published
Duolingo is categorized as an Education app in the App Store. In fact, they claim to be the "most downloaded education app". I take issue with that. Duolingo is not an education app, it is a game.
You may say that is a good thing. You may say it is both. You may say I'm being pedantic about categorization. My position is simple: if Duolingo wants to call itself an education app, it should be good at teaching people a language. However, the ways in which they 'gamify learning' are detrimental to learning. Driven by short-term monetization incentives, Duolingo chooses to sell a self-improvement fantasy and an ad machine, instead of the language tutor they market themselves as.
Duolingo is built to entertain, rather than help you gain fluency, because they realize they derive more value from you by keeping you hooked. Which brings us to...
Gamification
At its core, Duolingo is a fancy flash cards application. It has some built-in categorization of words and leverages audio and microphone features on your device to help with pronunciation (nominally. no dialect training or detection). They will ask you to fill in a blank or move some blocks around to reinforce sentence structure, and sufficiently advanced users will be asked to translate conversations and answer complete questions.
Lets break down the in-game mechanics of Duolingo: Currency/Reward (Gems, Chests), Engagement Drivers (Streaks, Quests, Leaderboards), and Punishment (Hearts).
Rewards and Engagement Drivers are meant to keep you coming back. Perversely, users aren't going to Duolingo to learn the language. They go because they feel compelled to 'keep the streak going', or 'stay high up on the leaderboard'.
When Duolingo offers rewards to users for doing what they were already going to do (people download the app through intrinsic desire to learn), users start to value the rewards i.e. streaks or leaderboard position rather than the attainment of language.
anything presented as a prerequisite for something else—that is, as a means toward some other end—comes to be seen as less desirable. “Do this and you’ll get that” automatically devalues the “this.” The recipient of the reward figures, “If they have to bribe me to do this, it must be something I wouldn’t want to do.” Alfie Kohn (Punished By Rewards - p76)
This means that the learner's intention at time of engagement is not to learn language, since they have been distracted by the reward i.e. the motivation is no longer intrinsic which lowers retention and efficacy. This research has been around since the 90s and has been intuitively known for much longer.
People who have been led to think in terms of what they will get for doing something can be described as extrinsically motivated. The opposite of this is intrinsic motivation, which basically means enjoying what one does for its own sake. If our goal is quality, or a lasting commitment to a value or behavior, no artificial incentive can match the power of intrinsic motivation. Alfie Kohn (Punished By Rewards - p68)
Chests are rewards you earn for logging into Duolingo every day, learning their way, and telling all your friends about it. They are meant to control user behavior, besides being another extrinsification of intrinsic motivation. They reinforce behaviors that benefit the application, not the user. For example, you can get a chest for 'staying on your learning path', but that is the same path that Duolingo lays out for everyone else. What if it is not right for you? They don't want to hear about it, so they distract you with a shiny Chest.
What rewards and punishments do is induce compliance, and this they do very well indeed. If your objective is to get people to obey an order, to show up on time and do what they’re told, then bribing or threatening them may be sensible strategies. Alfie Kohn (Punished By Rewards - p41)
The velocity with which Duolingo hijacks intrinsic motivation and turns it into rewards-based-blackmail is shocking. Transforming learners into users (in the worst sense of that word).
Punishment (Hearts) - You lose them when you make mistakes. If you run out of hearts, you can't continue the lesson. Purportedly a mechanic to prevent guessing and exhaustion. But, people learn through mistakes. The exact moment of observing a mistake, is an ideal time to correct and practice. But in Duolingo, you are penalized and told to go sit in a corner; no matter how motivated you may be to practice. Well, except you can buy more hearts. Watch an ad (distract yourself) to earn hearts or spend gems (rewards or real money) to buy hearts. So what is the point of the hearts except for feeding the capitalist owl its gems?
(Here’s a video of a first-time user surprised to see he is immediately being asked to pay for this free app.)
Imagine your language tutor telling you the lesson is over because you made 3 mistakes. "Come back tomorrow!"
Everything is a Distraction
The thing is... this form of gamification is not about learning, it is about monetization. Duolingo wants you to either look at ads, make in-game purchases, buy a subscription, or just log in every day so that they can flout their Daily Active Users numbers to shareholders. They could not care less if you learn a language.
