Lost in Translation: Duolingo Makes Language Learning Mobile

Duolingo democratizes language learning to create a crowdsourced translation business.

Ah, language. Perhaps the most human and most analog thing around. Think back to when you first learned a language. Oh wait, you probably can’t remember because you were a baby. Alright, think back to when you first learned a second language. If you were like most Americans, you probably sat in a classroom with a blackboard and drilled verb conjugations. But the traditional classroom approach doesn’t work. Indeed, only about 25 percent of Americans can speak a foreign language and only 7 percent of those learned that language in school [1].

Enter Duolingo. Launched in 2011 by the same man who invented Captcha [2], Duolingo’s mission is twofold: first, they want to make free, high quality language learning software available to everyone. Second, they want that software to be adaptive to each and every user, such that language learners receive a uniquely tailored educational experience. This is delivered through a mobile app that serves up gamified language lessons for use on a daily basis.

While the aims are noble, how well does a mobile app replace a language teacher? Just as importantly, how does a company offering “free language education – no hidden fees, no premium content, just free” [3] make money?

The Product

img_2032
The “Skill Tree” which maps out your journey to fluency. Source: Author

So, if you’re the average Duolingo user, what exactly do you get when you fire up the app? Duolingo breaks down a language into compact and discrete modules, covering both thematic content (food, clothing, colors) and grammatical concepts (the preterite, question words). Users run through a battery of vocabulary quizzes and translation, conjugation, and pronunciation exercises (see the author’s feeble attempts at Spanish below) that build on each other. As you progress through the exercises, you’re rewarded with “Lingots,” a sort of gamified currency which you can redeem for some, um, interesting in-app digital prizes, such as a “flirting” lesson in the target language, as well as a new outfit for the owl mascot. Kind of lame, but the joy is in the learning, right?

Attempting to learn clothing and colors. Source: Author
Attempting to learn clothing and colors.
Source: Author

The Business Model

So, we get that Duolingo is a fairly cool way to learn a language. And there’s all that great talk about “free and open for all.” But how does Duolingo keep the lights on? The answer comes back to that old adage about the Internet: if you’re not paying for the product, you are the product.

Duolingo might seem like a straightforward language learning app, but it’s in the business of translation. Turns out, all those exercises and practice sentences generate an enormous amount of crowdsourced data, data which can then be packaged and sold as translations to sites such as Buzzfeed and CNN [4]. That’s right, all those important and ground-breaking articles (15 Hedgehogs With Things That Look Like Hedgehogs, What Is Your Inner Potato, Which Ousted Arab Spring Ruler Are You?) are translated by Duolingo users for the entire world’s edification.

While the company is still young (and because they don’t disclose financials), it’s hard to say how effective the model has been. Globally, translation and language services is a huge industry, worth approximately $40 billion annually [5]. On the old-fashioned side, most translators available today charge a relatively expensive fee (around $0.21 per word [6]) for their craft. Duolingo is betting that it can produce translations of comparable quality at a fraction of the price, as their translation staff (that’s you, by the way) works for nothing but Lingots. If traditional translators are the skilled artisans, Duolingo and its legion of 120 million users [7] want to disrupt the industry with mass production.

The virtuous Duolingo Cycle Source: Author
The virtuous Duolingo Cycle
Source: Author

But this whole model is predicated upon two big assumptions. The first is that crowdsourced translation can hold up against bespoke professional translation. The second is that enough users will keep coming back to Duolingo for language instruction to satisfy demand for translations. The first question is hard to answer, as it’s yet to be proven that there’s a market for crowdsourced translation. Recent moves by Duolingo towards other channels of monetization, such as fees for language fluency exams [8], suggest translation consumers have yet to come around.

However, the second question seems to be testable even today: does Duolingo work for learning a second (or third, or fourth) language? Duolingo commissioned a study which found that 34 hours using Duolingo was the equivalent of one semester of a college-level language class [9]. There’s likely just a bit of bias there and professional educators have much more mixed reviews [10]. Certainly, if Duolingo is to keep growing little linguists, such that an army of translators can rise from the crowdsourced depths, they’ll need to keep delivering a consistent and high quality product that both works for users and incentivizes them to contribute more and more data. If they can’t, it might be the traditional translators who get the final word.

