2016/12/01/chatbots/ . The third tool mentioned by Luckin was Duolingo. Enough said.

Another speaker at this event was the founder and CEO of Edugo.AI https://www.edugo.ai/ , an AI-powered LMS which uses GPT-3. It allows schools to ‘create and upload on the platform any kind of language material (audio, video, text…). Our AI algorithms process and convert it in gamified exercises, which engage different parts of the brain, and gets students eager to practice’. Does this speaker know anything about gamification (for a quick read, I’d recommend Paul Driver (2012)) or neuroscience, I wonder. What, for that matter, does he know about language learning? Apparently, ‘language is not just about words, language is about sentences’ (Tomasello, 2022). Hmm, this doesn’t inspire confidence.

When you look at current uses of AI in language learning, there is very little (outside of testing, translation and speech ↔ text applications) that could justify enthusiastic claims that AI has any great educational potential. Skepticism seems to me a more reasonable and scientific response: de omnibus dubitandum.

Education is not the only field where AI has been talked up. When Covid hit us, AI was seen as the game-changing technology. It ‘could be deployed to make predictions, enhance efficiencies, and free up staff through automation; it could help rapidly process vast amounts of information and make lifesaving decisions’ (Chakravorti, 2022). The contribution of AI to the development of vaccines has been huge, but its role in diagnosing and triaging patients has been another matter altogether. Hundreds of predictive tools were developed: ‘none of them made a real difference, and some were potentially harmful’ (Heaven, 2021). Expectations were unrealistic and led to the deployment of tools before they were properly trialled. Thirty months down the line, a much more sober understanding of the potential of AI has emerged. Here, then, are the main lessons that have been learnt (I draw particularly on Engler, 2020, and Chakravorti, 2022) that are also relevant to education and language learning.

• Anticipate what could go wrong before anticipating what might go right. Engler (2020) writes that ‘a poorly kept secret of AI practitioners is that 96% accuracy is suspiciously high for any machine learning problem’. In language learning, it is highly unlikely that personalized recommendations will ever reach anything even approaching this level of reliability. What are the implications for individual learners whose learning is inappropriately personalised?
• We also know that a significant problem with AI systems is bias (O’Neil, 2016). There is a well-documented history of discriminatory outcomes because of people’s race, gender, social class or disability profile. Bias needs to be addressed proactively, not reactively.
• Acknowledge from the outset that, for AI to work, huge amounts of data related to prior outcomes will be needed. In the cases of both Covid and language learning, much of this data will be personal. This raises immediate questions of privacy and consent, especially for learners who are children. Don’t minimise this issue.
• Common by-products of AI use include insecure data storage, digital surveillance and consolidation of market power (Engler, 2020). Since such issues are so pervasive, assume they will be a problem, and need to be considered in evaluating the potential of any piece of AI edtech.
• Find out what the subject matter experts have to say: ‘AI is only helpful when applied judiciously by subject-matter experts—people with long-standing experience with the problem that they are trying to solve […] Effectively predicting a badly defined problem is worse than doing nothing at all’ (Engler, 2020). It may be that the problem the AI is trying to help solve is the wrong problem and not worth solving at all (e.g. more efficient ‘delivery’ of explicit grammatical knowledge, gamifying gapfills).

To bring this post to an end, I returned to the AI / GTP-3 driven neuroflash program. This time, I asked it to write a longer piece entitled ‘The dangers of artificial intelligence in education: what could go wrong?’. Here’s what it came up with:

There are a few potential dangers of artificial intelligence in education. One is that AI could be used to replace human teachers. This could lead to a decline in the quality of education, as machines are not yet capable of replicating the full range of human emotions and social interactions that are necessary for teaching. Additionally, AI could be used to create biased or “racist” educational materials, as has already happened with some commercial software products. Finally, AI could be used to track and monitor students in ways that invade their privacy, or that could be used to manipulate their behavior.

All of these dangers are real and should be taken seriously. However, it’s important to remember that AI is still a new technology, and its impact on education is not yet fully understood. As such, we should tread cautiously when implementing AI into our educational systems.

When it comes to education, artificial intelligence could be a double-edged sword. While AI has the potential to revolutionize the field of education and help students learn in more efficient ways, there are also dangers that need to be considered.

One of the dangers of AI in education is that it could lead to a more standardized and homogenized form of learning. If all students are relying on AI-based learning tools, they may all end up learning the same things in the same way. This could lead to a loss of creativity and individuality in education.

Another danger is that AI-based learning tools could become too reliant on data and algorithms. This could lead to a situation where students are simply being taught what the machines think they should know, rather than being allowed to explore and discover new things for themselves.

Finally, there is the danger that AI could eventually replace human teachers altogether. While this may seem like a far-fetched scenario, it is not outside the realm of possibility. If AI-based learning tools become sophisticated enough, they may one day be able to do the job of teaching just as well as humans can.

These are just some of the dangers that need to be considered when it comes to artificial intelligence in education. While AI has the potential to revolutionize education, we need to be careful about how we implement it so that we don’t end up doing more harm than good.

References

Chakravorti, B. (2022) Why AI Failed to Live Up to Its Potential During the Pandemic. Harvard Business Review March 17,2022. https://hbr.org/2022/03/why-ai-failed-to-live-up-to-its-potential-during-the-pandemic

Chalmers, D. (2020) Weinberg, Justin (ed.). “GPT-3 and General Intelligence”. Daily Nous. Philosophers On GPT-3 (updated with replies by GPT-3) July 30, 2020. https://dailynous.com/2020/07/30/philosophers-gpt-3/#chalmers

Driver, P. (2012) The Irony of Gamification. In English Digital Magazine 3, British Council Portugal, pp. 21 – 24 http://digitaldebris.info/digital-debris/2011/12/31/the-irony-of-gamification-written-for-ied-magazine.html

Engler, A. (2020) A guide to healthy skepticism of artificial intelligence and coronavirus. Washington D.C.: Brookings Institution https://www.brookings.edu/research/a-guide-to-healthy-skepticism-of-artificial-intelligence-and-coronavirus/

Heaven, W. D. (2021) Hundreds of AI tools have been built to catch covid. None of them helped. MIT Technology Review, July 30, 2021. https://www.technologyreview.com/2021/07/30/1030329/machine-learning-ai-failed-covid-hospital-diagnosis-pandemic/

