Listening in Chinese means catching which of four tones landed on a syllable that has already gone past. Listening in German means holding a clause in your head, unfinished, until the verb arrives at the end and tells you what happened in it. Both are called listening. Neither one is practice for the other.
That is why there are twenty four topics here and not five. Grammar, vocabulary, pronunciation, listening and speaking are real categories, and they are the wrong top level, because the name is shared and the problem under it is not.
Three findings that hold across all twenty four
- input has to be understood to count. Hours of audio you cannot follow build almost nothing; material where you get most of it and miss a little is where acquisition happens. This is true of Arabic and of Dutch alike, and it is the finding that decides how every branch below is written.
- each level costs roughly twice the last. The first thousand words appear constantly and the next thousand do not, so identical effort produces visibly less progress as you climb. The plateau every branch below names is arithmetic rather than a failure of character.
- 600 to 750 hours for a close language and 2200 for a distant one, to professional working proficiency for an English speaker. The factor of three is the single most useful number on this page, and it is a fact about distance rather than about difficulty.
The split is by language, because the five skills do not transfer
Almost everything published about language learning is filed the other way, by skill: one article on remembering vocabulary, one on improving listening, one on getting past the fear of speaking, each written to serve every language at once. Filed that way, an article can only say what is true of all of them, and what is true of all of them is a short list. That is why so much of it reads like advice you already had before you opened it.
Filed by language, the same five headings start saying something. Vocabulary in German is mostly a decoding problem, because a compound you have never seen is built out of parts you already have. Vocabulary in Arabic is a root problem: learn one pattern and a dozen words arrive with it. Vocabulary in Chinese is neither, because the unit carrying the meaning is the character, and there are a few thousand of those.
There is exactly one place where the skill split is the right one, and it has a topic of its own. Spacing, review, immersion and the business of finding somebody to practise with do not change when the language changes, so how to learn a language holds them once instead of twenty four times. Everything else lives inside the language it belongs to.
Of the eight pillars listed under Knowledge has a shape, this is the only one that divides by its object rather than by its craft, and the widest as a result.
Month two
The six weeks everybody passes through, and most people stop inside
The top is wide because month one was designed to be. The waist is narrow because nothing was designed for month two.
- Everyone who installs the appMonth one, where every session ends with something new
- Still opening it in week sixThe sentences they want are now longer than the exercises
- Looking for something else to doThe reward stopped arriving and nothing took its place
- Somebody to talk to who will not get boredThe thing that pays weekly instead of daily
- Material made for natives, with two subtitle tracksListening, which most people leave until last
- A review schedule you did not invent yourselfNow that there is enough in your head to schedule
- B1, and a language that answers backWhere the next stall is waiting, and it is a different animal
It does not say how many get through. Nobody counts the people who quietly stopped, because quietly stopping produces no data.
One language to B1 before a second one starts
The order that costs the most is not grammar before sounds or sounds before grammar. It is two languages started three weeks apart, on the belief that the second is nearly free because the first is going well. It is not free. Two languages interfere precisely where they resemble each other, which is why Spanish and Portuguese together is a worse plan than Spanish and Japanese together, and the people who try this are almost always trying the first pair.
The quieter wrong order is reading about method instead of doing any. There is a version of this reader who has compared six apps, read about spaced repetition, chosen a note system and learned forty words. Method is worth an hour in week one, and then worth nothing again until around month three, when there is finally enough in your head for a schedule to have something to schedule.
Inside a language the order is less contested than it looks. Sounds first, and this is the one thing worth being dogmatic about, because whatever your mouth does in the first month is what it will keep doing. A pronunciation habit costs far more to remove at month twelve than to install at month one. Arabic's throat consonants and Chinese's four tones are not advanced material. They are week one, in both.
Month two is where people quit, and it does not feel like quitting
The stall has a date on it. Month one is built for you: everything is new, every session ends with something you could not do when it started, and the free apps are tuned to exactly that feeling, because month one is where most of their users are. Month two takes it away. The sentences you now want to say are longer than the exercises. The words you meet are ones you half know rather than ones you do not know, and consolidating something you half know shows no result on any particular day. Six weeks of that is the shape of it.
Nobody decides to quit. They get busy, and the app moves down the home screen. The accurate description is that the reward stopped arriving and nothing took its place, so the fix is not discipline. It is putting something into that gap that pays weekly rather than daily: a person to talk to, or material made for natives instead of for learners. The stalls every learner hits covers the later ones, which are different animals and arrive around B1 and again at B2.
