For my recent trip to Norway I thought I'd prepare by taking some language lessons. As it was already my third journey to the country it seemed appropriate to make a bit of an effort this time to not just be another bumbling tourist idiot. For lack of a proper language class anywhere near me, I decided to give Duolingo a try. Friends had been recommending the platform and I was curious about the experience of learning a language without human interaction.

I signed up in January 2026. Using mostly the web version and occasionally the Android app, I finished the Norwegian course a bit more than 200 days later. Surprisingly, they never charged me a dime for it nor did it ever display ads; I suspect a glitch in the subscription. The course has plenty of content compared to other Duolingo offerings: split in seven sections of uneven length, it appears much larger than for example the Dutch class.

Signing up I was put off by the lack of information regarding the depth of those courses. Duolingo assesses proficiency in "levels" which is kind of an arbitrary measure and mainly tracks course completion, not competence. E. g. finishing the Norwegian class one advances to "level" 115 in the language. There is however no clear mapping between Duolingo levels and standards like the European Reference Framework, so what that number 115 is supposed to mean can only be guessed.

Duolingo didactics is quite different compared to traditional in-person language classes, and the result feels different as well: while I did start my vacation with decent passive language comprehension, I almost completely lacked routine in responding in the language. Not that I couldn't. For most standard situations my proficiency was more than sufficient, I just didn't intuitively use Norwegian when addressed in it. What could this be attributed to? Most likely the culprit is that Duolingo attempts to teach through translation. No matter how far one progresses, the class remains bilingual throughout: Most exercises consist of translating from and to English, and vocabulary is trained using English as the reference. Contrast this with an in-person class which after a few introductory lessons is usually taught exclusively in the target language. No English middleman. This way, if you're given a task or asked a question, you learn to respond directly without the roundtrip through another language. Duolingo however always requires context switching back and forth with English which may explain the lack of intuive responses even after many hours of practice.

It's not a complete waste of time though. Below are some observations about Duolingo that may help in the decision whether it might be worth trying out.

The Good

Duolingo is amazing at two things in particular:

1. Rote learning through repetition. It will generate exercises on the fly, sprinkle in new words with recently learned ones and make you actively use them in several exercises. Plus they do seem to have done their research as the average success rate appears to be fine tuned to 94 % which agrees with studies on the optimal error rate for learning. In combination this is incredibly effective at making the human brain retain vocabulary.
2. Individual pace. While the mobile app nags you to log on and interact with it daily, it's still up to the user to decide about frequency and schedule of lessons. How much progress you make in a day is entirely up to you. This is unbeatable by a language class, even an expensive individual one.

These two aspects alone make the platform worthwhile. The flexibility is incredibly motivating as it allows taking lessons during downtime e. g. when traveling or waiting in line.

Further highlights are speaking exercises although for some inexplicable reason they're only available in the mobile app, and the implementation is rather janky. The lessons can also be quite entertaining and witty; often they feature quotations from pop culture and colloquialisms ("It ain't over till the fat lady sings").

Also the mere availability of classes for "small" languages with just a few million speakers is incredible. The Norwegian course by Duolingo seems to reach up to about advanced intermediate level. Contrast that with the competing commercial online offerings (let alone on-site classes) which usually only cover up to level A2. The situation varies depending on the language however; for example the Dutch and Swahili classes feature fewer sections than Norwegian.

The Bad

Progress

Duolingo progress is hard to assess as it lacks classification according to standards like the ERF. Instead they have a few metrics of their own which are BS for the most part. What does it mean to achieve "level 115"? Or 140000 experience points? Without tying these stats to a commensurable standard they remain utterly useless.

English Creep

As already touched on above, Duolingo classes are always bilingual between the target language and a teaching language, usually English. This by far the most egregious downside compared to regular classes and probably acccounts for the lack of fluency even after many hours spent using a language.

Lack of Personalization

While Duolingo's strength is in concerted repetition, it doesn't appear to leverage personal statistics to focus on the user's weaknesses. There's "personalized exercises" alright that challenge the user on the content of earlier lessons. There's no way however these are based on past mistakes. It seems more like an arbitrary subset of phrases that appeared at one time which get repeated over and over again, even after the user gave the correct solution dozens of times. While most of the content never appears again after it was used in an exercise. For e.g. in the Norwegian course I must have gotten an exercise about someone's boyfriend's gray hooded sweaters about once per day. Despite getting that one right for dozens of times it still kept reappearing which I take as conclusive evidence that Duolingo's claims of using statistics to create individual lessons is complete BS. If there's any application you'd expect machine learning to help with in language learning, it's figuring out the student's weak spots and drilling down on them. Duolingo just does not do anything of the kind.

