Open the app

The science

A language app built out of what actually works

Taalhammer is built to get you speaking, and it does not look like this because we liked the layout. It is assembled out of the five findings that hold up in memory research and language acquisition, running together on your own sentences in any of 76 languages.

See plans

The claim

Every principle below is public knowledge. You can read Ebbinghaus yourself, and a dozen apps will tell you they use spaced repetition.

What almost nobody does is run them all together, on the same material, inside a day you would live anyway. Spaced repetition operating on your own full sentences, which are chunks, delivered as input you can just about follow, resurfaced interleaved rather than blocked — and, when there is no screen, in your ears. That combination is the product. The rest of this page is what each part does and why it earns its place.

The five findings

  1. 01

    Spaced repetition

    Forgetting is not a character flaw. It is a curve, and a curve can be scheduled. A review timed just before the memory fades buys far more than the same review done twice as often, which is why endings and agreement stop being a decision you make mid-sentence and start being reflex.

    The forgetting curveSpaced repetition in language learning

  2. 02

    Chunks, not words

    Fluent speakers do not assemble every utterance from parts. They reach for a ready-made sequence and bend it to fit. A word list gives you a glossary; a chunk gives you the behaviour of the language, which word wants which preposition and what order things go in. That is why a chunk comes back to you at conversation speed and a rule does not.

    Formulaic sequences: language consists of chunksInput, learning and acquisition

  3. 03

    Comprehensible input

    Acquisition happens on material one step past what you already understand. Too easy teaches nothing, too hard teaches nothing, and the band between them is narrower than it feels. This is the part that decides whether the other three have anything to work on.

    Krashen's comprehensible input

  4. 04

    Desirable difficulties

    Review that goes smoothly has taught you very little. The effort of pulling something back out of memory is what makes it hold, so the version that feels worse usually works better. Material arrives mixed rather than in neat blocks, so you have to work out which rule applies before you can apply it. We build the harder version on purpose.

    Desirable difficultiesRobert A. Bjork (1994)Interleaving: grammar, verbs and stories

  5. 05

    Learning while moving

    A review that needs only your ears fits the hours a screen cannot: the walk, the gym, the commute. And memory settles after the session, much of it while you sleep — so the hours around study are part of the work too. A method that fits the day you already live simply gets more of them.

    Memory consolidation and why we learn during sleepListening exercises that actually workThe hands-free listening mode

What happens when you run them together

Alone, each finding is a study. Together they are a loop, and the loop is what you actually use:

You write a sentence you genuinely need, or take one from a course. It is stored as a chunk, with audio and an automatic phonetic transcription, so nothing in the loop stalls waiting for you to look something up. It is scheduled against your own forgetting curve. It comes back interleaved with other material rather than in a block, at a level you can still follow, and it comes back slightly too soon to be comfortable.

No single step there is novel. Running all of them, on your own sentences, in one place, is.

And it adapts to you, not to an average

The schedule adapts to your recall: get something wrong and it returns sooner, get it right repeatedly and it backs off. The material adapts to what you put in, because the sentences can be yours. The difficulty adapts to what you already hold, so the same course does not move at the same speed for two different people.

You are not walking a fixed path that was laid out for somebody else and lightly personalised at the edges.

How the hands-free listening mode works And the technology underneath it

What we have not solved

We are researching whether a system can guarantee that each new sentence introduces exactly one unfamiliar element. It is a real question and it is not in the product.

Our own measurements put it well short of the target we set for it, and we would rather publish that than a claim you would catch us on. If it starts working, it ships, and we will show the numbers either way.

What you get

76 languages

Courses and your own material in any of them, with a real voice on every sentence and automatic phonetics.

Courses from A1 to C2

The Core collections run one continuous path from absolute beginner to fluent.

Your own sentences

Add the language you personally need and the same scheduling, audio and review machinery applies to it.

Learn and Listen

On the screen when you have one, hands-free in your ears when you do not.

This is for you if

  • You want to know why a method works before you will give it your evenings.
  • You have read about spaced repetition and are tired of apps that name-drop it and then show you a streak counter.
  • You have learned a language before and know the difference between recognising a word and being able to say it.
  • You would rather be told what is still unsolved than sold a guarantee.

What users say

Over 24,000 people forge a language with Taalhammer.

As a polyglot I tested all the other tools like Anki, Quizlet, and Memrise. But Taalhammer is in a league of its own. — Martin

Read the research, then go and test it on yourself. That is the only experiment that settles it.

See plans Open the app