Langdogu

This site in

Get Lingua Layer

A map of what you know

Langdogu is built around a personal ontology: a map of what you know, what you don’t, and how words connect. A word list can only say that you saw a word. The map is designed to know which sense you know, the sentence you met it in, what it shares with your other languages, and when you are about to forget it.

Being rebuiltThe ontology core is being rebuilt. Every diagram on this page is a sample that shows the design.

Concepts sit above languages

Each word hangs from a concept, not from its spelling. Read the map like a transit map: every concept is a line, and the words for it in each language are its stations.

Choose a word
Being rebuiltSample

  1. Conceptletter

    • Japanese手紙tegamiSame characters as 手纸 on the toilet paper line.
    • Korean편지pyeonji
    • Chinese信xìn

  2. Concepttoilet paper

    • Japaneseトイレットペーパーtoiretto pēpā
    • Korean화장지hwajangji
    • Chinese手纸shǒuzhǐSame characters as 手紙 on the letter line.

手紙 is a letter in Japanese. 手纸 is toilet paper in Chinese, where 纸 is the simplified form of 紙. Same characters, two lines.

Sample map, written for this page. Readings in Hepburn, Revised Romanization and pinyin.

発車 in Japanese, 발차 in Korean and 发车 in Chinese sit on one line, departure. They share their characters, in Japanese, traditional and simplified forms, and readings you can hear in all three, so a word you learn once has somewhere to land in every language.

The same characters can also mean something else. 手紙 is a letter in Japanese; 手纸 in Chinese is toilet paper. The map keeps them on two different lines, joined only by a dim dotted link that says “same characters”, so knowing one never marks the other as known.

A finish line, not an endless feed

Every step starts from the smallest unit whose pronunciation is certain and ends at a finish line you can see: a complete sentence. The sentence you finish becomes the material for the next step.

On its own, 発 can be read more than one way. Inside 発車 it can only be read one way, so the word is where this line starts. Each station adds a little, and the line ends at a sentence you can read whole.

Levels stack the same way: word, phrase, sentence, conversation. The finish line of one level is the start of the next, so progress has a shape, and an end you can see from the start.

Stations on the lineBeing rebuiltSample

Station 1 of 4Start: the smallest unit you can pronounce for certain

発車

はっしゃ

departure (of a train)

New at this station

Sample line, written for this page.

Same word, a sense you don’t have yet

Knowing is tracked per sense, inside sentences. You can know 読む in one sentence and not in the next.

You marked 本を読む, to read a book, as known. Then a manga balloon says 空気を読む: to read the room, literally “to read the air”. It is the same 読む in a sense you don’t have yet, inside an expression whose meaning is not the sum of its words.

Langdogu proposes it as a fixed expression, one unit with its own review, and you confirm. In the first app, “known” is what you mark yourself; judging it from your reading, sentence by sentence, comes later.

Sense of 読む

Being rebuiltSample

  • An e-book

    毎晩、寝る前に本を読む。

    Every night I read a book before bed.

    本を読む
    hon o yomu, to read a book

    Known: you marked it

  • Manga balloon

    ちょっとは空気を読んでよ。

    Read the room a little!

    空気を読む
    kūki o yomu, to read the room
    to read the room
    Literally: to read the air

    A sense you don’t have yet

Proposed: keep 空気を読む as one fixed expression

Nothing changes until you confirm.

Sample sentences, written for this page.

Every word remembers where you met it

The page, the manga panel, the line. Review brings back the moment, not a card with only a definition on it.

Each sighting is kept with its sentence and its source. A sighting in another language lands on the same concept line, so 发车 on a Chinese bus timetable is the same idea you saved from a Japanese rail notice.

Being rebuiltSample

  1. Day 0Saved with its sentence

    Rail service notice, read in Firefox

    強風の影響で、発車が10分ほど遅れています。

    Departures are running about 10 minutes late because of strong winds.

  2. Day 12

    Manga balloon

    発車まであと5分！走れ！

    Five minutes to departure! Run!

  3. Day 30

    Bus timetable, in Chinese

    末班车23:00发车。

    The last bus leaves at 23:00.

Sample sightings of one concept, departure, over 30 days.

Review comes back just before you would forget

Every word and every fixed expression has its own spaced repetition (FSRS) state, scheduled from your own answers.

FSRS estimates how likely you are to recall an item and brings it back when that chance is about to fall below 90%. Each review that goes well pushes the next one further out.

Your history is appended, never overwritten. Sightings and reviews only ever add a line, so how you came to know a word stays readable years later, and you can still delete any item, its source or your whole account.

Chance of recalling 発車

Being rebuiltSample

Day 0Days since you saved itDay 120

  • Today
  • A review
  • If today’s review is skipped

Sample chance of recalling 発車 over 120 days. Each review, on day 2 (Good), day 8 (Hard) and day 17 (Good), came when recall reached 90%, and each gap was longer than the one before. Today is day 41, and a review is due.

Answer today’s reviewReview due today, day 41

HistoryLines are only ever added.

  1. Day 0Saved with its sentence from a rail service notice
  2. Day 2Reviewed: Good
  3. Day 8Reviewed: Hard
  4. Day 12Met again in a manga balloon
  5. Day 17Reviewed: Good
  6. Day 30Same concept met in Chinese, as 发车

Sample. The curve uses the FSRS forgetting curve; the memory strength after each answer is illustrative.
Planned

The dictionary is shared. What you know is yours.