This is evident in the first minutes of using Duolingo. Every time you finish a lesson, you are thrust into 2 ads and 4 additional screens showing statistics about changes in engagement drivers resulting from the lesson you just finished. The app's usefulness is hindered by the half dozen screens/checkboxes/distractions you have to hurdle over to get from one lesson to the next. Duolingo knows that focus is important to learning, yet they don't let the learner stay in focus to internalize what they have just learned.
After the very first lesson, these are the screens I had to go through so I could keep learning.
I thought maybe it was a one-time thing, so I did the second lesson and had the same set of intermediate screens. Then the third. Same result.
I installed Candy Crush Saga, the most popular mobile game I know of, to see what the gaming industry standard for ads is. I've played 8 levels and I still have yet to see a single ad. Maybe they are trying to hook me and then show me ads (makes sense). Maybe they are extremely confident with their in-game purchases strategy. Either way, I was pleasantly surprised by the initial user flow of Candy Crush Saga.
Duolingo is leaning on both strategies (ads and in-game purchases), which is common for gaming applications. It does so relentlessly, and without having actually drawn the user into the learning (gaming) experience. Its reliance on these strategies for monetization hinders the experience and stated objective of the application.
While Duolingo can serve in language acquisition, all the bells and whistles to drive engagement tend to hinder learning and retention. It seems they know this and don't care or, worse, actively count on it to keep people stuck in the app. If you successfully learn the language, you may stop coming back to watch ads. The incentives are misaligned.
Duo-better, Please
Apologies for the cheesy section title.
Duolingo has a truly profound opportunity, given its user-base, to create a culture of polyglots. That is, if they ever decide to move away from being a pop-culture game to being an actual language learning application. The argument could be made that people are only hooked to Duolingo because of the gamification aspects. That the gracious and generous owl really does care about language acquisition and the priority is the daily practice. Sure. So, what can they do now that they have this captive audience and $50,000,000 in R&D?
So much.
-
Um...flash cards. Duolingo is not even a good spaced repetition app. The questions are not well designed, the repetition algorithm is unclear to the user, and where are the pictures? There are studies that show including an image of the object/concept you are learning the vocabulary for can increase retention by providing another anchor point. They could add a free play mode where they let users choose categories, or frequency of repetition. There are several common sense features that a flash cards app should have. Build those.
-
Show users real data. Streaks and Leaderboards are smoke and mirrors. Its a language learning app, so show me what areas I am excelling in and what I struggle with. Then help me get more practice in concepts I need support with—in app and in the real world. Show me data like what words I have learned in the last week/month, and how my retention is affected by practice. This is Ed Tech 101. Bring me along on the journey of my own learning, please!
-
Give users control. The learning pathways in Duolingo are 1-dimensional. Literally. They show you a progression path, and it is a line from lesson to lesson. We all know that learning is more complex than that. Help me formulate and explore my unique and nuanced learning pathways.
-
Conversation. How does this multi-billion dollar application not teach people language in the one way we know works? Through conversation. Many adults' language learning goal is to get to a conversational level. Conversation requires presence, listening, processing, and recall. An environment that flash cards drills cannot replicate. We can see that Duolingo's primary motive is profit, but even so there is economic opportunity in creating a marketplace of language tutors. Now, since AI has become an important fundraising buzzword, Duolingo is all about conversation, aiming to automate away the need for language tutors. (More on this in Note on Gen AI below)
-
Provide meta advice. Learning a language requires deep engagement with the concepts/vocabulary. It requires embodiment. Not all learning will happen in app, and that is okay. Duolingo can help create mindfulness around language acquisition outside the app. A notification like "Think of the Spanish words for objects you encounter today" or "Try to use the phrase 'Comme-ci comme-ça' in conversation today." How about pointing out etymological commonalities? It's not hard. And they supposedly have a team of pedagogical experts working for them?
This list is common sense, inexhaustive, unordered, and low-depth. Just a taste of what Duolingo could develop if they care to accelerate your language journey.
In their most recent earnings call, the CEO spent 90% of his statement talking about Revenue, Engagement, and the Market. Even when he was talking about language learning, it was through an economic growth lens. Maybe they care and are codeswitching because this was, after all, an earnings call. But, I doubt it.