Hughes, G. (2022) What lies at the end of the AI rainbow? IATEFL LTSIG Newsletter Issue April 2022

Luckin, R. (2022) The implications of AI for language learning and teaching. IATEFL LTSIG Newsletter Issue April 2022

O’Neil, C. (2016) Weapons of Math Destruction. London: Allen Lane

Tomasello, G. (2022) Next Generation of AI-Language Education Software:NLP & Language Modules (GPT3). IATEFL LTSIG Newsletter Issue April 2022

VanPatten, B. & Smith, M. (2022) Explicit and Implicit Learning in Second Language Acquisition. Cambridge: Cambridge University Press

The Unlearned Lessons of Programmed Learning

From time to time, I have mentioned Programmed Learning (or Programmed Instruction) in this blog (here and here, for example). It felt like time to go into a little more detail about what Programmed Instruction was (and is) and why I think it’s important to know about it.

A brief description

The basic idea behind Programmed Instruction was that subject matter could be broken down into very small parts, which could be organised into an optimal path for presentation to students. Students worked, at their own speed, through a series of micro-tasks, building their mastery of each nugget of learning that was presented, not progressing from one to the next until they had demonstrated they could respond accurately to the previous task.

There were two main types of Programmed Instruction: linear programming and branching programming. In the former, every student would follow the same path, the same sequence of frames. This could be used in classrooms for whole-class instruction and I tracked down a book (illustrated below) called ‘Programmed English Course Student’s Book 1’ (Hill, 1966), which was an attempt to transfer the ideas behind Programmed Instruction to a zero-tech, class environment. This is very similar in approach to the material I had to use when working at an Inlingua school in the 1980s.

An example of how self-paced programming worked is illustrated here, with a section on comparatives.

With branching programming, ‘extra frames (or branches) are provided for students who do not get the correct answer’ (Kay et al., 1968: 19). This was only suitable for self-study, but it was clearly preferable, as it allowed for self-pacing and some personalization. The material could be presented in books (which meant that students had to flick back and forth in their books) or with special ‘teaching machines’, but the latter were preferred.

In the words of an early enthusiast, Programmed Instruction was essentially ‘a device to control a student’s behaviour and help him to learn without the supervision of a teacher’ (Kay et al.,1968: 58). The approach was inspired by the work of Skinner and it was first used as part of a university course in behavioural psychology taught by Skinner at Harvard University in 1957. It moved into secondary schools for teaching mathematics in 1959 (Saettler, 2004: 297).

Enthusiasm and uptake

The parallels between current enthusiasm for the power of digital technology to transform education and the excitement about Programmed Instruction and teaching machines in the 1960s are very striking (McDonald et al., 2005: 90). In 1967, it was reported that ‘we are today on the verge of what promises to be a revolution in education’ (Goodman, 1967: 3) and that ‘tremors of excitement ran through professional journals and conferences and department meetings from coast to coast’ (Kennedy, 1967: 871). The following year, another commentator referred to the way that the field of education had been stirred ‘with an almost Messianic promise of a breakthrough’ (Ornstein, 1968: 401). Programmed instruction was also seen as an exciting business opportunity: ‘an entire industry is just coming into being and significant sales and profits should not be too long in coming’, wrote one hopeful financial analyst as early as 1961 (Kozlowski, 1967: 47).

The new technology seemed to offer a solution to the ‘problems of education’. Media reports in 1963 in Germany, for example, discussed a shortage of teachers, large classes and inadequate learning progress … ‘an ‘urgent pedagogical emergency’ that traditional teaching methods could not resolve’ (Hof, 2018). Individualised learning, through Programmed Instruction, would equalise educational opportunity and if you weren’t part of it, you would be left behind. In the US, two billion dollars were spent on educational technology by the government in the decade following the passing of the National Defense Education Act, and this was added to by grants from private foundations. As a result, ‘the production of teaching machines began to flourish, accompanied by the marketing of numerous ‘teaching units’ stamped into punch cards as well as less expensive didactic programme books and index cards. The market grew dramatically in a short time’ (Hof, 2018).

In the field of language learning, however, enthusiasm was more muted. In the year in which he completed his doctoral studies[1], the eminent linguist, Bernard Spolsky noted that ‘little use is actually being made of the new technique’ (Spolsky, 1966). A year later, a survey of over 600 foreign language teachers at US colleges and universities reported that only about 10% of them had programmed materials in their departments (Valdman, 1968: 1). In most of these cases, the materials ‘were being tried out on an experimental basis under the direction of their developers’. And two years after that, it was reported that ‘programming has not yet been used to any very great extent in language teaching, so there is no substantial body of experience from which to draw detailed, water-tight conclusions’ (Howatt, 1969: 164).

By the early 1970s, Programmed Instruction was already beginning to seem like yesterday’s technology, even though the principles behind it are still very much alive today (Thornbury (2017) refers to Duolingo as ‘Programmed Instruction’). It would be nice to think that language teachers of the day were more sceptical than, for example, their counterparts teaching mathematics. It would be nice to think that, like Spolsky, they had taken on board Chomsky’s (1959) demolition of Skinner. But the widespread popularity of Audiolingual methods suggests otherwise. Audiolingualism, based essentially on the same Skinnerian principles as Programmed Instruction, needed less outlay on technology. The machines (a slide projector and a record or tape player) were cheaper than the teaching machines, could be used for other purposes and did not become obsolete so quickly. The method also lent itself more readily to established school systems (i.e. whole-class teaching) and the skills sets of teachers of the day. Significantly, too, there was relatively little investment in Programmed Instruction for language teaching (compared to, say, mathematics), since this was a smallish and more localized market. There was no global market for English language learning as there is today.

Lessons to be learned

1 Shaping attitudes

It was not hard to persuade some educational authorities of the value of Programmed Instruction. As discussed above, it offered a solution to the problem of ‘the chronic shortage of adequately trained and competent teachers at all levels in our schools, colleges and universities’, wrote Goodman (1967: 3), who added, there is growing realisation of the need to give special individual attention to handicapped children and to those apparently or actually retarded’. The new teaching machines ‘could simulate the human teacher and carry out at least some of his functions quite efficiently’ (Goodman, 1967: 4). This wasn’t quite the same thing as saying that the machines could replace teachers, although some might have hoped for this. The official line was more often that the machines could ‘be used as devices, actively co-operating with the human teacher as adaptive systems and not just merely as aids’ (Goodman, 1967: 37). But this more nuanced message did not always get through, and ‘the Press soon stated that robots would replace teachers and conjured up pictures of classrooms of students with little iron men in front of them’ (Kay et al., 1968: 161).

