Language-learning tech for the 2020s
Which tech will actually cut it
I‘m aware that tech predictions and futurology are not very reliable endeavors: just look at how the future was seen in 1970’s pop culture, Jetson’s style: spacesuits for everyone, flying cars, robotic maids, and so on. But look at us now: not a single one of these visions is at sight, and the Internet, which was not predicted back then, rules our lives.
But I’ll try anyway, at least restricted to the EdTech for language learning (LL for short). EdTech, defined as the effective use of technology to education scenarios, will change very much from what it was in the second decade of this century forward to the third one.
As you know, from 2010 to now, some categories of EdTech, like learning management systems (LMS, like Blackboard, Moodle, Canvas, Google Classroom) went from experimental to the mainstream. Another category of EdTech that became mainstream was gamification (making the learner experience like a game), which has been around for a decade or so: instead of studying a boring grammar book, the language student gets immersed in an exciting video game-like experience.
But LL is somewhat special, and its tech is special too. The use of video for LL has boomed in the last decade, and it’s not unusual to find YouTube channels with subscribers in the millions. Many websites for LL integrate video content to the point that it has become a commodity.
In another post, I explained why the dimensions over which the effectiveness of EdTech can be measured in LL are just engagement and convenience. Engagement is broadly how fun and motivating is to use the method or tech, and convenience is about reducing the friction of getting and using it and is a combination of several aspects (if it’s easy to use; if it runs in many platforms; if it is not costly; if it gives you privacy so you can use it even in underwear –as opposed to video chat). For instance, VR headsets provide good engagement but convenience is hindered by costs and range of platforms on which it runs. Flashcard apps give great convenience (free, runs everywhere, flexible, private) but very low engagement.
More gamification
I know I already mentioned gamification as mainstream, but in the next years, we´ll see gamification in just about every language learning app. One app that has come a long way in the use of gamification is Duolingo. Recently I read an article by one of the Duolingo developers where he explained the so many ways they tried for getting effective gamification in their product. And yet, an independent study found that 60% of Duolingo users leave the app for good after just a couple of hours of use: this figure show extremely low engagement. To me, it proves that gamification, even if very well done, is not enough by itself to get engagement.
Voice recognition and synthesis
When Rosetta Stone started using voice recognition many years ago, its performance level was extremely low. To be fair, understanding a student learner is not easy even for today’s standards; sometimes Siri doesn’t get my phrases right (or is it the other way around?).
Voice synthesis, to the contrary, is a very mature technology. If you have asked Alexa or Google Assistant to read aloud a text you get a very natural voice, barely robotic. Did you see the video of the Google Duplex voice system, where an AI made a haircut appointment? It was an impressive show-off of both voice synthesis and recognition, along with conversation follow-up, context management and so much more. Real high-tech indeed. Not for LL though.
Currently, there are some apps with conversational capabilities, which additionally incorporate an evaluation of how well you spoke. In my opinion, the most advanced out there is ELSA, which I tried and even bought in full.
I think in the near future (this decade we are starting) we will have voice assistant-supported apps giving language lessons, especially for advanced students, because the talking part of LL is the most underserved skill right now, and the voice assistant will make you talk. The cost for the individual user will be orders of magnitude lower than with human teachers –remember, the cost is one dimension of convenience –and another convenience aspect is anytime availability, which current language schools don’t provide (in English Central and similar ones you have to schedule your session, subject to availability). Some negative consequences will be there as well, as a sharp employment reduction for language tutors.
VR — AR
A few years ago, Virtual Reality (VR) looked like “the next big thing”, making myriads of headlines, only to be later on kind of displaced by Augmented Reality (AR). The most compelling benefit of using VR is immersion: you feel like really being there (the virtual place, of course). It’s logical to assume that immersion has benefits on the engagement aspect, and it has scientifically proven to be the case. VR can also make you “live” a situation, preparing you to actually facing it in real life.
The reason why VR remains still a niche more than the mainstream is more a matter of economics than a matter of “coolness”: VR requires bulky (and costly) headsets, sometimes a high-end computer and cables, and all of this shrink the consumer base. Now, on the developer side, they need a big quantity of consumers to make their titles profitable. This is a chicken and egg problem.
AR, to the contrary, is supported by millions of smartphones that are already out there. This gives AR a huge advantage over VR. Theoretically, VR doesn’t need a headset for playing and a regular smartphone will be enough, but in reality, the kind of content played in VR is mainly videogames, which aren’t so immersive neither look so great on the smartphone screen.
