From knowing
to speaking

01

Conversation is game day

Why practicing with a tutor isn’t as efficient as it seems

If you want to get good at something, the advice goes, just do it. So once they’ve covered the basics of grammar, many learners start practicing for real with a native speaker or an AI tutor. It’s a worthwhile experience. But it isn’t an efficient way to reach the next level, for three reasons.

1A tutor can’t see your learning map

By a learning map, we don’t mean vague labels like “beginner,” “intermediate,” and “advanced.” We mean concrete data: every word and grammar pattern you need to learn and how much each one matters; which of them you know and which you don’t; when you last saw each one, and which ones you’re about to forget. No tutor, human or AI, talks to you with that map spread out in front of them. Which words you run into is left to chance. The ones you already know come up again and again, and the ones you should be learning right now don’t show up when you need them.

2You can only say what you already have

When there’s nothing to draw on, learners either steer around the expression they don’t know, or translate from their own language in their heads and slot Japanese words into the result. Researchers have noticed since the 1970s that learners often avoid difficult structures altogether rather than get them wrong (Schachter, 1974). And sentences built by mental translation may look fine at first glance, but very often they’re just not what a Japanese speaker would say.

A tutor can correct you, of course. But with the pressure to keep a conversation going, correction works less well than you might think. In one study of real classrooms, the most common kind of correction, repeating the learner’s sentence back with the mistake fixed, was the least likely of all to get learners to correct themselves (Lyster & Ranta, 1997).

It works better the other way around. Hear correct sentences and say them back often enough first, and correct sentences are what come out when you need them. It’s how a baby picks up a first language: by echoing what Mom says.

3Game experience pays off once you have the fundamentals

What you get out of conversation depends on what you bring to it. Think of sports: a player without solid fundamentals can play game after game and still hit a ceiling. In psychologist Anders Ericsson’s research on musicians, what separated the levels of skill wasn’t the hours spent simply playing, but the hours spent alone in deliberate practice, aimed squarely at their weak points (Ericsson et al., 1993). If you have a goal, you need serious training, not just play.

Games are won by the training you do off the field.

02

What fluency is made of

Why you know the grammar and still get stuck

You know the grammar. You know a fair number of words. Yet when it’s time to speak, you freeze. It’s the most common complaint among learners who have finished the basics, and the answer lies in how native speakers actually talk.

1Native speakers don’t build sentences from scratch

The linguists Andrew Pawley and Frances Syder pointed out that grammar rules alone can’t explain how fluent native speakers are (Pawley & Syder, 1983). Native speakers don’t put each sentence together word by word, following the rules. They pull out whole chunks they’ve already heard and said countless times. That’s why they’re fast, and that’s why they sound natural.

For English speakers, this is a very practical problem. Learn 取る as “to take,” and when you want to say you took your medicine, out comes 薬を取る, which sounds as if you just picked up the bottle. A Japanese speaker says 薬を飲む: in Japanese, you drink your medicine. You won’t find 飲む next to “take” in any word list. It only lives in sentences.

“You shall know a word by the company it keeps.”

— J. R. Firth, linguist (Firth, 1957)

That’s why MomSpeaking always teaches words inside sentences. Learn from sentences, and you pick up the company each word keeps, its particles, its conjugations, its place in the word order, all at once.

2Native speakers don’t translate

What you learn from a grammar book is knowledge you can explain. What speaking takes is knowledge you can use without thinking. What turns one into the other isn’t understanding. It’s a great deal of practice. As practice builds up, the time, the mistakes, and the attention a task demands all shrink, and the knowledge gradually becomes automatic (DeKeyser, 2007).

The translation filter in your head fades the same way. At first, when you hear 飲む, the English word “drink” comes to mind, and only then the meaning. The more fluent you become, the more Japanese opens a direct line to meaning (Kroll & Stewart, 1994).

It’s multiplication, not addition. Know plenty of sentences but take forever to find them, and the conversation stalls. Be quick with nothing to say, and it runs thin. And waiting for either one to build up by chance in conversation is far too slow. Both need to be trained, on their own and intensively.

03

Shadowing, and its two weak spots

When a proven method is used by guesswork

Listen, recall the meaning, and say it back. Shadowing is a long-established way to learn a foreign language, and research backs it up. Repeating speech the moment you hear it improved learners’ ability to make out speech sounds accurately (Hamada, 2016), and researchers describe shadowing as a bridge between listening and speaking (Kadota, 2019). It starts from correct sentences and has you repeat them aloud, one sentence at a time, so it meets every condition above. But shadowing by guesswork loses much of its efficiency.

1The material doesn’t fit you

If you’ve ever tried shadowing a scene from an anime or a drama, you know how it goes. With too many words you don’t know, it turns into mimicking sounds without meaning. With sentences you already know, nothing new sticks. The best-known studies put numbers on it: to understand a text on your own, you need to know about 98% of its words (Hu & Nation, 2000), and to follow speech, about 95% (van Zeeland & Schmitt, 2013). Once more than one word in twenty is unfamiliar, in other words, comprehension starts to slip.

The linguist Stephen Krashen called the best input i+1: input just one step beyond your current level (i) (Krashen, 1985). The chances that whatever you’re shadowing happens to be i+1 for you, at the moment you study it, aren’t high.

2You miss the moment to review

Repeat something the same number of times, and spreading those repetitions over several days makes it last far longer than cramming them into one sitting. This is the spacing effect, one of the most consistently replicated findings in memory research since Hermann Ebbinghaus, famous for the forgetting curve (Ebbinghaus, 1885). An analysis of more than 300 experiments reached the same conclusion (Cepeda et al., 2006).

The problem is scale. Once there are thousands, then tens of thousands, of sentences to learn, keeping track of what to review and when is more than anyone can manage by hand.

Choosing what to study, and timing each review so it reaches long-term memory:
that’s too big a calculation to do by hand.
MomSpeaking does the math.

04

Shadowing, made entirely personal

How MomSpeaking picks a single sentence

What MomSpeaking does fits in one sentence: it picks the one Japanese sentence that’s most efficient for you, right now. To do that, its engine does four things.

Words

14,000+

Example sentences

140,000+

Grammar patterns

900+

1It maps out Japanese

We’ve sorted and ranked more than 14,000 words by JLPT level and by how often they’re actually used. A small number of high-frequency words make up most of what people say in everyday conversation (Zipf, 1949), so learning words in order of frequency gets you the most out of the same amount of time (Nation, 2013).

2It keeps a record of you

You study in sentences, but your record is kept word by word. After you shadow a sentence, you tell the app how it went, that signal is passed on to each word in it, and your study data builds up one word at a time: when you last studied it, how hard it is for you, how solid the memory is, and when it’s likely to fade. The state of each memory is estimated with the spaced repetition algorithm FSRS, which grew out of research that modeled how memory changes using more than 200 million real study records (Ye et al., 2022).

3It picks the best next sentence

When the learning engine chooses your next sentence, it looks over your learning map and picks a sentence that takes you just a little past where you are now.

In a sentence chosen this way, there’s usually just one unfamiliar word, and everything else becomes context that helps you understand it. It’s Krashen’s i+1, worked out fresh for every card from your own record.

4Reviews at just the right time

Words you’ve learned come back for review at the best possible moment: at short intervals at first, then days, weeks, and months apart as the memory firms up. Meeting a word again in a new context, at just the right moment, is what turns a word you know in your head into one you can use without thinking.

As you get better,
the learning engine gets stronger too.