the build

Reproducing a country you're not in

AJATT's founding idea was "bring Japan to you." The internet made that cheap. Five layers, built in order. Each one costs a bit of setup once and pays exposure back forever.


layer 01 — set it once, forget it

Flip your environment

These are settings you change once and never think about again, and each one converts time you were already spending into input. Ellis's sampling argument is the reason: every flip makes the sample bigger without costing you a minute.

Phone and computer language

Switch the OS language on both. You already know where everything lives by muscle memory, so the first annoying week costs you almost nothing in function, and after that you get hundreds of incidental word-meetings a day from menus and dialogs you'd have skimmed in English anyway. Best exposure-per-effort ratio available anywhere.

Label the house

Sticky notes on objects, in characters plus pinyin. The classic beginner move, and it does work for a while, mostly because it forces the question "how do I say this?" fifty times a week. Retire it after a few months. Labels only teach nouns you already point at every day.

Narrate your own life

Inner monologue in the target language while making coffee, driving, in the shower. No equipment, no cost, works anywhere. When you hit a gap, and you'll hit one constantly, note the phrase and look it up later. This is Swain's noticing function running without a conversation partner.

Default audio

Make target-language audio the default for every slot that currently holds silence or English: driving, dishes, gym, walking the dog. Early on you won't understand most of it, and that's fine. You're training the ear to find word boundaries in the stream, which everything else depends on.

Search in the language

Google, YouTube, Wikipedia, recipe sites. Search in the target language for things you were going to look up anyway. Recipes, tech fixes, game guides and hobby content are ideal because you already know the domain, so context does the comprehension work Krashen needs.

Retrain your feeds

Make a separate YouTube or streaming profile used only for target-language content, and only interact with that content there. Within about a week the recommendation engine becomes a self-renewing comprehensible-input machine. For most people this is the highest-leverage item on the list, because it hijacks time you were already spending.


layer 02 — the main event

Choosing media that's actually comprehensible

The failure mode of home immersion isn't laziness. It's buying a native crime drama in week two, understanding twenty percent of it, and quietly deciding you're bad at languages. Nation's 98% rule from reading research transfers almost directly to listening: below roughly ninety percent comprehension you're not acquiring, you're enduring.

The subtitle ladder

RungSetupWhat it's forHow long to stay on it
0Native audio, English subsGetting used to the sound with zero comprehension demandWeeks zero to four, then get off. With English subs on, your eyes read English and your ears tune the audio out.
1Native audio, target-language subsBinding sound to written formThe workhorse. Most learners spend most of their first year here.
2Dual subsUnsticking yourself when rung 1 stallsSparingly. The English line competes for your attention and usually wins.
3No subsPure listening; the ear does all the workAdd gradually once rung 1 feels easy. Rewatching known content is the gentle way in.
4Re-watch, no subsKnown plot means near-total comprehensibilityAny time. A second viewing of a known episode beats a first viewing of a new one, per minute.

Picking the content


layer 03 — how to spend the hours

Active and passive immersion are different currencies

Active

Full attention on the content. Watching with subs, looking words up, mining sentences, reading with a dictionary open. This is where most acquisition happens, especially early.

high yield per minutecosts real attentionthe daily core

Passive

Audio in the background while your hands are busy: dishes, commute, gym, work. Comprehension is partial and intermittent, so the yield per minute is much lower, but the minutes are essentially free.

free minutesbuilds the earweak on its own

The ratio that works: thirty to sixty minutes of active immersion as the daily core, then as much passive as your life allows. And re-run yesterday's episode as today's background audio. Passive listening to content you've already watched actively is dramatically better than passive listening to new material, because you're consolidating known input instead of filtering noise.

What passive immersion is genuinely good for: prosody and rhythm, which is make-or-break for tonal languages; getting used to the speed and blur of real speech; and phonological segmentation, hearing where words begin and end, which adult beginners find nearly impossible at first. What it can't do: teach you vocabulary you've never met, or fix grammar you don't notice. It's a supplement with real value, not a substitute.


layer 04 — making it stick

Reading and sentence mining

Listening builds the ear and the gut feel. Reading builds the vocabulary and the precision, and Nation's numbers make it the highest-yield vocabulary source available, provided you read at the right level.