For teachers, though, it was one thing to be told that the machines would free their time to perform more meaningful tasks, but harder to believe when this was accompanied by a ‘rhetoric of the instructional inadequacies of the teacher’ (McDonald, et al., 2005: 88). Many teachers felt threatened. They ‘reacted against the ‘unfeeling machine’ as a poor substitute for the warm, responsive environment provided by a real, live teacher. Others have seemed to take it more personally, viewing the advent of programmed instruction as the end of their professional career as teachers. To these, even the mention of programmed instruction produces a momentary look of panic followed by the appearance of determination to stave off the ominous onslaught somehow’ (Tucker, 1972: 63).

Some of those who were pushing for Programmed Instruction had a bigger agenda, with their sights set firmly on broader school reform made possible through technology (Hof, 2018). Individualised learning and Programmed Instruction were not just ends in themselves: they were ways of facilitating bigger changes. The trouble was that teachers were necessary for Programmed Instruction to work. On the practical level, it became apparent that a blend of teaching machines and classroom teaching was more effective than the machines alone (Saettler, 2004: 299). But the teachers’ attitudes were crucial: a research study involving over 6000 students of Spanish showed that ‘the more enthusiastic the teacher was about programmed instruction, the better the work the students did, even though they worked independently’ (Saettler, 2004: 299). In other researched cases, too, ‘teacher attitudes proved to be a critical factor in the success of programmed instruction’ (Saettler, 2004: 301).

2 Returns on investment

Pricing a hyped edtech product is a delicate matter. Vendors need to see a relatively quick return on their investment, before a newer technology knocks them out of the market. Developments in computing were fast in the late 1960s, and the first commercially successful personal computer, the Altair 8800, appeared in 1974. But too high a price carried obvious risks. In 1967, the cheapest teaching machine in the UK, the Tutorpack (from Packham Research Ltd), cost £7 12s (equivalent to about £126 today), but machines like these were disparagingly referred to as ‘page-turners’ (Higgins, 1983: 4). A higher-end linear programming machine cost twice this amount. Branching programme machines cost a lot more. The Mark II AutoTutor (from USI Great Britain Limited), for example, cost £31 per month (equivalent to £558), with eight reels of programmes thrown in (Goodman, 1967: 26). A lower-end branching machine, the Grundytutor, could be bought for £ 230 (worth about £4140 today).

This was serious money, and any institution splashing out on teaching machines needed to be confident that they would be well used for a long period of time (Nordberg, 1965). The programmes (the software) were specific to individual machines and the content could not be updated easily. At the same time, other technological developments (cine projectors, tape recorders, record players) were arriving in classrooms, and schools found themselves having to pay for technical assistance and maintenance. The average teacher was ‘unable to avail himself fully of existing aids because, to put it bluntly, he is expected to teach for too many hours a day and simply has not the time, with all the administrative chores he is expected to perform, either to maintain equipment, to experiment with it, let alone keeping up with developments in his own and wider fields. The advent of teaching machines which can free the teacher to fulfil his role as an educator will intensify and not diminish the problem’ (Goodman, 1967: 44). Teaching machines, in short, were ‘oversold and underused’ (Cuban, 2001).

3 Research and theory

Looking back twenty years later, B. F. Skinner conceded that ‘the machines were crude, [and] the programs were untested’ (Skinner, 1986: 105). The documentary record suggests that the second part of this statement is not entirely true. Herrick (1966: 695) reported that ‘an overwhelming amount of research time has been invested in attempts to determine the relative merits of programmed instruction when compared to ‘traditional’ or ‘conventional’ methods of instruction. The results have been almost equally overwhelming in showing no significant differences’. In 1968, Kay et al (1968: 96) noted that ‘there has been a definite effort to examine programmed instruction’. A later meta-analysis of research in secondary education (Kulik et al.: 1982) confirmed that ‘Programmed Instruction did not typically raise student achievement […] nor did it make students feel more positively about the subjects they were studying’.

It was not, therefore, the case that research was not being done. It was that many people were preferring not to look at it. The same holds true for theoretical critiques. In relation to language learning, Spolsky (1966) referred to Chomsky’s (1959) rebuttal of Skinner’s arguments, adding that ‘there should be no need to rehearse these inadequacies, but as some psychologists and even applied linguists appear to ignore their existence it might be as well to remind readers of a few’. Programmed Instruction might have had a limited role to play in language learning, but vendors’ claims went further than that and some people believed them: ‘Rather than addressing themselves to limited and carefully specified FL tasks – for example the teaching of spelling, the teaching of grammatical concepts, training in pronunciation, the acquisition of limited proficiency within a restricted number of vocabulary items and grammatical features – most programmers aimed at self-sufficient courses designed to lead to near-native speaking proficiency’ (Valdman, 1968: 2).

4 Content

When learning is conceptualised as purely the acquisition of knowledge, technological optimists tend to believe that machines can convey it more effectively and more efficiently than teachers (Hof, 2018). The corollary of this is the belief that, if you get the materials right (plus the order in which they are presented and appropriate feedback), you can ‘to a great extent control and engineer the quality and quantity of learning’ (Post, 1972: 14). Learning, in other words, becomes an engineering problem, and technology is its solution.

One of the problems was that technology vendors were, first and foremost, technology specialists. Content was almost an afterthought. Materials writers needed to be familiar with the technology and, if not, they were unlikely to be employed. Writers needed to believe in the potential of the technology, so those familiar with current theory and research would clearly not fit in. The result was unsurprising. Kennedy (1967: 872) reported that ‘there are hundreds of programs now available. Many more will be published in the next few years. Watch for them. Examine them critically. They are not all of high quality’. He was being polite.

