During the stay-at-home grim days of 2020, I started learning Spanish on Duolingo. Having a working understanding of Spanish seemed like a sensible first step towards opening a taco truck in Mexico, in case I had to run away from my doctoral studies. This July, after about 5 years I decided to end the 1800 day streak that I managed to drag on with numerous streak freezes and minimal effort lessons. While Spanish words look less foreign, and with some focus, I am able to decipher small paragraphs of grammatically simple sentences; the effort was less than a smashing success – I certainly could not be writing this essay in Spanish.

If Duolingo is known for anything, it has to be their gamification approach. There is no shortage of gamification mechanics on the platform: XP; potions to double your XP; leagues of gold, silver, and all sorts of other metals and minerals; treasure chests; quests; monthly quests; and so on. I have only ever paid little attention to these mechanics. While I still haven’t entirely rejected the idea that a good RPG could be a good scaffolding to teach a language, I do not think Duolingo is one.

Games worth their salt are not created by bolting together a collection of numerical statistics. That is how you get cookie clicker. I did not have a good understanding of how the mechanics work: if I learn 10 words, how many XP do I get for my hard work? Is the Diamond League higher or lower than the Obsidian League? I could have viewed their documentation to figure it out, but there was nothing motivating me to do so. If I collect 100 XP, what does it mean for my language skills? For that matter, why do I collect extra XP when I receive a potion? Can the XP I collect be used in a way to carefully guide me towards the specific language skills I would explore next? Navigating the mechanical gameplay of Duolingo is neither rewarding for its own sake, nor is it helpful towards actually learning a language.

Duolingo is not just a poor simulacrum of the mechanical aspects of a game, but also of the social aspects of one. Who are all these people I am on the Silver league with? Having a comparable amount of XP does not give me a sense of social connection with them. When I click a button to congratulate a friend on Duolingo, I do not truly engage with their learning journey. Indeed, it is worse than hearting an instagram photo, or upvoting a reddit thread. In those cases, I am reacting to a sliver of expression from my acquaintance. Here, I am presented only with a pre-rendered text with an abstract numerical statistic. Reacting to it is deliberately frictionless: I am presented with a wall of buttons allowing me to click them with ease and without thought. When Duolingo tells me that so and so sent me a message saying “Hey, come back and learn Spanish with me!”, I don’t admire how thoughtful and encouraging my friend is; I just notice that they clicked a button to send me a pre-generated message.

Interactions on Duolingo were not always of the push-button variety. Duolingo had forums where users would discuss different aspects of their language learning journey. In fact, Duolingo would link each sentence to its own forum thread for discussion – discussion, which was at the very least, helpful, and at times, eye-opening. At first, these discussion threads were locked, and later removed. My hypothesis is that for the business geniuses running Duolingo, the forums were assesed to be a liability, having to moderate which were not worth spending the dollars for. The nature of interacting with people – friends or strangers, in person or online – is that sometimes bad things would happen. Even when there is no abuse or harassment going on, there is always the risk that the other person might greet you with disagreement, or worse, apathy. Many people tend to think that the risks outweigh the benefits.

The gamification mechanic that I did latch on to was the Streak. I generally have been critical of the green owl, but I do think that it did help me form a good habit – a net positive, despite the minuscule magnitude. Regular Duolingo users will know that the streak can be gamed away in more than one ways. Streak Freezes can be bought using gems (of which I happen to have 24,053 of, somehow) or be gifted by your friends, and equipped 2 at a time. Streaks wouldn’t have their social effect if there weren’t enough people with a moderate number of people with decent streaks to be sprinkled around. Maintaining the streak, even without freezes, does not have to mean that you are learning – repeating a simple lesson from several units ago would work. My 1800 day streak didn’t mean that I spent 1800 days learning Spanish; it meant that I spent a large number of days engaging with the platform. I later started peeking into the Japanese and Finnish courses, and the 1800 day number includes them. If you loose interest in languages, Duolingo tells us that spending time with math or music will count towards your Streak.

The deficiency of Duolingo’s pedagogy was first made obvious to me by the excellent audio lessons produced by Language Transfer. Going through the first few lessons of Language Transfer, I was unfazed, observing that I had already learnt what was being taught. Soon, what shocked me was how quickly Language Transfer caught up to what I had managed to learn in a couple years of time. While Duolingo is great at making sure that the user comes back to the app everyday, their pedagogy is subpar. Remaining true to gamification, Duolingo prefers to throw users head-first into translation exercises. If you do not know a word, you hover over it and you arrange a given bag of words into a sentence that is hopefully meaningful. Grammar lessons are extremely minimal. The removal of the forums dedicated to the discussion of specific sentences did not help. Understanding the course outline – knowing what is taught where, or reviewing lessons – is not easy.