A note about Play
So far, this has been a specific analysis of Duolingo's use of Gamification which, to me, is particularly offensive. Based on some of the evidence about reward and punishment cited above, I'm generally skeptical of the Gamification of Education.
Play, on the other hand, is an effective mode of learning. Gamification dilutes the learning experience by creating distraction and diverting intrinsic motivation. Play enhances the learning experience by directing focus to the concept itself. Allowing the learner to turn the concept into clay, shape and reshape it, and work with it from every angle. Deepening understanding in the process. With guitar, for example, it is why something like Yousician is ineffective, and actually playing the guitar yields better results. I hope to write more about Play.
Any Yanomami father knows that you don’t have to force young children to learn, you just give them the tools they need and then let them play. Carol Black - A Thousand Rivers
Minecraft, Universe Sandbox, and the Montessori classroom are all good examples of the power of Play in learning.
Please reach out if:
- You know Alfie Kohn and can make an intro :) I want to get his thoughts on Gamification
- You work at Duolingo and want to have words
- You don't work at Duolingo and want to have words
P.S. I intend look at less egregious instances of Gamification (ClassDojo, ClassCraft), and will share what I find to be interesting.
A note about Gen AI in Language Learning
Nothing beats an actual language tutor, who knows you, will prepare lessons catered to you, and adapt to your needs without some engineer somewhere having to update a prompt for it to work—for now.
Many, many people are trying. Including Duolingo, who thinks their market position gives them an advantage in the Gen AI Language Tutor space. But should you really pay $15/mo or whatever it is for Max?
I wrote a prompt for turning your Gen AI Chat into a language tutor. I've tested it on (best to worst): Llama 405B, Llama 3.1 70B, ChatGPT-4o, and Gemini. Feel free to use it. You can provide it with more instructions within your chat context. If you keep the chat alive and come back to it, you will have a language tutor that has some history of your interactions. The more you use it and tweak it, the better it gets.
You are a language teacher. You have extensive knowledge of the Spanish language and grammar, as well as different dialects in different regions where Spanish is spoken. You are patient but also a bit stern. You pay attention to the responses you get, and drive your students to improve. You will have several modes:
1. Conversation Mode
This is your default mode
You will have conversations with your students in Spanish. You also understand that your students are learning conversational Spanish, and don't necessarily need to be grammatically perfect as long as they are speaking clearly and are understood. You will teach them colloquialisms when necessary. Sometimes they will have questions, you will answer these questions patiently and then continue with the conversation.
By default, Provide the English translation of your Spanish statements in parentheses. If the student says 'Stop Translation', stop showing these. If the student says 'Start Translation', start showing these again.
Instruction: When your student uses an English word in box brackets, it means they don't know the Spanish translation of that word. Provide the translation and then continue the conversation.
Instruction: Feel free to correct the student when they make a mistake related to masculinity or femininity of the vocabulary
Instruction: Feel free to provide corrections for your student's spelling as well.
Instruction: Create a corrections section at the top of your responses. This will contain corrections, answers to the student's meta questions, translations for words they didn't know, etc. Use a code block for the corrections. then make a divider and continue the Spanish conversation.
Instruction: At the start of conversations, let the learner know what the objective of the conversation is.
Instruction: Remember the learner's questions and challenges for future conversations and sprinkle review moments into conversations as you go along.
2. Vocabulary Mode
The student can ask you to enter vocabulary mode by just saying Vocabulary Mode in chat
In this mode, you will act as a spaced repetition application. You will introduce words to the student one at a time and test them on previous words you have introduced to them. For each word, you introduce, you will provide at least 3 and at most 5 example usages of different contexts, conjugations, tenses, etc in which the word can be used. But never more than 5 examples total in a response.
Show only the spanish word first and wait for the user to write the English translation (or vice versa), then provide the correct translation and examples.
Regardless of the topic the user wants to learn vocabulary for, you should feel free to test them on past words you have taught them in Vocabulary Mode or in Conversation Mode.
I haven't found a good foreign language Speech to Text implementation built into a Gen AI Chat (shocking, considering Google Translate exists and works very well for translating conversations). So that would be Duolingo's only 'moat' as the techies like to say. Especially if they can train a model to recognize bad/accented pronunciation.