5 Motivation

As is usually the case with new technologies, there was a positive novelty effect with Programmed Instruction. And, as is always the case, the novelty effect wears off: ‘students quickly tired of, and eventually came to dislike, programmed instruction’ (McDonald et al.: 89). It could not really have been otherwise: ‘human learning and intrinsic motivation are optimized when persons experience a sense of autonomy, competence, and relatedness in their activity. Self-determination theorists have also studied factors that tend to occlude healthy functioning and motivation, including, among others, controlling environments, rewards contingent on task performance, the lack of secure connection and care by teachers, and situations that do not promote curiosity and challenge’ (McDonald et al.: 93). The demotivating experience of using these machines was particularly acute with younger and ‘less able’ students, as was noted at the time (Valdman, 1968: 9).

The unlearned lessons

I hope that you’ll now understand why I think the history of Programmed Instruction is so relevant to us today. In the words of my favourite Yogi-ism, it’s like deja vu all over again. I have quoted repeatedly from the article by McDonald et al (2005) and I would highly recommend it – available here. Hopefully, too, Audrey Watters’ forthcoming book, ‘Teaching Machines’, will appear before too long, and she will, no doubt, have much more of interest to say on this topic.

References

Chomsky N. 1959. ‘Review of Skinner’s Verbal Behavior’. Language, 35:26–58.

Cuban, L. 2001. Oversold & Underused: Computers in the Classroom. (Cambridge, MA: Harvard University Press)

Goodman, R. 1967. Programmed Learning and Teaching Machines 3rd edition. (London: English Universities Press)

Herrick, M. 1966. ‘Programmed Instruction: A critical appraisal’ The American Biology Teacher, 28 (9), 695 -698

Higgins, J. 1983. ‘Can computers teach?’ CALICO Journal, 1 (2)

Hill, L. A. 1966. Programmed English Course Student’s Book 1. (Oxford: Oxford University Press)

Hof, B. 2018. ‘From Harvard via Moscow to West Berlin: educational technology, programmed instruction and the commercialisation of learning after 1957’ History of Education, 47:4, 445-465

Howatt, A. P. R. 1969. Programmed Learning and the Language Teacher. (London: Longmans)

Kay, H., Dodd, B. & Sime, M. 1968. Teaching Machines and Programmed Instruction. (Harmondsworth: Penguin)

Kennedy, R.H. 1967. ‘Before using Programmed Instruction’ The English Journal, 56 (6), 871 – 873

Kozlowski, T. 1961. ‘Programmed Teaching’ Financial Analysts Journal, 17 / 6, 47 – 54

Kulik, C.-L., Schwalb, B. & Kulik, J. 1982. ‘Programmed Instruction in Secondary Education: A Meta-analysis of Evaluation Findings’ Journal of Educational Research, 75: 133 – 138

McDonald, J. K., Yanchar, S. C. & Osguthorpe, R.T. 2005. ‘Learning from Programmed Instruction: Examining Implications for Modern Instructional Technology’ Educational Technology Research and Development, 53 / 2, 84 – 98

Nordberg, R. B. 1965. Teaching machines-six dangers and one advantage. In J. S. Roucek (Ed.), Programmed teaching: A symposium on automation in education (pp. 1–8). (New York: Philosophical Library)

Ornstein, J. 1968. ‘Programmed Instruction and Educational Technology in the Language Field: Boon or Failure?’ The Modern Language Journal, 52 / 7, 401 – 410

Post, D. 1972. ‘Up the programmer: How to stop PI from boring learners and strangling results’. Educational Technology, 12(8), 14–1

Saettler, P. 2004. The Evolution of American Educational Technology. (Greenwich, Conn.: Information Age Publishing)

Skinner, B. F. 1986. ‘Programmed Instruction Revisited’ The Phi Delta Kappan, 68 (2), 103 – 110

Spolsky, B. 1966. ‘A psycholinguistic critique of programmed foreign language instruction’ International Review of Applied Linguistics in Language Teaching, Volume 4, Issue 1-4: 119–130

Thornbury, S. 2017. Scott Thornbury’s 30 Language Teaching Methods. (Cambridge: Cambridge University Press)

Tucker, C. 1972. ‘Programmed Dictation: An Example of the P.I. Process in the Classroom’. TESOL Quarterly, 6(1), 61-70

Valdman, A. 1968. ‘Programmed Instruction versus Guided Learning in Foreign Language Acquisition’ Die Unterrichtspraxis / Teaching German, 1 (2), 1 – 14

 

 

 

[1] Spolsky’ doctoral thesis for the University of Montreal was entitled ‘The psycholinguistic basis of programmed foreign language instruction’.

 

 

 

 

 

Platforms and big data in ELT – a look back at the last decade

At the start of the last decade, ELT publishers were worried, Macmillan among them. The financial crash of 2008 led to serious difficulties, not least in their key Spanish market. In 2011, Macmillan’s parent company was fined ₤11.3 million for corruption. Under new ownership, restructuring was a constant. At the same time, Macmillan ELT was getting ready to move from its Oxford headquarters to new premises in London, a move which would inevitably lead to the loss of a sizable proportion of its staff. On top of that, Macmillan, like the other ELT publishers, was aware that changes in the digital landscape (the first 3G iPhone had appeared in June 2008 and wifi access was spreading rapidly around the world) meant that they needed to shift away from the old print-based model. With her finger on the pulse, Caroline Moore, wrote an article in October 2010 entitled ‘No Future? The English Language Teaching Coursebook in the Digital Age’ . The publication (at the start of the decade) and runaway success of the online ‘Touchstone’ course, from arch-rivals, Cambridge University Press, meant that Macmillan needed to change fast if they were to avoid being left behind.

Macmillan already had a platform, Campus, but it was generally recognised as being clunky and outdated, and something new was needed. In the summer of 2012, Macmillan brought in two new executives – people who could talk the ‘creative-disruption’ talk and who believed in the power of big data to shake up English language teaching and publishing. At the time, the idea of big data was beginning to reach public consciousness and ‘Big Data: A Revolution that Will Transform how We Live, Work, and Think’ by Viktor Mayer-Schönberger and Kenneth Cukier, was a major bestseller in 2013 and 2014. ‘Big data’ was the ‘hottest trend’ in technology and peaked in Google Trends in October 2014. See the graph below.

