Every language,
written in Hangul.

The Stroke-Adding Rule

Letters shaped like the sounds they make

Hangul's five basic consonants are literal sketches of your mouth and throat mid-sound. Add one stroke for a stronger sound — that's the whole rulebook. Nothing to memorize, everything to derive.

Heaven, Earth, Human

Three strokes, infinite vowels

Every vowel starts from just three shapes: the roundness of the sky, the flatness of the earth, a person standing tall. Combine them and you get every vowel below.

Sejong's Full Design

Modern Korean uses 24 letters. Sejong designed 28 — and more.

The 1446 Hunminjeongeum Haerye lists 28 letters: 17 consonants and 11 vowels. Four fell out of everyday Korean — three consonants (ㅿ ㆆ ㆁ) and one vowel (ㆍ). Korean didn't need them, so they faded. But the system was never limited to Korean: it also defines ways to combine letters into sounds no single letter covers. This section shows that toolbox, and how Hangulpia proposes to use it for Japanese, Mandarin and English.

雖風聲鶴唳 雞鳴狗吠 皆可得而書矣

"Even the sound of wind, the cry of a crane, a rooster’s crow, a dog’s bark — all can be written." — Jeong In-ji, preface to the Haerye

Source: Hunminjeongeum Haerye (1446)

Note: ㅿ ㆆ ㆁ ㆍ are four letters but only three of them are consonants — the fourth, ㆍ, is a vowel.

Combination Rules

Letters are building blocks

The Haerye does not stop at 28 shapes. It gives rules for stacking, doubling and clustering them — which is how a small alphabet reaches a very large sound space.

Why it works

A phonetic alphabet, 442 years before the IPA

Most alphabets grew out of pictures: Latin letters descend from Egyptian hieroglyphs, and their shapes say nothing about how a sound is made. Hangul was designed the other way round. Each consonant is drawn from the place in the mouth where it is produced, and the Haerye explains the whole logic. The International Phonetic Alphabet, the linguists’ standard, was first published in 1888 (its association was founded in Paris in 1886). Hunminjeongeum was promulgated in 1446. Linguist Geoffrey Sampson called Hangul a “featural” alphabet for this reason: its letter shapes encode sound features.

The Haerye also pairs each place of articulation with one of the five elements, tying the sounds of speech into the cosmology of its day.

1. 1443

   Sejong creates 28 letters

   The Annals record that the king created 28 letters that could write both Chinese-character words and native speech.

2. 1446

   Hunminjeongeum promulgated

   The Haerye explains how the letters were designed, including the combination rules above.

3. 1447–48

   Dongguk Jeongun

   A dictionary that fixes the standard reading of Chinese characters in Hangul. Only two of its six volumes survive. Later critics found it artificial, and its system was dropped by the end of the 1400s.

4. 1455

   Hongmu Jeongun Yeokhun

   Standard Chinese pronunciation written in Hangul — direct evidence that Hangul was used to transcribe a foreign language.

5. 1888

   The IPA is first published

   Europe’s linguists agree on a shared phonetic alphabet, built from Latin letters.

6. 1933

   24 letters fixed

   The Hangul Spelling Unification Plan settles on the modern 24 letters. The lost letters were not needed for Korean, so they were dropped.

Beyond Korean

What the 24-letter version cannot write

Today, Japanese, Mandarin and English all get squeezed through the 24 modern letters: ざ and さ collapse, r and l merge, f and p merge. Sejong’s lost letters and combination rules fit many of these gaps surprisingly well. Below is our proposed notation for each language, with every row marked as historical or as our own extension.

Honest status: this is a research proposal. The letters and combination rules are Sejong’s; the sound-to-letter assignments for Japanese, Mandarin and English are our design, to be tested with linguists and native speakers. The downloadable app and the converter below do not use these letters yet — they still output ordinary hiragana from the 24 modern letters.

Can you type it today?

The letters exist. The keyboard does not.

Good news: Unicode already encodes almost every letter above, so nothing new has to be standardized for most of them. Bad news: a standard Korean keyboard has no key for any of them, and the Hangulpia app cannot produce them yet.

* From a community guide; not tested by us. It also reports that the keyboard only worked reliably in Firefox.

Current limits

• No default keys. Standard Dubeolsik has no key for ㅿ ㆆ ㆁ ㆍ or the light-sound letters. We could not confirm any built-in way to type them on macOS.
• No precomposed syllables. Unicode precomposes only the 11,172 modern syllables. A syllable like "ㅿ+ㅏ" must be built from conjoining jamo, which needs a font and renderer that support it. Community reports mention broken display in some browsers.
• Two kinds of jamo. Compatibility jamo (ㅿ U+317F, …) are standalone and do not combine into syllables; conjoining jamo combine but look different when alone.
• Our engine is modern-Hangul only. The decoder and its training table are built on the 24 modern letters, so extended letters are simply not part of what it handles today.
• New clusters have no key or code. ð (ㅿ+ㄷ) is our own cluster. ㅅㄷ exists in Unicode, but as the old s+t cluster, not as θ.
• Typing cost. The Windows layout leans on Shift for rare letters, and its own guide warns of frequent typos.

Where we want to take it

Sources & further reading: Hunminjeongeum text (Wikisource) · Obsolete Hangul jamo · Hunminjeongeum · Hangul Compatibility Jamo · Hangul Jamo (Unicode) · Old Hangul typing guide (Wikisource user page) · Lecture “Why is Hangul considered a world-class script?” (Design House) — basis of the Hanja-sound bridge table. Reconstructed sound values of obsolete letters are scholarly estimates, not settled facts. Typing instructions come from a community guide and have not been tested by us.

Live Now

First up: Japanese

Type on your regular Dubeolsik keyboard, get real-time Hiragana. A statistical engine trained on a 4,153-word vocabulary, plus explicit grammar rules for particles and verb endings, fills in what a plain rulebook alone would miss.

• Real-time conversion — Dubeolsik in, Hiragana out, no retraining needed
• Statistically trained — a 4,153-word vocabulary plus explicit particle/conjugation grammar rules catch what character-level statistics alone miss
• Manual override markers — ~ for long vowels, ^ for dakuten, for the cases statistics alone gets wrong
• Built for macOS — native InputMethodKit app, ready today. Windows is next.

Heads up: this is a research prototype, not a finished product. It works best on common everyday words (about 4,000 in its vocabulary) and can get particles, small っ, long vowels and voiced sounds wrong. Please don't rely on it for important writing yet — feedback is very welcome.

Setup

Up and running in 3 minutes

No App Store, so macOS needs a few extra steps to trust it. Do these once, in order.

Want it gone? Same folder, run bash uninstall.sh.

Fine control

When the engine guesses wrong, correct it

The engine is statistical, not a fixed rulebook — so it occasionally misses a long vowel or picks the unvoiced version of a consonant. Two markers let you override it, syllable by syllable, without retraining anything.

Stack both on the same syllable: 카^~ → があ. Everything else in the sentence still goes through the trained engine as usual — markers only touch the syllable they're attached to.

One Script, Every Language

What's next?