The Hardest Languages to Learn, and Why AI Can't Learn Them for You
The hardest languages by study hours, next to Erin Meyer's culture scale: where meaning sits outside the words, and what a translation leaves you without.
Chief Language Pedagogy Geek @ Reelang

The hardest languages for an English speaker to learn are Arabic, Mandarin, Cantonese, Japanese and Korean. The US State Department, which trains American diplomats, gives each of them 88 weeks of full-time study, about 2,200 hours in class. Spanish and French take 30 weeks. Italian, Portuguese and Dutch take 24.
So why not skip the 2,200 hours and talk through an AI translator? Because a translator works from the words, and in many cultures a large part of the message is never put into words. You have to read the room, and the translator isn't in it.
DW News recently ran a report called "Does it still make sense to learn a new language in the age of AI?". In it the linguist Ema Ushioda, whose own background is Japanese, explains that Japanese speakers leave much of the important content unspoken, so a translation built from the words can miss it. That sent me back to a book I know well, Erin Meyer's The Culture Map, which maps the same idea across countries.
The hardest languages by study time
The State Department's language school, long known as the Foreign Service Institute, sorts languages into four groups by how long its students usually need to reach level 3 in speaking and listening on the US government's proficiency scale. That is professional working proficiency. A typical week is 23 hours in class and 17 hours of self-study.
| Group | Class time | Languages |
|---|---|---|
| I | 24 to 30 weeks (552 to 690 hours) | Spanish, French, Italian, Portuguese, Dutch, Swedish, Romanian, Danish, Norwegian |
| II | About 36 weeks (828 hours) | German, Haitian Creole, Indonesian, Malay, Swahili |
| III | About 44 weeks (1,012 hours) | Russian, Polish, Czech, Greek, Turkish, Hebrew, Hindi, Vietnamese, Thai, Finnish, Hungarian and about 30 more |
| IV | 88 weeks (2,200 hours) | Arabic, Mandarin, Cantonese, Japanese, Korean |
Source: US Department of State, Foreign Language Training.
Group IV takes more than three times as long as group I. The State Department calls these five "super-hard". What makes each one hard:
- Arabic. The script runs right to left and usually leaves out short vowels. The written standard differs from every spoken dialect, so a learner needs both. Our guide to Levantine Arabic covers one.
- Mandarin. Four tones change the meaning of a syllable, and reading takes thousands of characters.
- Cantonese. Six tones, the same character load, and a spoken language that differs from the written standard.
- Japanese. Three scripts used together, including 2,136 standard kanji, verbs at the end of the sentence, and separate verb forms for levels of politeness.
- Korean. The alphabet takes a few days. The grammar is far from English: verbs come last, and the ending changes with the listener's age and status.
The State Department describes group III as languages with "significant linguistic and/or cultural differences from English". Culture is already part of how it explains difficulty.
What study time leaves out
I taught English as a second language for a dozen years. My students included C-suite executives who dealt with counterparts from North America and Europe to India and China, where a misread email or a silence in a meeting had real consequences. I had every one of my students buy The Culture Map. Nothing I've read explains intercultural communication, and how languages differ from one another, as well as this book does. Every one of them loved it, and they told me it changed how they understood their own colleagues.
I also know both sides of her scale from home. My family comes from Russia, a relatively high-context culture, and I grew up in the United States, the lowest-context one she measures.
Meyer is a professor at INSEAD. Her first chapter, "Listening to the Air", ranks countries on one scale, which she calls Communicating. The idea comes from the anthropologist Edward Hall, and she explains it herself in a short video on low-context and high-context societies.
- In a low-context culture, a good communicator is precise and explicit. You say what you mean, and the listener takes it at face value.
- In a high-context culture, a good communicator is subtle and layered. Part of the message is spoken and part is read between the lines.
On her scale the United States is the lowest-context culture in the world and Japan is the highest. The chapter opens with Kenji Takaki, a Japanese HR executive, explaining Japan's word of the year, "KY". It is short for kuuki yomenai, someone who can't read the air. As he puts it: "In Japan if you can't read the air, you are not a good listener."