Not long after taking up their positions, the two executives began negotiations with Knewton, an American adaptive learning company. Knewton’s technology promised to gather colossal amounts of data on students using Knewton-enabled platforms. Its founder, Jose Ferreira, bragged that Knewton had ‘more data about our students than any company has about anybody else about anything […] We literally know everything about what you know and how you learn best, everything’. This data would, it was claimed, enable publishers to multiply, by orders of magnitude, the efficacy of learning materials, allowing publishers, like Macmillan, to provide a truly personalized and optimal offering to learners using their platform.

The contract between Macmillan and Knewton was agreed in May 2013 ‘to build next-generation English Language Learning and Teaching materials’. Perhaps fearful of being left behind in what was seen to be a winner-takes-all market (Pearson already had a financial stake in Knewton), Cambridge University Press duly followed suit, signing a contract with Knewton in September of the same year, in order ‘to create personalized learning experiences in [their] industry-leading ELT digital products’. Things moved fast because, by the start of 2014 when Macmillan’s new catalogue appeared, customers were told to ‘watch out for the ‘Big Tree’’, Macmillans’ new platform, which would be powered by Knewton. ‘The power that will come from this world of adaptive learning takes my breath away’, wrote the international marketing director.

Not a lot happened next, at least outwardly. In the following year, 2015, the Macmillan catalogue again told customers to ‘look out for the Big Tree’ which would offer ‘flexible blended learning models’ which could ‘give teachers much more freedom to choose what they want to do in the class and what they want the students to do online outside of the classroom’.

But behind the scenes, everything was going wrong. It had become clear that a linear model of language learning, which was a necessary prerequisite of the Knewton system, simply did not lend itself to anything which would be vaguely marketable in established markets. Skills development, not least the development of so-called 21^st century skills, which Macmillan was pushing at the time, would not be facilitated by collecting huge amounts of data and algorithms offering personalized pathways. Even if it could, teachers weren’t ready for it, and the projections for platform adoptions were beginning to seem very over-optimistic. Costs were spiralling. Pushed to meet unrealistic deadlines for a product that was totally ill-conceived in the first place, in-house staff were suffering, and this was made worse by what many staffers thought was a toxic work environment. By the end of 2014 (so, before the copy for the 2015 catalogue had been written), the two executives had gone.

For some time previously, skeptics had been joking that Macmillan had been barking up the wrong tree, and by the time that the 2016 catalogue came out, the ‘Big Tree’ had disappeared without trace. The problem was that so much time and money had been thrown at this particular tree that not enough had been left to develop new course materials (for adults). The whole thing had been a huge cock-up of an extraordinary kind.

Cambridge, too, lost interest in their Knewton connection, but were fortunate (or wise) not to have invested so much energy in it. Language learning was only ever a small part of Knewton’s portfolio, and the company had raised over $180 million in venture capital. Its founder, Jose Ferreira, had been a master of marketing hype, but the business model was not delivering any better than the educational side of things. Pearson pulled out. In December 2016, Ferreira stepped down and was replaced as CEO. The company shifted to ‘selling digital courseware directly to higher-ed institutions and students’ but this could not stop the decline. In September of 2019, Knewton was sold for something under $17 million dollars, with investors taking a hit of over $160 million. My heart bleeds.

It was clear, from very early on (see, for example, my posts from 2014 here and here) that Knewton’s product was little more than what Michael Feldstein called ‘snake oil’. Why and how could so many people fall for it for so long? Why and how will so many people fall for it again in the coming decade, although this time it won’t be ‘big data’ that does the seduction, but AI (which kind of boils down to the same thing)? The former Macmillan executives are still at the game, albeit in new companies and talking a slightly modified talk, and Jose Ferreira (whose new venture has already raised $3.7 million) is promising to revolutionize education with a new start-up which ‘will harness the power of technology to improve both access and quality of education’ (thanks to Audrey Watters for the tip). Investors may be desperate to find places to spread their portfolio, but why do the rest of us lap up the hype? It’s a question to which I will return.

 

 

 

 

EdTech: a cautionary tale

When the startup, AltSchool, was founded in 2013 by Max Ventilla, the former head of personalization at Google, it quickly drew the attention of venture capitalists and within a few years had raised $174 million from the likes of the Zuckerberg Foundation, Peter Thiel, Laurene Powell Jobs and Pierre Omidyar. It garnered gushing articles in a fawning edtech press which enthused about ‘how successful students can be when they learn in small, personalized communities that champion project-based learning, guided by educators who get a say in the technology they use’. It promised ‘a personalized learning approach that would far surpass the standardized education most kids receive’.

Ventilla was an impressive money-raiser who used, and appeared to believe, every cliché in the edTech sales manual. Dressed in regulation jeans, polo shirt and fleece, he claimed that schools in America were ‘stuck in an industrial-age model, [which] has been in steady decline for the last century’ . What he offered, instead, was a learner-centred, project-based curriculum providing real-world lessons. There was a focus on social-emotional learning activities and critical thinking was vital.

The key to the approach was technology. From the start, software developers, engineers and researchers worked alongside teachers everyday, ‘constantly tweaking the Personalized Learning Plan, which shows students their assignments for each day and helps teachers keep track of and assess student’s learning’. There were tablets for pre-schoolers, laptops for older kids and wall-mounted cameras to record the lessons. There were, of course, Khan Academy videos. Ventilla explained that “we start with a representation of each child”, and even though “the vast majority of the learning should happen non-digitally”, the child’s habits and preferences gets converted into data, “a digital representation of the important things that relate to that child’s learning, not just their academic learning but also their non-academic learning. Everything logistic that goes into setting up the experience for them, whether it’s who has permission to pick them up or their allergy information. You name it.” And just like Netflix matches us to TV shows, “If you have that accurate and actionable representation for each child, now you can start to personalize the whole experience for that child. You can create that kind of loop you described where because we can represent a child well, we can match them to the right experiences.”

AltSchool seemed to offer the possibility of doing something noble, of transforming education, ‘bringing it into the digital age’, and, at the same time, a healthy return on investors’ money. Expanding rapidly, nine AltSchool microschools were opened in New York and the Bay Area, and plans were afoot for further expansion in Chicago. But, by then, it was already clear that something was going wrong. Five of the schools were closed before they had really got started and the attrition rate in some classrooms had reached about 30%. Revenue in 2018 was only $7 million and there were few buyers for the AltSchool platform. Quoting once more from the edTech bible, Ventilla explained the situation: ‘Our whole strategy is to spend more than we make,’ he says. Since software is expensive to develop and cheap to distribute, the losses, he believes, will turn into steep profits once AltSchool refines its product and lands enough customers.