Supposedly, the Duolingo philosophy is that if you are exposed to enough sentences, you will eventually learn how to use them. I do believe this is true, and I do believe the exercises are indispensable. However, the whole process could be greatly improved by a few more lessons interspersed in the curriculum telling the student what is going on. To see this, I would ask myself, did I always internalize the grammatically correct structures even in my own mother tongue? I think not, and my language skills have improved when my parents or teachers would point out simple grammatical mistakes. Eggcorns are a closely related amusing phenomena.

I cannot tell if Duolingo repeats different concepts in exercises adaptively based on your mastery, or are simply fixed in the course material. Repetition is good for learning but Duolingo’s repetition can be frustrating. The platform’s interface is largely built around clicking a bag of jumbled words one at a time to input a translation. Once you learn a concept well enough, most of your mental energy is spent on the finding and clicking of the words rather than the translation. Thankfully, this situation can be made better by dictating your answer, as pointed out by the official blog.

Duolingo does include a few other formats for their lessons. There are some stories, which are interspersed in the course. The stories are short, but silly and enjoyable. There are also audio only lessons, which are also shorter and unfortunately, not as fun. From time to time, the regular lessons also ask you to speak to the microphone but in my experience, the audio recognition seems to accept the answer even if I mumble through the words. Duolingo is also known for its usage of bizarre phrases, whose shock value generates social media buzz and may or may not have a positive pedagogical effect.

Duolingo is a neat case study in Silicon Valley ideology. Big tech embraces blitz-scaling: the primary goal is neither financial sustainability nor the quality of materials but making the number of users grow. The faux gamification and passive-aggressive messaging may be helpful with little else, but is good for user retention. The expansionism does not stop at growing the number of users; Duolingo has decided that they must loop in music and math learners as well. As we have discussed, the maxim of friction reduction has guided them towards optimizing away authenticity in the user interactions on the platform.

In April 2025, Duolingo decided to go AI-first. Supposedly, “to teach well, [they] need to create a massive amount of content” – so much so that “doing [it] manually does not scale”. For the top ten languages, I cannot imagine any reasonable person saying that the lack of study material is the main obstacle towards learning. This statement spells out what the Duolingo executives value. The Duolingo CEO is not shy to admit it. In an interview with NPR, he said the following.

[I]f it’s our content, as in, like, our learning content, there’s so much of that - thousands and thousands and thousands of kind of sentences and words and paragraphs. That is mostly done by computers, and we probably spot-check it. But if it’s things like the user interface of Duolingo, where we say - like, you know, the button says quit, and we have to translate, that is all done with humans. And we spend a lot of effort on that, but that’s because each one of those is highly valuable.

Yes, the button that says ‘quit’ is more valuable than the learning material, which is only ‘probably’ spot-checked.

After I moved to Japan, I dialed up my efforts to learn Japanese. For a while, I shifted over my Duolingo habits to Japanese. Because Duolingo wasn’t my only learning material for Japanese, it was glaringly obvious very soon that the Duolingo pedagogy is unhelpful and often misleading. While the Spanish learner has to introduce themselves to a few new concepts (e.g, ser vs estar, or reflexive verbs), the Japanese learner faces an explosion of differences. Japanese has a writing system with three components; generally uses a topic-comment structure and often omits the topic; has a subject-object-verb order; has adjectives which conjugate; a lot of counting suffixes; sentence ending particles and famously, a complex honorific system. Duolingo does not break its gamification façade to teach the user some of these concepts head-on. Instead, it pretends that translating between Japanese and English is a matter of substituting phrases and shuffling them around.

Since I gave up on my Duolingo streak, I have started exploring other avenues to continue learning Japanese. I participate in group lessons with a tutor once a week for an hour. Believe it or not, the tutor has more charm than Falstaff. I regularly do my flashcard kanji study with Wanikani. A newer addition to my study routine is Bunpro. My progress has been slow but evident: when I recognize that the names of the metro stations I frequent break down into simple words, they lose a little bit of their mystery but it is a satisfying revelation.

These platforms are a welcome contrast against the techno-accelerationist attitude of Duolingo. Instead of trying to do it all, they are extremely niche: they only teaching one language and Wanikani is focused at teaching a very specific element of it. Wanikani maintains a public API, which makes third-party apps and scripts possible. I praise them for their welcome attitude towards interoperability instead of trying to build a closed ecosystem. Both Wanikani and Bunpro have vibrant user forums. Bunpro makes actual lessons part of their critical path, instead of hoping that the user will eventually figure it out. When a Bunpro user feels that their lesson was not adequate, they do not have to rely on AI generated slop – Bunpro directs users to carefully-crafted lessons by other people (see the ‘resource’ section at the end of this page, for example).