Her scale places national cultures, not languages. Language still shows in it. In the text she compares them directly: English is lower-context than French, Spanish, Italian and Portuguese, and those are lower-context than most Asian languages. Her figure below draws the scale with the countries grouped by language: an English-speaking cluster on the left, a Romance-language cluster in the middle, and an Asian cluster on the right.

Figure 1.2, "Communicating", from The Culture Map by Erin Meyer (PublicAffairs, 2014), chapter 1.
Inside each cluster, she writes, the countries with the most mixed populations and the shortest shared history sit lowest: the United States among English speakers, Brazil among the Romance languages, Singapore and India in Asia. Russia sits outside all three clusters, next to France.
Here is her scale beside two measures of study time, for languages you can learn on Reelang. The first column of hours is from our own study of learning time: hours of video at your level until your vocabulary matches a B2 learner's.
| Language | Our hours to a B2 vocabulary | State Department class hours | Where Meyer puts its country |
|---|---|---|---|
| Dutch | 180 | 552 | Netherlands: far low-context end |
| French | 225 | 690 | France: high-context side of the middle |
| Spanish | 250 | 690 | Spain, Mexico, Argentina: right of the middle |
| Italian | 280 | 552 | Italy: right of the middle |
| Portuguese | 290 | 552 | Brazil: right of the middle |
| German | 325 | 828 | Germany: far low-context end |
| Russian | 500 | 1,012 | Russia: high-context side of the middle, level with France |
| Arabic | not modeled | 2,200 | Saudi Arabia: high-context |
| Mandarin | not modeled | 2,200 | China: far high-context end |
| Korean | not modeled | 2,200 | Korea: far high-context end |
| Japanese | not modeled | 2,200 | Japan: the highest-context culture on the scale |
Meyer's positions are averages for a country's business culture. People inside one country differ, and she says so.
Two things stand out.
All five of the hardest languages are spoken in high-context countries. Saudi Arabia, China, Korea and Japan sit on the high-context side of her scale, and Cantonese is spoken in southern China and Hong Kong. These are two separate rankings placed side by side. Nothing here shows that one causes the other.
The two measures don't move together. Spanish is in the easiest group and sits well to the high-context side of the United States. German takes longer than Spanish and sits at the low-context end. Study time tells you how long the language takes. The context scale tells you how much its speakers leave unsaid.
One word, two meanings
Before culture comes into it, a language can ask more of the listener at the level of a single word. Meyer gives three examples in chapter 1:
- Japanese ashi means both "leg" and "foot".
- Hindi kal means both "tomorrow" and "yesterday".
- French ennuyé means "bored" or "bothered".
Her conclusion about the first two: when speaking Japanese or Hindi, "you really do have to 'read the air' to understand the message."
Every language has words like these, English included. What matters for a learner is that a translation has to pick one meaning, and the speaker didn't.
Russian works like her Japanese example. Рука is both "hand" and "arm", and нога is both "leg" and "foot". I also teach beginner Russian, one of my native languages, on my YouTube channel, ThatRussianWord, using comprehensible input: simple speech made understandable by what you can see. In a lesson on parts of the body, filmed at the gym, I say the same sentence twice, "У меня две руки и у меня две руки". Only what I'm pointing at tells you that the first means two arms and the second means two hands.
What people leave unsaid
High context is a different thing from a word with two meanings. It is a habit of speakers: how much they expect the listener to work out.
Meyer explains the French sous-entendu, saying something without saying it, and notes that Spanish has the same word, sobrentendido, and Portuguese has subentendido. Her example: a man tells his wife there are a lot of calories in the toffee ice cream she bought. He hasn't said she has gained weight. She hears it anyway.
Context matters even when nobody is deliberately leaving anything unsaid. In Mexican Spanish, ahorita, a form of ahora ("now"), can refer to the present, the near future, or even the recent past, depending on the sentence and situation. BBC News Mundo made a short video about it. The presenter says that words for time don't mean the same thing to everyone, and that many are understood "por contexto o hasta por cultura": by context, or even by culture. Then come four scenes to test yourself:
- Someone running late: "No me tardo, ¿eh? Ahorita llego a la junta."
- A waiter: "Ahorita le traigo su orden."