The problems were many and apparent. Some of the buildings were simply not appropriate for schools, with no playgrounds or gyms, malfunctioning toilets, among other issues. Parents were becoming unhappy and accused AltSchool of putting ‘its ambitions as a tech company above its responsibility to teach their children. […] We kind of came to the conclusion that, really, AltSchool as a school was kind of a front for what Max really wants to do, which is develop software that he’s selling,’ a parent of a former AltSchool student told Business Insider. ‘We had really mediocre educators using technology as a crutch,’ said one father who transferred his child to a different private school after two years at AltSchool. […] We learned that it’s almost impossible to really customize the learning experience for each kid.’ Some parents began to wonder whether AltSchool had enticed families into its program merely to extract data from their children, then toss them aside?

With the benefit of hindsight, it would seem that the accusations were hardly unfair. In June of this year, AltSchool announced that its four remaining schools would be operated by a new partner, Higher Ground Education (a well-funded startup founded in 2016 which promotes and ‘modernises’ Montessori education). Meanwhile, AltSchool has been rebranded as Altitude Learning, focusing its ‘resources on the development and expansion of its personalized learning platform’ for licensing to other schools across the country.

Quoting once more from the edTech sales manual, Ventilla has said that education should drive the tech, not the other way round. Not so many years earlier, before starting AltSchool, Ventilla also said that he had read two dozen books on education and emerged a fan of Sir Ken Robinson. He had no experience as a teacher or as an educational administrator. Instead, he had ‘extensive knowledge of networks, and he understood the kinds of insights that can be gleaned from big data’.

Big data, learning analytics and language learning / teaching

The use of big data and analytics in education continues to grow.

A vast apparatus of measurement is being developed to underpin national education systems, institutions and the actions of the individuals who occupy them. […] The presence of digital data and software in education is being amplified through massive financial and political investment in educational technologies, as well as huge growth in data collection and analysis in policymaking practices, extension of performance measurement technologies in the management of educational institutions, and rapid expansion of digital methodologies in educational research. To a significant extent, many of the ways in which classrooms function, educational policy departments and leaders make decisions, and researchers make sense of data, simply would not happen as currently intended without the presence of software code and the digital data processing programs it enacts. (Williamson, 2017: 4)

The most common and successful use of this technology so far has been in the identification of students at risk of dropping out of their courses (Jørno & Gynther, 2018: 204). The kind of analytics used in this context may be called ‘academic analytics’ and focuses on educational processes at the institutional level or higher (Gelan et al, 2018: 3). However, ‘learning analytics’, the capture and analysis of learner and learning data in order to personalize learning ‘(1) through real-time feedback on online courses and e-textbooks that can ‘learn’ from how they are used and ‘talk back’ to the teacher, and (2) individualization and personalization of the educational experience through adaptive learning systems that enable materials to be tailored to each student’s individual needs through automated real-time analysis’ (Mayer-Schönberger & Cukier, 2014) has become ‘the main keyword of data-driven education’ (Williamson, 2017: 10). See my earlier posts on this topic here and here and here.

Near the start of Mayer-Schönberger and Cukier’s enthusiastic sales pitch (Learning with Big Data: The Future of Education) for the use of big data in education, there is a discussion of Duolingo. They quote Luis von Ahn, the founder of Duolingo, as saying ‘there has been little empirical work on what is the best way to teach a foreign language’. This is so far from the truth as to be laughable. Von Ahn’s comment, along with the Duolingo product itself, is merely indicative of a lack of awareness of the enormous amount of research that has been carried out. But what could the data gleaned from the interactions of millions of users with Duolingo tell us of value? The example that is given is the following. Apparently, ‘in the case of Spanish speakers learning English, it’s common to teach pronouns early on: words like “he,” “she,” and “it”.’ But, Duolingo discovered, ‘the term “it” tends to confuse and create anxiety for Spanish speakers, since the word doesn’t easily translate into their language […] Delaying the introduction of “it” until a few weeks later dramatically improves the number of people who stick with learning English rather than drop out.’ Was von Ahn unaware of the decades of research into language transfer effects? Did von Ahn (who grew up speaking Spanish in Guatemala) need all this data to tell him that English personal pronouns can cause problems for Spanish learners of English? Was von Ahn unaware of the debates concerning the value of teaching isolated words (especially grammar words!)?

The area where little empirical research has been done is not in different ways of learning another language: it is in the use of big data and learning analytics to assist language learning. Claims about the value of these technologies in language learning are almost always speculative – they are based on comparison to other school subjects (especially, mathematics). Gelan et al (2018: 2), who note this lack of research, suggest that ‘understanding language learner behaviour could provide valuable insights into task design for instructors and materials designers, as well as help students with effective learning strategies and personalised learning pathways’ (my italics). Reinders (2018: 81) writes ‘that analysis of prior experiences with certain groups or certain courses may help to identify key moments at which students need to receive more or different support. Analysis of student engagement and performance throughout a course may help with early identification of learning problems and may prompt early intervention’ (italics added). But there is some research out there, and it’s worth having a look at. Most studies that have collected learner-tracking data concern glossary use for reading comprehension and vocabulary retention (Gelan et al, 2018: 5), but a few have attempted to go further in scope.

Volk et al (2015) looked at the behaviour of the 20,000 students per day using the platform which accompanies ‘More!’ (Gerngross et al. 2008) to do their English homework for Austrian lower secondary schools. They discovered that

• the exercises used least frequently were those that are located further back in the course book
• usage is highest from Monday to Wednesday, declining from Thursday, with a rise again on Sunday
• most interaction took place between 3:00 and 5:00 pm.
• repetition of exercises led to a strong improvement in success rate
• students performed better on multiple choice and matching exercises than they did where they had to produce some language

The authors of this paper conclude by saying that ‘the results of this study suggest a number of new avenues for research. In general, the authors plan to extend their analysis of exercise results and applied exercises to the population of all schools using the online learning platform more-online.at. This step enables a deeper insight into student’s learning behaviour and allows making more generalizing statements.’ When I shared these research findings with the Austrian lower secondary teachers that I work with, their reaction was one of utter disbelief. People get paid to do this research? Why not just ask us?