 

 

Word Count (797)

Sources:

[1] Pew Research Center: Learning a foreign language a ‘must’ in Europe, not so in America

[2] The Verge: Duolingo will translate the internet while teaching languages

[3] Duolingo Company Site

[4] PC Mag: Duolingo Users Translate BuzzFeed, CNN Into Spanish, French

[5] Common Sense Advisory

[6] USD 0.21 per Word: America’s Translation Rate

[7] Pittsburgh Business Times: Duolingo moving to East Liberty, plans to add employees

[8] Duolingo English Test Info Page

[9] Duolingo Effectiveness Study

[10] International Journal of English Linguistics: Learning a Language for Free While Translating the Web. Does Duolingo Work?

Previous:

Telematics: How Flo is Tracking Your Every Move

Next:

From Paper to Parametric Vectors: Gehry’s Transformation of Architectural Modeling

Student comments on Lost in Translation: Duolingo Makes Language Learning Mobile

  1. Thanks, Conor, for this interesting topic. The move of language learning to online/mobile platforms does not seem to be a great jump from the Rosetta stone audio era a decade or so back. I do find especially interesting 1) the effectiveness of such a pedagogy and 2) the monetization/use case.
    On the first, as someone to whom second language learning did not come naturally, I believe that digital methods can be very helpful in making the routinized part of language learning (especially vocabulary memorization, grammatical conjugations, etc.) more engaging. Digital also seems to be a big boon for children, to get them excited and engaged. It also enables adults greater flexibility than having to show up at a certain class at a certain time. However, digital modes alone will not get people to language proficiency (think pronunciation, odd turns of phrase, grammatical irregularities, linguistic manifestations of culture…). The regular practice in person with a language partner or class (or full immersion) seems necessary, though digital services like duolingo can help interested individuals get part of the way there on their own time.
    The monetization piece is also fascinating. First, it’s slightly frightening that they are using language novices to translate, but hopefully with a great enough n the group will prevail! Second, as you mention, it is quite reminiscent of (re)captcha’s orientation of solving a business need but also having a broader purpose, in that case captchas were used to digitize books. I am somewhat skeptical of web translation holding up as a revenue stream, but I am impressed by the model’s creativity (especially in today’s online ad-driven ethos). Others in the space, like Babbel, use freemium models to try to collect premium client revenue [1].

    [1] https://techcrunch.com/2009/04/06/when-startups-monetize-babbel-switches-on-paid-courses/

  2. As an avid user of Duolingo (and someone who enjoys buying new outfits for the Owl mascot), I actually had no idea that they were generating revenues via crowd-sourced translations. To answer your second question – I see the value of Duolingo in its ability to meaningfully develop someone who already possesses a foundational understanding of the language, but I’m skeptical of its benefits to someone who’s never had any prior exposure to a language. For example, I use Duolingo to brush up on my French (I’ve taken several years of French in school), but found it utterly unhelpful when I was attempting to learn Russian for the first time.

    Separately, I’m interested in Duolingo’s decision making process in choosing the languages to build modules for. For example, Esperanto and Klingon (in development) are offered, but not Japanese or Chinese?

  3. Fascinating to hear translation may become a key revenue stream for a company like Duolingo. I fear the quality of the translation would not be better than Google translate. Would Duolingo be able to provide effective quality assurance for the translation? Furthermore, with sites like upwork (https://www.upwork.com/) already significantly reduce the price of translation by native speakers using a large freelance network.
    While I am worried about the revenue stream, I would like to find a way to keep Duolingo alive. Perhaps integrating it to formal education with public support is a viable option here?