- A teenager: "Ay, sí, mamá, ahorita recojo mi cuarto."
- A friend: "Ahorita vi a tu ex."
The same word, four different situations: "I'll be there shortly," "I'll bring it in a moment," "I'll clean my room later" (perhaps much later), and "I just saw your ex." The verb tense offers clues, but tone and context tell you how much urgency the speaker really intends.
Easy to learn, and still not said out loud
Spanish is one of the fastest languages for an English speaker (our roadmap): 30 weeks at the State Department, and about 250 hours of video in our model before your vocabulary matches a B2 learner's. It is also the language of sobrentendido and ahorita. Knowing the words and the grammar gets you the sentence. It doesn't always get you the message.
"You" is the clearest case. Every Spanish course teaches tú for friends and usted for strangers and elders, and the verb forms take a week or two. Who gets which one depends on the country:
- Spain: tú with almost everyone your age, usted to mark distance or respect.
- Colombia: in Bogotá, usted is the default, used even between couples, friends and with children, alongside tú and sumercé. In Medellín and Cali you'll hear vos, tú and usted. On the Caribbean coast it is mostly tú.
- Costa Rica: usted is the everyday form, including inside the family.
- Argentina: vos replaces tú.
The linguist Cristal Yeseidy Cepeda Ruiz recorded how people in Bogotá switch between these forms with the same person, to soften an order, to show affection, or to match the other speaker. Her 2021 paper takes its title from one such sentence: "¿Eres feliz con su abuela?", with a tú verb and an usted possessive side by side. A 2023 comparison of Bogotá, Cali and Medellín finds all three cities moving toward the closer forms.
So the same usted that keeps a stranger at a distance in Madrid can be warm between a couple in Bogotá. A translation renders both as "you".
Russian has the same split, ты and вы, and I use the formal one with strangers and older people without thinking. The habit follows me into Spanish, where I err on the side of usted.
In this clip a Costa Rican speaker advises foreigners on which form to use there.
The hours buy you the forms. Which one to use, with whom, in which country, is a separate skill, and you get it by hearing people from that country talk to each other.
German is the reverse (our roadmap). It takes more hours than Spanish, but Germany and the Netherlands sit at the low-context end of the scale. A Dutch executive in one of Meyer's classes put it plainly: in the Netherlands, someone who doesn't say it straight isn't trusted.
Why not just use an AI translator?
Because of context. A translator gives you what was said. In a high-context culture the listener is expected to work out what was meant, from the situation, the relationship and the tone, and that job stays with you.
AI translation is good, and it keeps improving. It reads a menu, a sign or an email, and earbuds now translate speech as you listen. Current systems also use the surrounding conversation to fill in what a sentence leaves out. For a train ticket in Germany, translate.
In the DW report, Ema Ushioda gives the reason it isn't enough for Japanese (at 4:28): an AI bases its translation on the words, and "the important message might not be actually verbalized". She adds that machine translation works at a transactional level, getting information across to complete a task, and doesn't help you build friendships or feel you belong in a community.
Three things remain yours to do, however accurate the translation is.
- A correct translation can still miss the message. Meyer describes a guest in Iran who is offered food late at night and answers "No, thank you". A polite host asks twice more, because both know the first two refusals are manners. A translator gives you "No, thank you", correctly, and the guest goes to bed hungry.
- It happens inside one language too. A German manager in chapter 2 has a British boss who suggests he "think about" doing something differently. He thinks about it and decides not to. The boss meant: change it now. Both were speaking English, and there was nothing to translate.
- A translator only sees the text it is given. Japanese sentences often drop "I", "you" and "they", and English can't, so a translator has to choose. Researchers built a test set for this problem in 2020, and a context-aware system of the time chose wrongly. The more of the conversation a system sees, the better it does. It still doesn't see the room, the relationship or the history between the two people.
There is also the conversation itself. Speaking through a phone is slower than speaking, and people talk to the phone, not to you.
The answer to DW's question depends on the communication culture you're navigating and what you want from the language. A translation gives you the words. A working relationship, a friendship or a family in Japan, Korea, China, Russia, France or Mexico also needs what the words leave out, and you can only read that yourself.