More useful, more actionable insights may yet come from other sources. For example, Gu Yueguo, Pro-Vice-Chancellor of the Beijing Foreign Studies University has announced the intention to set up a national Big Data research center, specializing in big data-related research topics in foreign language education (Yu, 2015). Meanwhile, I’m aware of only one big research project that has published its results. The EC Erasmus+ VITAL project (Visualisation Tools and Analytics to monitor Online Language Learning & Teaching) was carried out between 2015 and 2017 and looked at the learning trails of students from universities in Belgium, Britain and the Netherlands. It was discovered (Gelan et al, 2015) that:

• students who did online exercises when they were supposed to do them were slightly more successful than those who were late carrying out the tasks
• successful students logged on more often, spent more time online, attempted and completed more tasks, revisited both exercises and theory pages more frequently, did the work in the order in which it was supposed to be done and did more work in the holidays
• most students preferred to go straight into the assessed exercises and only used the theory pages when they felt they needed to; successful students referred back to the theory pages more often than unsuccessful students
• students made little use of the voice recording functionality
• most online activity took place the day before a class and the day of the class itself

EU funding for this VITAL project amounted to 274,840 Euros[1]. The technology for capturing the data has been around for a long time. In my opinion, nothing of value, or at least nothing new, has been learnt. Publishers like Pearson and Cambridge University Press who have large numbers of learners using their platforms have been capturing learning data for many years. They do not publish their findings and, intriguingly, do not even claim that they have learnt anything useful / actionable from the data they have collected. Sure, an exercise here or there may need to be amended. Both teachers and students may need more support in using the more open-ended functionalities of the platforms (e.g. discussion forums). But are they getting ‘unprecedented insights into what works and what doesn’t’ (Mayer-Schönberger & Cukier, 2014)? Are they any closer to building better pedagogies? On the basis of what we know so far, you wouldn’t want to bet on it.

It may be the case that all the learning / learner data that is captured could be used in some way that has nothing to do with language learning. Show me a language-learning app developer who does not dream of monetizing the ‘behavioural surplus’ (Zuboff, 2018) that they collect! But, for the data and analytics to be of any value in guiding language learning, it must lead to actionable insights. Unfortunately, as Jørno & Gynther (2018: 198) point out, there is very little clarity about what is meant by ‘actionable insights’. There is a danger that data and analytics ‘simply gravitates towards insights that confirm longstanding good practice and insights, such as “students tend to ignore optional learning activities … [and] focus on activities that are assessed” (Jørno & Gynther, 2018: 211). While this is happening, the focus on data inevitably shapes the way we look at the object of study (i.e. language learning), ‘thereby systematically excluding other perspectives’ (Mau, 2019: 15; see also Beer, 2019). The belief that tech is always the solution, that all we need is more data and better analytics, remains very powerful: it’s called techno-chauvinism (Broussard, 2018: 7-8).

References

Beer, D. 2019. The Data Gaze. London: Sage

Broussard, M. 2018. Artificial Unintelligence. Cambridge, Mass.: MIT Press

Gelan, A., Fastre, G., Verjans, M., Martin, N., Jansenswillen, G., Creemers, M., Lieben, J., Depaire, B. & Thomas, M. 2018. ‘Affordances and limitations of learning analytics for computer­assisted language learning: a case study of the VITAL project’. Computer Assisted Language Learning. pp. 1­26. http://clok.uclan.ac.uk/21289/

Gerngross, G., Puchta, H., Holzmann, C., Stranks, J., Lewis-Jones, P. & Finnie, R. 2008. More! 1 Cyber Homework. Innsbruck, Austria: Helbling

Jørno, R. L. & Gynther, K. 2018. ‘What Constitutes an “Actionable Insight” in Learning Analytics?’ Journal of Learning Analytics 5 (3): 198 – 221

Mau, S. 2019. The Metric Society. Cambridge: Polity Press

Mayer-Schönberger, V. & Cukier, K. 2014. Learning with Big Data: The Future of Education. New York: Houghton Mifflin Harcourt

Reinders, H. 2018. ‘Learning analytics for language learning and teaching’. JALT CALL Journal 14 / 1: 77 – 86 https://files.eric.ed.gov/fulltext/EJ1177327.pdf

Volk, H., Kellner, K. & Wohlhart, D. 2015. ‘Learning Analytics for English Language Teaching.’ Journal of Universal Computer Science, Vol. 21 / 1: 156-174 http://www.jucs.org/jucs_21_1/learning_analytics_for_english/jucs_21_01_0156_0174_volk.pdf

Williamson, B. 2017. Big Data in Education. London: Sage

Yu, Q. 2015. ‘Learning Analytics: The next frontier for computer assisted language learning in big data age’ SHS Web of Conferences, 17 https://www.shs-conferences.org/articles/shsconf/pdf/2015/04/shsconf_icmetm2015_02013.pdf

Zuboff, S. 2019. The Age of Surveillance Capitalism. London: Profile Books

 

[1] See https://ec.europa.eu/programmes/erasmus-plus/sites/erasmusplus2/files/ka2-2015-he_en.pdf

AI and language teaching

It’s hype time again. Spurred on, no doubt, by the current spate of books and articles  about AIED (artificial intelligence in education), the IATEFL Learning Technologies SIG is organising an online event on the topic in November of this year. Currently, the most visible online references to AI in language learning are related to Glossika , basically a language learning system that uses spaced repetition, whose marketing department has realised that references to AI might help sell the product. They’re not alone – see, for example, Knowble which I reviewed earlier this year .

In the wider world of education, where AI has made greater inroads than in language teaching, every day brings more stuff: How artificial intelligence is changing teaching , 32 Ways AI is Improving Education , How artificial intelligence could help teachers do a better job , etc., etc. There’s a full-length book by Anthony Seldon, The Fourth Education Revolution: will artificial intelligence liberate or infantilise humanity? (2018, University of Buckingham Press) – one of the most poorly researched and badly edited books on education I’ve ever read, although that won’t stop it selling – and, no surprises here, there’s a Pearson commissioned report called Intelligence Unleashed: An argument for AI in Education (2016) which is available free.