There are many different ways of using VR in LL, the simplest one being to use 360º YouTube videos for the video lessons in an otherwise traditional online course, like the course of “Dynamic Spanish” or ImmerseMe. Other more sophisticated VR platforms use apps, either intended for headsets or for smartphones. In the case of LL apps, full immersion is not as important as for videogames, so smartphone apps can actually cut it.
Last year we got the very polished Mondly app (Oculus, Android, iPhone), but I guess they are having problems to get traction because their New Year’s sale is promoted with a 95% discount. In Mondly you practice by being transported to a simulated situation (for instance, the reception of a Paris hotel), and as the avatar speaks you try to speak back, then its speech recognition capability tries to continue the conversation –within a very limited choice of options. I bought the Oculus Go version (now called MondlyVR), and though the graphics are impressive (to the level of a videogame), the LL aspect is not very well developed, nor the engagement is. After a while, I felt bored with its too limited experience. I haven’t yet tried the very promising Immerse system for a headset, which they send to you with the app preloaded; it has only English learning for the moment.
VR can also be used to put together people in order to just hang out and talk, like in AltspaceVR (recently bought by Microsoft) or Oculus Spaces or Rooms. This could be of use for LL, but not using the software we just mentioned: Oculus software requires that people that want to talk first establish a friendship link. This requirement is just friction, from the LL perspective, so convenience reduction. I have myself participated in Facebook groups for Oculus owners, and it was just too hard to establish a simple Spanish practice session: find a partner, then become friends, then agree on a schedule, then hope the session will actually take place. AltspaceVR is no better: people of any level and any interests are put together, with mixed results. Anyway, the headset-based systems have very little traction and won’t get into the mainstream of LL anytime soon. Because of the VR headset struggles, many VR apps are used over the smartphones instead, obviously losing a degree of immersion, but making the user base much wider. This has been done by Mondly (the mobile version) and by Avalinguo.
Disclaimer: I’m the CEO of Avalinguo startup, an app now in closed beta, where users appear represented as avatars in a virtual room, using just the cellphone. Not much novelty here, but read below how we will incorporate AI into the mix.
AI for LL
Of course, one use case of AI in action for LL is the use of conversational capabilities in apps, that we covered above. But there are many others, not very widely used today, like automated student assessment, recommendation systems, match-making, and more.
Normally language learner assessment is made by humans. Last month I took a test session in English Central, and the teacher Shang placed me at the advanced level; we spent half an hour for obtaining a 3-bit information item (8 levels can be encoded in just 3 bits). But I have also directed the Master Thesis of some students about measuring the talking fluency and pronunciation quality of a language learner with Machine Learning (ML).
ML is a branch of AI. Many people get confused with all those terms like AI, Robotics, Data mining, ML, etc., and it’s not their fault, because AI has so many branches, and sometimes, like in the case of ML, it has in turn sub-branches, like supervised ML, reinforcement learning, unsupervised ML, etc. In our case, it’s even worse, because LL also involves the “learning” word…
So, my graduate students at the university developed systems for automatically classifying the performance of a speaker into 3 categories (like basic, intermediate, advanced). They do so by first building a dataset of already classified speech audios, then slicing the audios into segments, then for each segment calculating the numeric “features”, then training an ML classifier with the features vectors. I don’t want to get too technical here, the folks who are into ML already know what I’m talking about, and the rest could take a free ML course at Coursera, EdX or Udacity. Anyway, the results of my students were impressive: the trained classifier predicted the right level of a given language learner with an accuracy well above 90%. Automated speech assessment could sound just like another technological feat, but it could have a big practical impact on lowering the cost of the initial student-level placement.
Another aspect that could be improved with AI is match-making, for instance automatically placing people in virtual rooms in such a way that they share common interests, and engagement is likely to be improved. Compare the convenience of automatically find a group with compatible levels and interests with the friction of using Oculus Spaces, where you have to manually find a partner, then register as friends, then agree on a day/hour, then show up, etc. You guessed it right, we’ll use AI-supported match-making in Avalinguo.
We can conclude that AI is going to play a major role in the LL tech to come in this decade we’re starting. Every single aspect of LL, from student placement to fluency assessment will be done with AI intervention. This doesn’t mean that tech is enough for good learning, you need also good educational principles, but these principles can be enforced and made practical with the help of the tech.
So, what will we see in the 2020s?
All of the techs we covered here (voice assistants, VR/AR, AI/ML, together with gamification) will play a major role in shaping how we’ll learn languages in the 2020s. In many cases, those techs will go combined, like having VR avatars with AI matchmaking and ML assessment in a gamified platform (which is what Avalinguo intends to be), or immersive roleplaying in VR with conversational voice assistants (lightyears beyond Mondly).
What do you think? Does this sound like a “Black Mirror” dystopic episode or as a bright future? Please write below in the comments.
