Duolingo: A 1930's Method for a 2030's Platform

TL;DR: It Sucks

Duolingo sucks. It's the product of pure programmers trying to be educators. There's no innovative idea. There's no modernly researched approach. There's just naivety. And naive programmers do what they always do:
  • What seems intuitive to them, or
  • What others have already done, so long as it's easy to code.
Now, my series against top-down approaches should make this clear: Programmers' intuitions about language create crappy models for natural-language learning. To explain why, I'll start with a little bit about me:

As I've mentioned before, I trained for years to become a logician. That means I learned the same intuitions programmers apply. I spent years becoming familiar and competent with such top-down definitions and rules. But, I also learned how they restrict admissible natural language for limited, formal purposes. Relatively few logicians work on expanding logical expressiveness, anymore. Montague was probably the last major logician who did. However, Montague's approach turns off most programmers. They're perhaps right to not like it. It's not an elegant solution. Programmers don't want a programming language that's as complex as the language they want to model. Well, too bad for them. That's the deal. If a natural language is higher-order, only a higher-order logic will capture it.

So, these pure programmers are stuck in AI purgatory. The correct leaps go against their likely intuitions, and their intuitive leaps go against what's likely correct. And, it's obvious that Luis von Ahn and Severin Hacker were programmers first and applied linguists a distant, distant, distant second.

I get it, Duolingo programmers. It's hard. You nearly have saw your leg off to escape it. Your created worlds are so elegant. Then, some Italian flips you off, and it fucks up your whole universe:

Una salida elegante para von Ahn...
"Wittgenstein was insisting that a proposition and that which it describes must have the same 'logical form', the same 'logical multiplicity'. Sraffa made a gesture, familiar to Neapolitans as meaning something like disgust or contempt, of brushing the underneath of his chin with an outward sweep of the finger-tips of one hand. And he asked: 'What is the logical form of that?'"

Okay, so they'll hire some linguists, and we'll get a better solution, right?

You'd think so. You'd hope so. So far, though, no. It's clear at this point that Duolingo has been institutionalized. Even if the best SLA experts at their company promote a paradigm shift, it will probably never happen.

This is because of the designers' second naivety. They didn't come up with any original pedagogy. They didn't even research modern language pedagogy well. Instead, they stood on the shoulders of midgets.

Anyone who's read a foreign-language textbook since the 1930's recognizes Duolingo's structure. It's a bastardization of the situational and structural syllabus (SSS), and we've long known that there are shitloads of problems with it.

SSS Treats Humans Like Programmable Devices

An SSS is a drill-and-kill approach. It has you practice one grammatical structure or feature, 20 or so vocabulary terms, and a few key phrases. Then, the rest of the syllabus assumes or ignores it. That's only a sane strategy if you're a computer. Real humans have flimsy memories. They need reinforcement, but not blind repetition. Writers of SSS materials rarely address this issue. (They're often not paid enough to give a shit.) Yet, even when they do…

Every SSS Is Full of Prescriptive, Unnatural Language

When SSS developers create their materials, they usually follow this protocol:
  1. Pick some situational keywords or grammar features.
  2. Shove the elements of (1) them into a situational dialogue or narrative.
  3. Build exercises around the unnatural product of (2).
Just think from your own life. Imagine writing a short story. It's a good, interesting story. Then, some asshole comes to you and says, "Hey, make sure the word 'falafel' is in there at least three times Also, it needs to contain at least four sentences with prepositional-phrase complements." The correct response, of course, is, "What? Why? Who the hell cares if that stuff is there or not!" One correct answer: editors of language-learning magazines and textbooks. Another correct answer: retards.

Language learners recognize how fake and stupid SSS materials are. They endure them, though, for their  perceived benefits. They also have few other choices in texts and apps.

But, it's not like they have to do things this way. Real language is already out there. It's just that SSS course developers are too lazy or stupid for corpus linguistics. Plus, it's cheaper for them to make stuff up than to look stuff up. Sadly, that cost-cutting, top-down design also means that…

Every SSS Treats Language Use Like Subroutines

Here is an artist's rendering of ideal Duolingo users.
An SSS doesn't even treat human behavior as dynamic. To an SSS developer, we're all bland creatures of habit. We all fly to some foreign country, book a hotel room there, eat in its restaurants, visit some tourist traps, and engage in vapid small talk. The creators of SSS weren't Parisian. With that level of condescension to our humanity, though, they might as well have been.

Again, it's totally ass-backwards. It doesn't even prioritize situations by urgency. Take this example:
  1. Emergencies: visiting hospitals, reporting crimes, warning bystanders, etc.
  2. Daily survival: getting food, finding places, arranging shelter, doing transactions, etc.
  3. Personal sanity: expressing feelings, socializing, describing personally relevant things, etc.
  4. Self-improvement: gaining knowledge, power, wealth, etc.
I just made that up, and it's better than 99% of situational organizations. To make it, all I had to do was ask, "What situations demand language use?" not,  "What situations are most common?" However, even then, it's inadequate. There's no individuality, spontaneity, or unexpected element. It's a pre-planned, phrasebook attempt to impart language. Machine translators, medical interpretation, and more killed off phrasebook demand. It should have killed off the SSS approach, as well.