Likewise Duolingo doesn't add variety to practice material either. A sentence will always remain in the exact form it first appears in in an exercise. Considering how often some of the sentences are repeated, this results in memorization of exact phrases instead of the language patterns behind them. It would be another great use case for machine learning to prevent the user from learning a set of exercise-response pairs instead of the rules of a language.

Another missed opportunity at personalization is the hard limit to vocabulary. It’d be great if one could extend it e. g. with words picked up while reading a book or a news article. That would allow leveraging Duolingo’s main strength, memorization, even beyond the official classes. There is no way to enter additional words or phrases however, so the platform has a definitive end to its usefulnes. Nor is there a way of dumping the vocabular one has learned so far for use with other tools.

Pronounciation Lessons

While the idea of speaking lessons is great, the actual implementation in Duolingo leaves a lot to be desired.

First off, why only offer this feature in the mobile app version? For the most part I've been using the website, reaching to the Android version mainly in constrained settings like on trains where speaking isn't an option. Thus I must have skipped about 95 % of the available speaking exercises. What's more, speaking isn't generally available on the mobile app either. At least on Android requires enabling a some creepy Google service and infuriatingly won't even give a useful error message when that isn't running.

The lessons themselves follow Duolingo's approach of not explaining anything and refusing assistance in improving oneself. There's no feedback on what your mistakes are or how to correct them. You either get the pronounciation "right" (whatever the platform accepts) or you fail. For lack of feedback you might get the idea of trying out different pronounciations to get a feeling for what's within the rules and what isn't - too bad Duolingo doesn't allow repeating completed lessons. You either get it right in three attempts or you won't get another chance.

And the worst of all, there's a few lessons that are impossible to get right. The audio input just fails to be accepted by the platform even for simple phrases you couldn't possibly get wrong. It'd be an interesting experiment to pit Duolingo's speaker voices against the voice recognition model. Given the numerous occasions when their speakers pronunce words so obviously wrong that a complete beginner in the language will pick it up, I wouldn't be entirely surprised if the speakers fail to get a perfect score. (The voices tend to drop to English pronounciation for words that might have been missing in their training set, e.g. "tampongene" ("the tampons") gets rendered as "tampon-gene" with gene pronounced like the English word.)

As they're implemented now, speaking lessons make almost no difference to learning progress and might as well not exist at all.

Miscellaneous Bad Stuff

• The tiresome insistence on word order exercises. Granted, that may be an issue for native English speakers but that doesn't apply to a significant share of learners for most courses. In particular, coming from another Germanic language you shouldn't require that much emphasis on word order when learning Norwegian. It's also something a machine learning model would pick up on: those learners that do have trouble getting the word order right could expect getting relevant exercises more frequently.

• Listening exercises are presented by only two speaker voices, a female one and a male one. Given their crystal clear pronounciation and the artificial setting free of background noise, listening exercises cap out on difficulty (and thus effectiveness) early into the course, leaving the user wholly underprepared for real-life situations.

• New words are introduced without the grammatical gender which is borderline useless. You end up looking up most nouns on bokmålsordboka anyways.

• Speaking of vocabulary, the list of "learned words" just sucks. It presents a long list of thousands of items without anything that might help finding an entry: no search, no alphabetic sections. The only options are alphabetic sorting and progressively loading more items. Good luck finding some word starting with the letter "v"! How this "feature" made it through QA is puzzling. No wonder it doesn't even show up for some courses (Swahili for e.g. doesn't have it).

• Omission of much of the standard repertoire of language classes: There's no discussion of the country or region a language is spoken in; no excursions into etymology, the origin of loanwords or related languages; no info about dialects; no discussion of literature and the monuments of a language, etc. As a result it feels like being presented the language in a vacuum.

• Refusing to analyze words into stem and endings. I.e. the plural and definitive forms of a word, no matter if the declension is regular or not, each get treated as separate words. It's almost infantilizing how Duolingo hand-waves away morphology. I wonder how that works for languages that have a case system. Is each Polish noun treated as fourteen different "words"?

• The inconsistent use of contractions. Duolingo is rather anal about not accepting "we have" for "we've" and vice versa etc. The former is easier to type at least on my keyboard as it doesn't require dropping to the third layer. Which of the two options an exercise uses seems to be random and there's no specific didactic goal underlying the use either. Just allow both variants everywhere FFS.

• Synonyms are handled with extreme inconsistency as well. "Synes", "tror", "tenker" may all mean "think" but it's down to dice roll which one a given translation exercise accepts. If you're gonna reject one of them, then at least give a rationale for it. Any human language teacher would.