  4. Thanks for an interesting post. I’ve been following Duolingo since it launched and even used it to learn Spanish. However, after a few weeks, the game mechanics failed to work in my situation and I dropped off. Given my personal experience, it led me to think that Duolingo has issues with user engagement and retention. However, I was wrong and was surprised to learn that they crossed 110mn engaged users earlier this year [1]. Moving from engagement to monetization, Duolingo has succeeded in using an ad-free business model so far but I don’t think it is sustainable. Having worked in the SaaS area for a long time, it’s essential to have diverse revenue streams to ensure the longevity of the company. Thus, I believe Duolingo will soon need to reconsider their decision of keeping the app free of ads. Another potential revenue stream could be premium subscriptions. With a premium account, users could take notes when they learn new concepts and even when they make mistakes. By storing user generated data, Duolingo could create stickiness just like Dropbox and Evernote have with their applications.

    [1] Velayanikal, M. (2016, March 03). Connecting Asia’s startup ecosystem. Retrieved November 20, 2016, from Tech in Asia, https://www.techinasia.com/how-duolingo-got-110-million-users

  5. Connor, thanks so much for the interesting and witty read. I especially enjoyed your Buzzfeed example! Given the above comments, it sounds like there is quite some variability on the usefulness of the platfrom. However, I’m not sure if I found that all together discouraging. Language learning is indeed a process, and Duolingo seems to aid in that process in different ways for different users. On a larger note, I really appreciate your highlighting and describing a business model that monetizes the contributions of its free services to the masses. I’ll be keeping my eyes open for when I’m the product!

  6. I’ve used Duolingo but had no idea this is how their business model is structured. It reminds of what ReCaptcha is doing and clearly demonstrates the power of large groups of people. Essentially what ReCaptcha does is it gives two words when it needs to authenticate a human. One of the words is know by ReCaptcha, the other it uses inputs from a large number of humans to figure out what it is. In this way ReCaptcha is transcribing several old handwritten books. It is very interesting the Duolingo is following a similar model to harness the knowledge of large groups of people

    1. Thanks, Shiv! The folks who invented Captcha and who also work on ReCaptcha are the same people behind Duolingo!

  7. Echoing several others above, I am also a former user of Duolingo and was really interested to finally understand their monetization strategy. I’m interested in digging a bit more into your second assumption- is it absolutely necessary that people come back to the product in order for them to collect sufficient data to generate meaningful translations? Could they achieve the same goals by focusing on breadth rather than depth, especially since stickiness within an app is one of the hardest things to accomplish?

    1. I think this is where pedagogy comes into the model. It doesn’t work if you only have beginning-level users who have a narrow range of linguistic abilities translating simple words like “red,” “horse,” or “bicycle.” Translation involves nuance and judgment and those skills can only be developed over time and with practice. Certainly, on a per-person basis, if you’re crowdsourcing translations you need less judgment and skill, but, in aggregate you still need users with a wider range of abilities, especially as the concepts being translated move further and further away from the introductory material. It might be interesting to see how Duolingo could reconfigure their lessons to cover the materials they need translated though. It might fit into their adaptive strategy.

  8. Connor, thanks for such an interesting article! I am the user of Duolingo but I never know I’m also the “translator” working for Duolingo! This model is self-sustaining, but I’m a bit worried about the quality of the translation and its impact on readers. On the other hand, I’m also intrigued by the process – how is Duolingo selecting the translation output? I think there’s opportunity here for Duolingo to better utilize its courseware and the ratings or information of each user, so that only the contents from high-rating users are selected and released as the “product” to the public.

  9. Thanks Connor for this very interesting post. I didn’t know about the existence of Duolingo and will I certainly like to download it and start using it. I think that you make a very interesting point in your first question and I would say that as a reader of an article, magazine or book would feel less confident in what I am reading if I was told that the translation was done by a crowdsource translation instead of a professional translator. It is important to understand how they are certifying the quality of the translations and how they plan to maintain that quality after they start to grow at a faster pace. If they are able to do this well, then their value would be impressive and actually it could change the whole translations industry.

Leave a comment