This doesn't sound quite like Duolingo's syllabus.

"It's alive!"
That's because Duolingo's programmers changed the SSS protocol to make programming it easier. That is to say, they took what's broken, and then added their broken perspectives. Their biggest change? They cut out narratives and dialogues. You know, the part of the SSS model that provides meaningful context? Yeah, that's not easy to code. So, instead, they created "Frankenstein sentences" — syntactic skeletons with random lexical transplants. Unfortunately, they didn't implant a brain. The result is a bunch of weird sentences in Duolingo's modules.

Hell, their Chinese practice sentences don't even separate words correctly! Did I say a distant second? I take that back. Some people are linguists, some people are not linguists, and some people are not even linguists. You can guess where Duolingo's creators and designers fall.

Worse still is where they sit. They're too big for the necessary, radical changes. More than that, they seem unmotivated to rebuild their busted vessel.

That brings me to my biggest gripe with Duolingo:

It's Nothing But a Busted Language Quiz!

It's incorrect to call what Duolingo offers "lessons". Lessons imply teaching. Duolingo doesn't teach anything (at least, not for free). What it offers are scaffolded piles of quizzes. It doesn't monitor your accuracy as you work. It doesn't give you pointers as you stray off course. It doesn't review your submissions with any nuance. Duolingo is not a platform in which you are taught. It is a platform on which you are judged. To Duolingo, your work receives one of four judgments:
  1. It's right (i.e., what we expect).
  2. It's right, but not what we expected.
  3. It's wrong, but forgivable.
  4. It's wrong.
Since (II) and (IV) piss users off the most, I'll keep my gripes to them.

Let's start with (IV). Have a look at the image below. Imagine it's your quiz. How does it make you feel?
"Follow my way, and my way only!" -- Anonymous Prick
Wronged? I'm going to guess wronged. That's how it feels to be fully bilingual, and then to be told by Duolingo, "Well, our shitty program expected a different result, even though yours is correct, so fuck you! Minus one heart." It just gets worse from there.

Me: "What if a dozen native speakers take this 'lesson' and give you the same translation?"

Duolingo: "Fuck them! Minus one heart."

Me: "What if I correct your sentence, and then get support from natives on your own forum?"

Duolingo: "Fuck you! Minus one heart. I mean, unless we approve your correct translation days later. Then, sorry for the 'fuck you'. As a token of thanks, here's some spam."

Me: "Can you at least give us the L2 audio? That would be fair, given how often you're wrong when you say we're wrong."

Duolingo: "No! How else are we going to force learners and bilinguals to correct us?"

Now you know how the data for (II) is built. It should be the Duolingo motto: "Spammed if you do, slammed if you don't." If we generalize from this, it points out my outrage at (II). Duolingo basically un-shits itself on the backs of actual bilinguals. You see, when a new Duolingo beta release comes out, lots of bilinguals flock to it. As with all of their beta releases, it's godawful. We make good-faith efforts to fix their mistakes. They "approve" our translations, but they give us no credit. One by one, we lose patience with Duolingo. Most of us abandon it. Some of us tell everyone else to stay away from it. We roll our eyes when new learners tell us they're using it.

"But, it's free!" yell the ignorant masses. First of all, just because there's no up-front cost doesn't mean that it's free. These users don't reason through the opportunity costs. There are hundreds of things you could instead be doing. Dozens of those things are better for language acquisition than Duolingo. You can do better than to be judged by incompetent judges.

If it's so bad, then why do so many people use it?

I'm afraid that's a question for another post. One thing should be clear, though. I'm not disputing its popularity. I'm criticizing its misguided approach and design. I'm blasting against every computer scientist who thinks that he's a language guru eo ipso. Programming languages aren't natural. They're artificial. Experts in artificial languages shouldn't pretend to be SLA experts. That's what von Ahn and Hacker did. That's why their product is such a shit pile. The millions who take whiffs and bites of it do not change that.

Hey, Duolingo, iss meinen Arsch!
Or, maybe Duolingo is a language app for masochists. That would make more sense.


  1. As a quadrilingual person and also a upper secondary language teacher I couldn't agree more. There are good sides to duolingo, and those are that it's free and that it's a good introductory platform before you start with your 'real' language learning. Doing 4-6 weeks of duolingo before some real language course is beneficial.

    That's about it though. As you mentioned, the pedagogy it's based on is extremely dated. This is how language tuition looked in the 50s and the 60s. That's when mere repetition was seen as a virtue. And then more effective approaches came along. And during all this time, there's been non-stop research to find out what approach is the best. Research that form the basis for how languages are thought in schools (not all) nowadays. And this is great, language tuition is now more effective than ever before. For some reason, duolingo seems to ignore that this research or knowledge even exists.

    It's my firm belief that five minutes of active, contextualised learning equals an hour spent with duolingo.