The Ugly

These were the merely bad issues. Now for the design flaws of Duolingo that are unforgivable and hard to reconcile with the purported objective of being a platform for learning languages:

There's no going back on exercises after completing them. That's right: Duolingo prevents you from going back and forth between units. Once done an exercise becomes inaccessible and there is no way for the user to revisit it. Considering how skipping between pages is a trivial feature of learning with a book, it's ridiculous for a language course not to allow this.

The mobile app requiring the user to opt into intrusive Google services globally (it cannot be done selectively for just Duolingo) is unacceptable.

There is no clear description of the course content. Units are structured into sections but there's no reference as to what level of proficiency a section targets.

In Duolingo, grammar does not exist. It's not just that the lessons avoid terminology, they also seem to use an overly simplistic model of morphology where different forms of a noun or verb get treated as separate words. Which ends up needlessly blowing up unit length as to each word the plural form is introduced separately instead of just pointing to the class it belongs to. This approach is tedious and error prone compared to learning a few rules and applying them to the vocabulary, and as a user one has to consult a grammar book or website anyways to complement the amorphous presentation in Duolingo.

Time based exercises are messed up and smell of engagement metrics. There's no way that gamified crap actually contributes to language acquisition. Thankfully they can be ignored altogether. Likewise for "experience points". While presented as though they were some kind of metric, XP turns out another manipulation device aimed to increase engagement. It's telling that the platform gives out about tree times as many "experience points" on average for the same exercise when using the Android app than on the website - a classic dark pattern rewarding the behavior that's most profitable to the vendor. I wouldn't be surprised if the reward was even more on IOS considering the higher ROI for users locked into the Apple ecosystem.

All content is chopped into short fragments. Some sentences appear like they've been taken from longer dialogs and prose, but they're never presented in an order that would make sense. This is in stark contrast to regular courses where assignments often include reading and summarizing texts that span multiple paragraphs. It's quite obvious that this format gets in the way of didactics: Exercises often insist on a single meaning of a word which would be evident from the preceding context. Duolingo omits any context and papers over the issue by forcing the user to select a translation from a short list of choices, refusing to acknowledge the ambiguity.

There's no writing exercises beyond one to one translation, no essay assignments, no creative writing even on the higher "levels". It's unclear how one is supposed to acquire a profound understanding of a language without experimenting with tone and different registers. Duolingo just skips over anything that goes beyond rote memorization. This may be the most glaring flaw in Duolingo's approach, and possibly the hardest to fix.

Also I understand the platform has ads. I can't confirm that as it never showed me any. Though it's clear the moment it starts doing that, I will delete the account.

Neutral

There's a number of minor issues and annoyances that can be opted out of or ignored but which should be mentioned here:

I found the gamification approach useless at best: the endless quests, leagues, gems, treasure chests appear shoehorned in and in no way tied to language instruction. The corsett would fit a gamified accounting app just as well.

I'd count the animations as part of gamification too. They're far worse in the Android app and cannot be disabled despite a switch in the settings. The website is mostly free of animated content though and what's there can be largely ignored without missing out on content. Not disruptive per se, merely distracting.

In some ways the course content hasn't kept up with reality: in the case of Norwegian it teaches outdated administrative divisions. For Duolingo, Norway is still comprised of 11 fylker in 2026 which has only been the case from 2020 to 2024. This is a minor issue that language books exhibit as well, but books at least state their "knowledge cutoff" on the title page. Considering this is easy to spot by running the assignments through an up to date LLM, it should still be treated as a QA failure.

Finally for a platform that claims its approach to language learning is based on science and statistics, one aspect that Duolingo thoroughly disappoints in is stats. No insights into one's own error rate, no most frequent mistakes, no global percentile of proficiency or vocabulary acquisition rate, no difficult to learn words and phrases - all the things you might intuitively mine Duolingo's gigantic database of user behavior for just aren't there. Instead the few numbers that do appear in the UI are fictional quantities: total "experience" points, points per week, as well as some awards that get bumped everytime some counter increases.

What gives?

To capture the essence of Duolingo: The areas where it has the most potential (statistics) it leaves the most to be desired. The hard challenges it pretends don't exist (writing). As soon as there's ambiguity (synonyms), its limitations become manifest.

What remains is an effective tool for rote learning in an annoying and unflexible wrapping. But that effectiveness ends where the official classes are exhausted. If Duolingo allowed entering new words that would turn it into a decent replacement for flashcards, but alas there’s no way for users to extend it in any way.

tl;dr: Duolingo is ok to use as one tool in your repertoir for learning a language, as long as it's free and without ads.