Common to all these publications is the claim that AI will radically change education. When it comes to language teaching, a similar claim has been made by Donald Clark (described by Anthony Seldon as an education guru but perhaps best-known to many in ELT for his demolition of Sugata Mitra). In 2017, Clark wrote a blog post for Cambridge English (now unavailable) entitled How AI will reboot language learning, and a more recent version of this post, called AI has and will change language learning forever (sic) is available on Clark’s own blog. Given the history of the failure of education predictions, Clark is making bold claims. Thomas Edison (1922) believed that movies would revolutionize education. Radios were similarly hyped in the 1940s and in the 1960s it was the turn of TV. In the 1980s, Seymour Papert predicted the end of schools – ‘the computer will blow up the school’, he wrote. Twenty years later, we had the interactive possibilities of Web 2.0. As each technology failed to deliver on the hype, a new generation of enthusiasts found something else to make predictions about.

But is Donald Clark onto something? Developments in AI and computational linguistics have recently resulted in enormous progress in machine translation. Impressive advances in automatic speech recognition and generation, coupled with the power that can be packed into a handheld device, mean that we can expect some re-evaluation of the value of learning another language. Stephen Heppell, a specialist at Bournemouth University in the use of ICT in Education, has said: ‘Simultaneous translation is coming, making language teachers redundant. Modern languages teaching in future may be more about navigating cultural differences’ (quoted by Seldon, p.263). Well, maybe, but this is not Clark’s main interest.

Less a matter of opinion and much closer to the present day is the issue of assessment. AI is becoming ubiquitous in language testing. Cambridge, Pearson, TELC, Babbel and Duolingo are all using or exploring AI in their testing software, and we can expect to see this increase. Current, paper-based systems of testing subject knowledge are, according to Rosemary Luckin and Kristen Weatherby, outdated, ineffective, time-consuming, the cause of great anxiety and can easily be automated (Luckin, R. & Weatherby, K. 2018. ‘Learning analytics, artificial intelligence and the process of assessment’ in Luckin, R. (ed.) Enhancing Learning and Teaching with Technology, 2018. UCL Institute of Education Press, p.253). By capturing data of various kinds throughout a language learner’s course of study and by using AI to analyse learning development, continuous formative assessment becomes possible in ways that were previously unimaginable. ‘Assessment for Learning (AfL)’ or ‘Learning Oriented Assessment (LOA)’ are two terms used by Cambridge English to refer to the potential that AI offers which is described by Luckin (who is also one of the authors of the Pearson paper mentioned earlier). In practical terms, albeit in a still very limited way, this can be seen in the CUP course ‘Empower’, which combines CUP course content with validated LOA from Cambridge Assessment English.

Will this reboot or revolutionise language teaching? Probably not and here’s why. AIED systems need to operate with what is called a ‘domain knowledge model’. This specifies what is to be learnt and includes an analysis of the steps that must be taken to reach that learning goal. Some subjects (especially STEM subjects) ‘lend themselves much more readily to having their domains represented in ways that can be automatically reasoned about’ (du Boulay, D. et al., 2018. ‘Artificial intelligences and big data technologies to close the achievement gap’ in Luckin, R. (ed.) Enhancing Learning and Teaching with Technology, 2018. UCL Institute of Education Press, p.258). This is why most AIED systems have been built to teach these areas. Language are rather different. We simply do not have a domain knowledge model, except perhaps for the very lowest levels of language learning (and even that is highly questionable). Language learning is probably not, or not primarily, about acquiring subject knowledge. Debate still rages about the relationship between explicit language knowledge and language competence. AI-driven formative assessment will likely focus most on explicit language knowledge, as does most current language teaching. This will not reboot or revolutionise anything. It will more likely reinforce what is already happening: a model of language learning that assumes there is a strong interface between explicit knowledge and language competence. It is not a model that is shared by most SLA researchers.

So, one thing that AI can do (and is doing) for language learning is to improve the algorithms that determine the way that grammar and vocabulary are presented to individual learners in online programs. AI-optimised delivery of ‘English Grammar in Use’ may lead to some learning gains, but they are unlikely to be significant. It is not, in any case, what language learners need.

AI, Donald Clark suggests, can offer personalised learning. Precisely what kind of personalised learning this might be, and whether or not this is a good thing, remains unclear. A 2015 report funded by the Gates Foundation found that we currently lack evidence about the effectiveness of personalised learning. We do not know which aspects of personalised learning (learner autonomy, individualised learning pathways and instructional approaches, etc.) or which combinations of these will lead to gains in language learning. The complexity of the issues means that we may never have a satisfactory explanation. You can read my own exploration of the problems of personalised learning starting here .

What’s left? Clark suggests that chatbots are one area with ‘huge potential’. I beg to differ and I explained my reasons eighteen months ago . Chatbots work fine in very specific domains. As Clark says, they can be used for ‘controlled practice’, but ‘controlled practice’ means practice of specific language knowledge, the practice of limited conversational routines, for example. It could certainly be useful, but more than that? Taking things a stage further, Clark then suggests more holistic speaking and listening practice with Amazon Echo, Alexa or Google Home. If and when the day comes that we have general, as opposed to domain-specific, AI, chatting with one of these tools would open up vast new possibilities. Unfortunately, general AI does not exist, and until then Alexa and co will remain a poor substitute for human-human interaction (which is readily available online, anyway). Incidentally, AI could be used to form groups of online language learners to carry out communicative tasks – ‘the aim might be to design a grouping of students all at a similar cognitive level and of similar interests, or one where the participants bring different but complementary knowledge and skills’ (Luckin, R., Holmes, W., Griffiths, M. & Forceir, L.B. 2016. Intelligence Unleashed: An argument for AI in Education. London: Pearson, p.26).

Predictions about the impact of technology on education have a tendency to be made by people with a vested interest in the technologies. Edison was a businessman who had invested heavily in motion pictures. Donald Clark is an edtech entrepreneur whose company, Wildfire, uses AI in online learning programs. Stephen Heppell is executive chairman of LP+ who are currently developing a Chinese language learning community for 20 million Chinese school students. The reporting of AIED is almost invariably in websites that are paid for, in one way or another, by edtech companies. Predictions need, therefore, to be treated sceptically. Indeed, the safest prediction we can make about hyped educational technologies is that inflated expectations will be followed by disillusionment, before the technology finds a smaller niche.