The signal
Fluency is measured in the overlap, and vocabulary drilling grows the wrong circle
Both circles are made of the same words. Only one of them has been practised in the direction a conversation asks for.
- What you understandGrows fastest, and is what a course tests
- What you can produceGrows slowest, and is what somebody waiting for you tests
The circles are drawn the same size, which is flattering. In most learners the understanding one is several times the other, and that is the shape people mistake for a confidence problem.
What to use in 2026
Anki is still the spaced repetition tool, which is a strange thing to write about software that looks the way it does. It replaced paper flashcards and has outlived a decade of startups built to replace it, because the scheduling is the product and the interface is not. Free on desktop and on Android; the iOS app is a paid one-off purchase, and that price is worth mentioning rather than skipping past. The mistake people make with it is filling it with words they have never met in a sentence. A card for a word you have not encountered is a card you will fail for a month.
Language Reactor, and the browser layers like it, is the second one, and it is worth more than any app. It puts two subtitle tracks on the same video, the target language and yours, and lets you stop on a line and take it apart. It replaced pausing, rewinding and a dictionary in your other hand. This is the technique that fixes listening, and listening is the skill most people leave until last and then find out was the whole thing.
The genuinely new tool, since 2023, is a model you can talk to. It replaced the scheduled tutor hour and the exchange partner who gets bored, and it is the specific answer to month two, because what is missing at month two is somebody at your level with unlimited patience. Every major assistant has a free tier that covers this much. The limits are worth saying plainly: it does not hear your pronunciation, because it is reading text, and it will accept a sentence no native speaker would produce, because accepting plausible text is what it does. Why that is true is enough of the mechanism to predict the behaviour, and getting a correction out of it instead of encouragement is getting a useful answer instead of a plausible one.
Duolingo is where most people reading this actually are, and it is fine. It replaced the phrasebook and the first chapter of a paper course, it is free with advertising, and it will carry you through month one better than a textbook would. What it does not do is month two. Its exercises stay short because short exercises are what keep a streak alive, and the sentence you want to say in week nine is longer than any of them.
Book to street
How far the language in the textbook is from the language in the room
Every course teaches a written standard. How far that standard sits from what a room actually says is a property of the language, not of the course.
Placed by how much of an ordinary conversation stays out of reach after a textbook course. A judgement, made here, not a measurement.
- SpanishWhat you read and what you hear sit close together
- GermanThe standard is genuinely spoken, and the dialects ignore it
- PersianThe same gap, smaller, and carried in the verbs
- ArabicA written standard nobody grew up speaking at home
It says nothing about difficulty, which is a separate axis. Arabic is far from its street form and Spanish is close to it, and neither fact tells you how long either one takes.
The signal is retrieval speed, not vocabulary size
Fluency is not an amount of words. It is how fast you can reach a smaller amount of them while somebody waits for you to finish the sentence. That distinction matters because the two things grow at different rates: nearly everyone can read and follow far past what they can produce, the gap widens rather than closes as the reading improves, and the standard misreading is that the gap is confidence. It is not confidence. It is that those words have only ever been retrieved in one direction.
The checkable version, at three months: say something out loud that you have not said before, about something that happened to you today, and watch whether an English sentence forms first and gets translated. If it does, you are translating, and more vocabulary will not stop you doing it. If it does not, even for one clumsy sentence, that is the thing that was not there at one month.
Twenty four topics, and where to go in
The world's second language is the widest of them, twelve clusters against five to seven everywhere else, because English is the only one here that also has to carry what each test scores for a reader whose deadline is a number on a certificate, the English that decides an interview, and teaching it to somebody else. Most people arriving at this subject arrive there.
The languages are not one topic with the name swapped out. Why the verb waits until the end is German's organising problem, and it makes German a listening difficulty long before it is a grammar difficulty. Chinese offers a straight trade: no conjugation, no plurals, no cases, and a writing system to learn instead. The gentlest big language to start is Spanish, and the reason is not that it is simple. It is that what you read and what you hear sit close together.
That last point separates these twenty four further than difficulty does. The distance from standard to street is widest in Arabic, where the written standard is a language nobody grew up speaking at home, so a year of it can leave you unable to follow a conversation in Cairo. Persian has a smaller version of the same gap, carried in the verbs. German has dialects that ignore the standard. Spanish barely has the problem.
For anyone here because of work, the neighbouring skill is being understood, in every form the job takes, and it is a separate skill: people with excellent English lose meetings every day.
What is left is which of the twenty four to open, and the useful form of that question is not which language is easiest. It is which one you will still be opening in week seven, because week seven is the only week that decides anything. The honest reasons people pick is where that gets argued out.