The reading ladder

  1. Graded readers. Books written inside controlled vocabulary bands, and the only way beginners get 98% coverage on real prose.
  2. Learner news and comics. Simplified current-affairs sites, webtoons, manga with reading aids.
  3. Web novels and fanfiction. Repetitive, plot-driven, endless. The mid-frequency vocabulary band lives here.
  4. Native material. News, essays, actual books. This is the destination, not a starting point.

Sentence mining

The retention engine from the AJATT and Refold world. While consuming media, pull out sentences containing one unknown item and put them in Anki with native audio.

  • Cards come from content you actually enjoyed, so context and motivation are built in
  • One unknown per card, never more
  • Audio on every card. You're training recognition, not translation
  • Ten to twenty new cards a day at most. Reviews always take priority

Nation's supporting techniques, all free: guess then confirm (guess a word from context, then check the dictionary; confirming a guess beats looking it up cold), narrow reading (one author or topic at a time, which halves the distinct new words), re-reading within a few weeks (free retrieval practice), and easy-reading sessions (about a third of your reading time on material with almost nothing unknown, purely for speed and fluency).


layer 05 — the part Canada forgot

Output: get humans involved

Swain's research and the Canadian data agree that comprehension without production plateaus. You need to be pushed to say things, and you need someone on the other end negotiating meaning with you. All of it now happens over a video call from your desk.

Paid tutors

italki and Preply. Community tutors run roughly ten to eighteen Australian dollars an hour for Mandarin, professional teachers more. Book two or three thirty-minute sessions a week and ask them to correct you. That correction is the pushed output and the negative evidence Long's research says drives form learning.

Language exchange

HelloTalk, Tandem, Discord servers. Free, and the reciprocity (you help them with English) is what makes the arrangement last. Less structured than a tutor, more social, and social is what keeps it going at month eight when the novelty has worn off.

AI voice practice

Voice-mode LLM conversation: infinite patience, zero social risk, available at eleven at night. It targets Krashen's affective filter directly, because no judgment means no anxiety means no blocked intake. Good for drilling the twenty conversation scripts you actually need.

Shadowing

Repeat audio a beat behind the speaker, matching rhythm, intonation and, for Mandarin, tones exactly. Builds the motor patterns for pronunciation without a partner, and forces the syntactic processing Swain says comprehension alone skips.

Journaling

Three to five sentences a day about what you did. Writing gives you time to run the Monitor properly, since Krashen's three conditions all hold when you write, so accuracy improves faster here than in speech and then transfers.

Recording yourself

Once a week, record two minutes of free speech and listen back. It's unpleasant and it's the fastest way to hear your own gaps. You'll catch fossilised errors instantly that a tutor has been correcting for months.

!

Timing. Don't force output on day one. Krashen's silent period is real, and speaking before you can hear the difference between tones just practises errors. Rough guide: two to three months of mostly input to build the ear, then start speaking and never stop. For tonal languages, listen longer than feels necessary before you open your mouth.


calibration

Adjusting the build as you move

StageFocusMediaMining and SRSOutput
Absolute beginner
weeks 0–8
Sound system first: tones and pinyin, or the alphabet. Then the highest-frequency few hundred words.Pronunciation videos, beginner comprehensible-input channels, kids' songs. Passive audio from day one.A beginner Anki deck plus pronunciation pairs.None. Quiet shadowing only.
Beginner
months 2–6
Listening stamina and the core thousand words.Learner content, graded readers, easy kids' animation. Rungs zero and one.Mine ten to fifteen sentences a day from what you watch.Self-talk, journaling, first tutor session around month four.
Intermediate
months 6–18
The long slog: one to five thousand words, native-speed listening. Where most people stall.Native content you enjoy, re-watches of known shows, web novels, podcasts.Mining continues; the review queue becomes the main workload.Tutor two or three times a week. This stage needs it most.
Advanced
year 2+
Depth, register, domain vocabulary, accent.News, books, film, technical material in your field, unscripted native media.Mining tapers. Most new words now come from volume alone.Regular conversation and writing, no scaffolding.