Concepts, senses and fixed expressions are common ground. What you saved, what you know and how you reviewed it stay in your own record and never write back into the shared dictionary.

PlannedYour data is yours: a full export in open formats at any time, and cancelling a paid plan never blocks reading, reviewing or exporting what you have.

Shared dictionary

  • Concepts
  • Senses
  • Fixed expressions

Your record points at the dictionary

Nothing in your record writes back

Your record

  • What you saved
  • What you know
  • Where you met it
  • How you reviewed it

Each related language should cost less

What you know in one language should give you a head start in the next, through shared meanings, roots and characters, and show you exactly where two languages differ.

French, Spanish, Italian and Portuguese share most of their middle grammar layer, and the same characters connect Japanese, Korean and Chinese words. Langdogu will measure that head start rather than assume it. Sharing rules between related languages existed in the earlier prototype and is being rebuilt.

Show
Being rebuiltSample

  • FrançaisFrench

    Jelisunlivre.

  • EspañolSpanish

    no pronounLeounlibro.

  • ItalianoItalian

    no pronounLeggounlibro.

  • PortuguêsPortuguese

    no pronounLeioumlivro.

I’m reading a book.

The verb and the noun come from the same roots in all four, and the words come in the same order.

One word, the same characters, three readings

  • 학생haksaeng, Korean
  • 学生gakusei, Japanese
  • 学生xuésheng, Chinese
  • student
Sample sentences, written for this page.

Reading at the edge of what you know

Two pieces of the earlier prototype are being rebuilt on top of the map.

Known-word highlightingBeing rebuilt

Sentence by sentence, an AI model would judge which words you already know, so the ones you don’t stand out, and level filters such as JLPT would set the edge. JLPT has published no official vocabulary list since 2010, so the filters would use clearly labelled third-party lists. A list-based version was prototyped earlier as a Firefox extension.

Treat as known: words up to
Being rebuiltSample

大雨の影響で、電車に遅れが出ています。現在、一部の区間で運転を見合わせています。

40% of this notice is words you know (4 of 10)

Sample notice, written for this page. Levels are illustrative.

Model routingBeing rebuilt

Light models such as Claude Haiku would handle preprocessing and most sentences; idioms and ambiguous passages would go to stronger models such as Claude Opus. The earlier prototype routed by cost and had a stronger model spot-check a sample of a cheaper model’s output; escalating by difficulty is new.

Being rebuiltSample

Light models, such as Claude Haiku

Preprocessing and most sentences

  • まもなく、3番線から電車が発車します。A plain sentence
  • 末班车23:00发车。A plain sentence

Stronger models, such as Claude Opus

Idioms and ambiguous passages

  • ちょっとは空気を読んでよ。An idiom: read the room
  • 結構です。Ambiguous: “that’s fine” or “no, thank you”
Sample sentences, written for this page.

11 languages to start. Every language is the goal.

The concept layer was designed for eleven languages in six writing systems, written left to right and right to left. Eleven is where the map starts, not where it ends.

Today, Lingua Layer reads Chinese and Japanese into Korean. The app’s first audience is Korean speakers reading Japanese on the web; Simplified Chinese and English are also in scope.

  • 한국어읽다Hangul
  • EnglishreadLatin
  • 中文读Chinese characters
  • 日本語読むKana and kanji
  • FrançaislireLatin
  • EspañolleerLatin
  • ItalianoleggereLatin
  • PortuguêslerLatin
  • DeutschlesenLatin
  • РусскийчитатьCyrillic
  • العربيةقرأArabic, right to left
  • Every languageThe goal

Where this comes from

The map was designed and partly prototyped before Langdogu had a name. That prototype is retired; the design is being rebuilt.

  1. Early 2026

    Past prototype

    An earlier ontology prototype was built around an 11-language graph. It is now retired.

  2. Mid-2026

    Being rebuilt

    The method was redesigned: a history that is only appended to, and languages as data.

  3. August 2026

    Past prototype

    A list-based known-word highlighter was prototyped as a Firefox extension.

  4. October 2026

    Being rebuilt

    Langdogu app development started, and the ontology core is being rebuilt for it.

One record, many doors

The map is the asset. The extension, the app, the AI assistant you already talk to and your own server are doors into one record.

  • Live

    Lingua Layer

    Where you meet a word on the web. Version 0.2.1 translates Chinese and Japanese into Korean in Firefox. It does not keep vocabulary or learning history yet.

    About Lingua Layer
  • In development

    The Langdogu app

    Keeps the record and the review: save an expression with its sentence, review it with FSRS, and see the expressions you know told apart on the next read.

    About the app
  • In development

    AI assistants, through MCP

    Claude, ChatGPT and other MCP-compatible assistants could open your vocabulary and talk with you about what you learned today, only after you add Langdogu and approve it.

    Open by design
  • In development

    Your own server

    The same learning features, API and MCP server on your own PC, NAS or server. Free forever; you bring your own AI key and pay the provider directly.

    Open by design

Everything you learn should have somewhere to stay.

The status of every piece

Questions or ideas about the map: [email protected]