So should you still learn a language in the age of AI? Probably, if you want to be an effective communicator and a good listener.
How to learn what isn't said
You pick up context by watching people use the language with each other: who pauses, who softens a refusal, what a raised eyebrow does to a sentence.
A French teacher says it herself in this clip: fluency includes the culture, the habits, the references, and "les petits sous-entendus" nobody explains to you.
Four habits that help:
- Watch people talking to each other, not only to the camera. Interviews, family vlogs and street conversations show how a "no" or a request is delivered.
- Watch the same short clip twice. The first time for the words, the second for the face and the voice. This is intensive listening.
- Ask two questions after each clip. What did the speaker say? And what did they mean, and what told me that?
- Stay near your level. You can't read between the lines while you're still decoding the lines. Our hours study shows how much video each level takes.
If you catch yourself translating every line as you listen, read how to stop translating in your head. For the wider method, see the best way to learn a language, according to research.
Reelang is built on short videos of native speakers, sorted by level, with a meaning under every word. Pick your language: Spanish, French, Russian, Japanese, Korean, Mandarin or Arabic.
Questions and answers
What is the hardest language to learn for English speakers?
By study time, Arabic, Mandarin, Cantonese, Japanese and Korean. The US State Department gives each 88 weeks of full-time study, about 2,200 class hours, against 24 to 30 weeks for Spanish, French or Italian.
Is Japanese hard to learn?
Yes. Japanese is in the State Department's hardest group at 2,200 class hours, with three scripts and 2,136 standard kanji. Erin Meyer's The Culture Map also ranks Japan as the highest-context culture in the world, so much of what people mean is left unsaid.
Is Russian hard to learn?
Russian is in the State Department's third group of four: about 44 weeks, or 1,012 class hours, roughly one and a half times Spanish. Russia also sits on the high-context side of Erin Meyer's scale, level with France.
Is Spanish easy to learn?
Spanish is in the fastest group for English speakers: 30 weeks, or 690 class hours, at the State Department. Spanish-speaking countries still sit on the high-context side of the United States, so tone and situation carry meaning that the words don't.
Can AI replace learning a language?
AI translation handles what is written or said, which is enough for signs, menus and simple requests. A correct translation can still miss what the speaker meant, and it can't build the relationship for you. In high-context cultures such as Japan, Korea, China and Russia, what is left unsaid is a large part of the message.
What is a high-context language?
High-context first describes a communication culture: people rely on shared knowledge, relationships, tone and what goes unsaid. Erin Meyer places Japan at the high-context end of her Communicating scale and the United States at the low-context end. She also uses the term for languages such as Japanese and Hindi, where many words take their meaning from the situation.
What is the difference between high-context and low-context cultures?
In a low-context culture, such as the United States, Germany or the Netherlands, good communication is explicit and taken at face value. In a high-context culture, such as Japan, Korea or China, messages are implied and read between the lines. The terms come from the anthropologist Edward Hall.
Begin met kijken
Kies een taal en start de Feed. Echte moedertaalspraak, ingedeeld op hoe moeilijk het te volgen is. Geen niveautest, geen aanmelding.
Kies een taalLees verder
Hoe lang duurt het om een taal te leren? Uren echte video per niveau, in 12 talen
We telden elk woord in 4.500 uur native video in 12 talen om in te schatten hoeveel uur input de woordenschat van elk ERK-niveau opbouwt.
De beste manier om een taal te leren, volgens onderzoek
Waar tientallen jaren taalonderzoek het over eens zijn: veel input die je begrijpt, eerst de meest voorkomende woorden, herhaling met tussenpozen en spreken met feedback.
Stoppen met vertalen in je hoofd: waarom begrijpen vóór spreken komt
Je stopt met vertalen in je hoofd door je brein zoveel begrijpelijke spraak te geven dat vertalen niet meer nodig is. Vertalen is een noodoplossing die je brein bouwt als het te weinig directe ervaring met een taal heeft. De oplossing is geen trucje en geen mentale discipline, maar volume: uren luisteren naar echte spraak die je grotendeels begrijpt, tot je brein een directe route van klank naar betekenis legt in plaats van een omweg via je moedertaal.