You only need to look at the tens of millions of subscribers to wellness apps like Strava and Sleep Cycle, tracking step counts and hours of shut-eye, to know that when it comes to achieving our goals, we want the data. Language learning is no exception. 

For the first time in language learning, AI can be used to gather data on students’ speech in real time. Until recently, it was only possible to analyze people’s language from self-study activities, but there was a black box around organic, conversational speaking. Now, AI can give us these insights – as long as you have access to the right data.  

And that’s where EF Education First is unique, delivering more than 5,000 hours of online classes every day. When it’s hooked up to a data set of this scale, AI can identify students’ strengths and areas for improvement, measure who’s talking the most and who’s hanging back in group lessons, and flag ‘critical learning incidents’ – “the ‘a-ha!’ moments where a student suddenly understands something,” explains Dr David Bish, Head of Academic Affairs at Hult EF Corporate Education, EF Education First’s B2B arm. 

It's also monitoring teachers’ language, providing another layer of quality assurance for training providers. Analyzing the flow of lessons allows EF’s EdTech team “to optimize our content, so we can deliver better and better lessons,” says Tim Hesse, VP of AI & Analytics at EF EdTech. “We use AI to supercharge our human superpower.”  

For teachers themselves, AI is creating huge time savings by accelerating processes like lesson planning and marking. One example of where AI-powered marking is being used is EF SET Workplace, a pioneering language test which assesses reading, writing, speaking, and listening skills. It would normally take teachers two days to turn around results – with EF SET Workplace, it’s instant. 

Tim Ackroyd, VP of Creative and Design at EF EdTech, highlights AI’s potential to personalize lessons, “so that content becomes more relevant and memorable to each individual student.” He identifies further opportunities outside the classroom. It’s well known that the most effective way to improve speaking – the number one skill that students want to work on – is to actually participate in conversations, whether in a class or out in the real world. “We now have AI tools that feel very close to speaking with a human,” part of what Ackroyd calls an AI-driven “evolution of self-study.” 

All combined, AI is enriching self-study and enabling teachers to deliver even more student-centered lessons. Factor in the convenience of 24/7 availability, and its role in and out of the classroom is sealed.  

AI is capable of mastering anything rules-based, so arguably it can achieve proficiency in any language. But language isn’t fixed. It reflects our culture and evolves over time – it’s the reason Americans say ‘fall’ while Brits say ‘autumn’, and the reason why Shakespeare’s English sounds so different to the way we speak today. Language is by its very nature in a constant state of flux – a human-driven process. Can AI really capture that full breadth of nuance?

In Hesse’s view, absolutely. “OpenAI [the company behind ChatGPT], Google and other big players are already incredible because their models are trained on such a huge data set of whatever is publicly available, with enormous variation of written and spoken language.”

“These powerful systems can then be trained to assess students’ capabilities at a very fine grain, because we can analyze hundreds of thousands of hours of our own audio.” 200,000 recorded lessons at EF, to be precise – the largest spoken dataset of English learners. And because it comes from learners, “it’s imperfect, so we measure it against this huge volume of language to help students understand their strengths and weaknesses against real-life English speakers.”

“The other cool thing is how we make this regionally fair,” says Hesse. While the levels defined by the Common European Framework of Reference for Languages [CEFR] are very broad, AI allows us to see a more granular picture. “What fluency means in one region is different than another, so we can benchmark students from China, Brazil, or Sweden, for example, against people in their region.”

But even with these capabilities, there’s more than vocabulary and grammar to consider when we’re thinking about languages. What about gestures and facial expressions that have different meanings in different places? What about sociolinguistic attitudes i.e. the fact that our choice of words, grammar and pronunciation can reveal different facets of our identity, like age, gender or socioeconomic background? These unspoken elements are embedded into our communication and vary language to language, even region to region. 

As Bish puts it, “a robot has no passport or cultural legacy” – people still own the culture and community aspects of communication. “Students tell us time and time again that they want to be in a class with people from around the world, so that it's like being in an international meeting. They want to practice in the real context of work.”  

AI is capable of absorbing language patterns sleeplessly, which means it hasn’t experienced the arduous process of actually learning a language. This is something that teachers who have acquired a second language can authentically share with students, even more so than ‘native’ speakers. 

“Language learning is a painful process,” says Bish. “You have to change behaviors and personality to some degree – to be bolder. Learners need an environment where they feel safe to take risks.” And that requires emotional intelligence – the ability to interpret gestures, expressions, and other small behaviors.  

What AI can provide students with is “a safe space to build up confidence before going into the classroom,” says Ackroyd. “I think being clear about when you interact with AI and when you interact with a human is really important.” 

AI works best when it’s supporting teachers, says Bish. “In language classes, we give primacy to human interaction, which we know from research is how people actually learn.” Far from replacing humans, “we actually need more teachers because that's where the demand is – one-to-one lessons and that high touch point with teachers.”  

Ultimately, “it’s a social process” in which teachers are managing expectations, delivering feedback, and establishing trust. “We slot in three-minute conversations with AI, but we wouldn’t dream of having it control the lesson,” says Bish. “AI can spot what a student is doing wrong, but that needs tying together with pedagogical skill.”  

That skill is the difference between knowing whether a student needs a gentle word of encouragement, or a bit of space to work something out themselves. Like an efficient co-pilot, “AI is helping teachers get just the right content at the right time to power learning, then it steps aside for the professional to do their thing.”  

This doesn’t mean AI’s role ends with grammar and vocabulary. “We find learners aren’t always plugged in between lessons,” says Bish. When the route to proficiency is non-linear, it’s difficult to stay motivated.

This is where AI can step in. EF’s AI systems have the capability to generate detailed data insights into people’s language skills, in everything from grammar to confidence, based on their performance during immersive online lessons. “Helping students see smaller increments in progress is really important, to regularly demonstrate how their efforts are paying off,” says Ackroyd. “It's a bit like how a BMI [body mass index] can show you what’s healthy and what’s unhealthy.” These results can really help students to understand their performance and the effects of their study behaviors.

And just like having a running buddy, accountability goes a long way. “We found in recent research that if you have a lesson with a person, you're more likely to keep that appointment than one with AI, because you don't want to let the person down,” says Bish. “That social affective relationship is actually what makes language learning work – it’s integral.”