Editorial
This article is based on a presentation summary published in IATEFL BESIG Bonn Conference Selections (spring 2015). My interest in adaptive learning has been inspired by Philip Kerr's blog "Adaptive Leaning in ELT" https://adaptivelearninginelt.wordpress.com

Adaptive Learning in ELT

Ania Kolbuszewska, Poland

Ania Kolbuszewska has been working in ELT for over 25 years as a language coach, trainer, manager as well as a business and academic consultant. A former Eaquals Board member and Director of Eaquals Accreditation and Consultancy Services, she now continues to work as an inspector for this international quality assurance organisation. She is the author of the "Eaquals Self-help Guide to Teacher Development" and co-author of Eaquals management competency framework. Ania is a founder member of IATEFL Poland and a member of IATEFL. E-mail: ania.kolbuszewska@intersection.pl

To explain what adaptive learning is, and how it can and in fact has already affected language teaching and learning, one of the underlying concepts in AL, namely that of big data, needs to be examined.

Big data is a relatively new concept. Data as such has always been there: our names, addresses, educational history, history of illnesses, phone numbers, employment history are all kinds of data, which has been collected and analysed – some for a surprisingly long time. There is also data relating to geography, physics, astronomy, weather, etc. However, it is only with the advances in technology that the amount of data produced has literally exploded: so much so, that we create a vast majority of currently existing data in the two year period preceding any present moment. This is also reflected in language change: we no longer talk about analysing data but about data mining and data miners.

By the year 2018, the number of people working specifically on and with big data is predicted to exceed a quarter of a million. This number is not accidental: it is only the analysis of these enormous amounts of data that allows us to see patterns which are otherwise impossible to detect. These patterns, in turn, are used to predict what may happen in the future in an increasing range of human activity. Examples of these include predictive policing, which allows for a reduction in crime rate by using data to predict where crime is most likely to happen and prevent it more effectively. Predicting customer behaviour and beaming just-in-time marketing of a particular product or service to the customer’s mobile device is already happening and is likely to become an increasingly important form of marketing. The world of finance uses data to find patterns in the seemingly unpredictable market trends.

Big data has also entered education and is the basis for a new generation online systems called adaptive learning systems. These are systems which, by analysing huge amounts of data, create individual learning paths for each individual user of the system. Early adaptive learning systems were only able to show what the learner was getting wrong; current adaptive learning systems claim to do much more: to show not only what the learner gets wrong, but, by analysing the reasons for the mistake, to be able to construct a learning path which will teach the learner to avoid making the same mistake in the future.

Adaptive learning systems have received a mixed reception in the world of education. They are rapidly gaining in popularity in some parts of the world, especially in tertiary education which has in recent years become a battleground for students-customers. The battle lines seem to be drawn between traditional, college-based tertiary education, for which AL systems may be an attractive, cost-cutting alternative, and free, massive open online courses. At the same time, the fact that AL may facilitate cost-cutting measures has sparked strong and well-articulated protests among parents and educators.

It is in the world of language teaching that adaptive learning systems have become the topic of one of the hottest professional debates in recent years. As in other areas of education, one of the major concerns is the issue of data collection – or rather, of data protection. The other concern touches an even more basic issue: are adaptive learning systems suitable for language learning and teaching at all.

From a purely business point of view, AL systems are very attractive for publishers: they allow publishers to use a subscription model for their materials on a much wider scale. This model is more profitable than selling printed materials (e.g. individual coursebooks); in addition, the amount of investment needed for modifying content is significantly lower than in case of printed materials. It is therefore not surprising that AL systems are very actively promoted as a solution to almost everything that ails education these days.

Let us analyse how adaptive learning systems work. In order to arrive at the “why” of making a mistake, an adaptive learning system needs to work on a set of prerequisites with regard to the subject matter taught. This set of prerequisites, a so-called knowledge graph, shows what learners need to know in order to perform a task at a given level correctly. In terms of language teaching and learning, this means that the knowledge graph will show what learners need to know in order to e.g. produce a correct sentence “I haven’t been here for 10 years”. By defining knowledge needed, the knowledge graph will contribute to determining why a learner makes a mistake if s/he says e.g. “I wasn’t here since 10 years”: is it a problem with the Present Perfect tense, or a problem with the auxiliary verb, or perhaps a problem with the use of “since” vs “for”.

To construct a knowledge graph, language would need to be broken down into minute elements, and a way of putting these elements back together in a certain order would need to be defined. This can be done if language is viewed as a wall constructed of individual bricks which are put together layer after layer. This seems to be a very simplistic view of language as a sum total of a finite number of quantifiable elements (bricks), with little or no regard for the human, unpredictable factor of language use, or the social, cultural, symbolic or creative dimensions of using language for communication.

There have been attempts at categorising what language is acquired in what order, such as the CEFR or Pearson’s Global Scale of English. Although very useful for assessing the level a learner has achieved in a foreign language, their value as knowledge graphs is not clear: the categorisation of certain language as belonging to a given level is debatable; there are linguists who claim that deciding what one level a certain piece of language belongs to is extremely difficult, if not impossible.

The “brick to wall” approach to language is also worrying on another level. As language teachers we know that of all the features successful language learners have (fluency, accuracy, creativity and confidence), the very last to develop is accuracy. Grammar and vocabulary are these language systems which are most easily broken down into minute units, and so most readily usable in adaptive learning. However, if adaptive learning systems focus too much on grammar and vocabulary, with the resulting strong bias towards accuracy, learners will either not develop, or will lose confidence, creativity and fluency and will thus become unsuccessful language learners – which in itself defeats the purpose of language teaching in the first place.

Another important controversy is that surrounding data collection, or rather data protection. Adaptive learning systems need big data in order to be reliable. The leaders in the adaptive leaning systems market claim that they will be collecting data from hundreds of millions of leaners. If this is the case, sufficient safeguards need to be in place to ensure that the data collected is appropriately protected and that individuals retain a degree of control over their personal data.

Education is becoming increasingly data-driven in an attempt to make it more quantifiable, in line with a market-driven demand for measurable returns for the money invested in education. On a practical level this may mean that education could be reduced to a money-making machine which does not have sufficient regard for its key human players: teachers and learners. This could also mean that, to those decision-makers who believe in education as a mechanical, additive experience rather than as a human, transformational experience, teachers would seem easily replaceable. Such a belief is likely to bring about a decidedly negative response to AL systems from teachers who might see these as a threat to their jobs.

There is a whole host of other practical issues which arise in the wake of adaptive learning systems entering education. These are connected with teacher training (who will pay for training teachers in using AL systems effectively?), potential widening of the ability gap in classrooms where AL systems are used as supporting systems, or dealing with the aftermath of the novelty effect wearing off once AL systems are more widely used.

Whether we like it or not, big data is here to stay – and so are adaptive learning systems. However, because AL systems seem to have rather serious limitations, it is our role as educators to make sure that we can maximise the potential of such systems for our students and for ourselves, while avoiding, or at least minimising, the dangers they may bring.

Please check the Methodology & Language for Secondary Teachers course at Pilgrims website.
Please check the Teaching Advanced Students course at Pilgrims website.
Please check the Practical Uses of Technology in the Classroom course at Pilgrims website.
Please check the How to Teach Business Professionals at Pilgrims website.
Please check the How to be a Teacher Trainer course at Pilgrims website.