Question 1

Why learn a language at all?

Accepted Answer

Because you almost certainly trust your own language more than it deserves. It feels like a window onto how things are, when it is a made object, assembled over thousands of years by people solving problems that may have nothing to do with your life. You cannot see that from the inside. A second language is the cheapest way outside there is: it shows you that the divisions you took for the shape of the world were one option among several, and after that you question your own reflexes more and other people's less.

Question 2

Does the language you speak really change how you think?

Accepted Answer

Yes, but less than the popular version claims, and it is worth being precise. What the evidence supports is that a language keeps making you notice certain things because you are forced to say them: grammatical gender, or which compass direction you are facing, or whether you saw something yourself or were told about it. Attention demanded every day leaves a mark. What the evidence does not support is the strong version, that your language sets a limit on what you are able to think. It does not. You can think anything in any language, you simply have to work harder for the things yours does not hand you on its own. Most of the famous stories, the hundred Inuit words for snow above all, are exaggerated and repeated by people who never checked the source.

Question 3

Which language should I learn?

Accepted Answer

The one you have a reason to hear. Not the most useful, not the easiest, the one where something is already waiting for you: a person, a place you keep going back to, a body of work you want without a translator standing between you and it. Motivation is the one ingredient nobody else can supply, and every method on earth lives or dies on it. If you honestly feel no pull in any direction, pick something close to what you already speak. The first success is worth more than the right choice.

Question 4

How should I use AI to learn a language?

Accepted Answer

As an infinitely patient explainer, not as a source of what is normal. It is very good at answering why is this wrong at two in the morning, at producing twenty variations of one pattern so you can drill it, at rewriting a text down to your level so you can actually read it, and at rehearsing a conversation you are genuinely about to have. Ask it to explain, to correct, and to drill. The one thing worth keeping for yourself is production. If it writes your sentences for you, you have outsourced precisely the effort that was doing the work.

Question 5

What is AI bad at, here?

Accepted Answer

Knowing what people actually say. A language model produces what is plausible, which is not the same thing as what is frequent or what is natural. It will hand you a grammatically perfect sentence that no native speaker has ever uttered, with complete confidence, and from the inside of a language you are learning you have no way of telling. It also flattens register: everything arrives at the same polite middle temperature, so you end up fluent in a version of the language nobody uses in a bar. This is exactly why the word lists here are counted from a corpus of real dialogue rather than generated. Use the model to explain, and use counted data to decide what is worth learning in the first place.

Question 6

Can AI replace a teacher?

Accepted Answer

For drilling, largely yes, and that is a genuine gain. Nobody in history has had unlimited patient practice before now. For the rest, no, and not for the reason people usually give. A teacher is not primarily a source of information, and that is precisely the part a model does replace. A teacher is someone who notices that you go quiet every single time you have to use the past tense, and asks you why. Most people who stall are not missing grammar. They are afraid of sounding stupid, and a system that praises every attempt cannot see that, let alone name it out loud.

Question 7

How many words do I actually need?

Accepted Answer

Fewer than you fear for a foothold, many more than you hope for comfort. The most common words of a language cover an absurd share of ordinary speech, somewhere between forty and sixty-five per cent, depending on the language. That sounds enormous and it is not comprehension: what it buys you is the ability to hear where one sentence ends and the next one starts, which is the thing that makes everything after it possible. Comfortable reading begins around ninety-eight per cent coverage, and that takes eight to nine thousand word families. Every language on this site has a page with its own curve, because the numbers differ a lot between them.

Question 8

How long does it take?

Accepted Answer

Longer than the apps imply and shorter than you fear, and honestly it depends on the distance between what you already speak and what you want. Something more useful than a number: two thirds of people quit within three months, and difficulty is almost never the reason. The reason is that progress in months two and three is real but invisible, and invisible progress feels like no progress. That is the actual argument for counting something, whether it is words known or hours listened. Not gamification. Evidence that the thing you cannot feel is happening anyway.

Question 9

Do I need to study grammar?

Accepted Answer

Some, later, and not the way you were taught it. Grammar is a description of patterns, and a description helps most once you have already bumped into the pattern a few times. Reading the rule first tends to produce people who can recite the conditional perfectly and freeze the moment they need it. Meet the thing, get confused by it, then read the explanation. At that point a single paragraph does more than a whole chapter would have done a month earlier.

Question 10

Is it too late if I am over forty?

Accepted Answer

For sounding native, probably, and there is no point pretending otherwise: the window for a fully native accent closes early. For everything else, no, and adults have advantages children do not. You can be told a rule and use it the same day. You already know how to study. You know what you want the language for, which children never do. What actually declines with age is not capacity, it is uninterrupted time and tolerance for feeling incompetent in front of other people.

Question 11

Do the big apps work?

Accepted Answer

They work for the thing they are built for, which is getting you to come back tomorrow. That is worth more than nothing: a habit you keep beats a method you abandon. Where they fail is the handover. They are decent up to roughly A2 and then leave you fluent inside the app and mute outside it, because tapping the right translation is not the same act as producing a sentence. Use one if it keeps you turning up, and start reading and speaking real material earlier than feels comfortable.

Question 12

Will I ever lose my accent?

Accepted Answer

Probably not entirely, and it matters considerably less than you have been led to believe. The thing worth working on is being understood, and that comes down to a small number of sounds and the rhythm, not to every phoneme in the inventory. Find the handful of sounds your own language does not contain, work on those, and let the rest go. An accent is evidence that you did something difficult.

Question 13

Why does all of this cost nothing?

Accepted Answer

Because the data underneath it is open and the work of arranging it is already done. The word lists come from an open corpus under a Creative Commons licence, I built the tools because I wanted to use them myself, and hosting a static site costs almost nothing. What does cost something is my time, so lessons are paid. Nothing here is a trial version of something else, nothing is behind an account, and your progress stays in your own browser.

Question 14

Where do the numbers come from?

Accepted Answer

The frequency lists are counted over the OpenSubtitles 2018 dialogue corpus, compiled by Hermit Dave and released under CC BY-SA 3.0. Subtitles lean towards conversation, which is the whole point: this is the language people speak rather than the language people publish. Speaker counts come from Ethnologue. Every figure on the site traces back to one of those, and where a well-known claim does not survive checking, it is marked as such instead of being quietly repeated.

Hi, I’m Vlada.

A language is an asset. A way of thinking you do not have yet.

Three things, in order of need

Things people actually ask

You do not need talent. You need a door and a reason.