Typed Korean

If you can write TypeScript, you can understand Korean!

New: Interactive Playground → — an in-browser TypeScript editor where the compiler resolves conjugations live, plus a Verb Lab, Adjective Lab, Phrase Builder, and Sentence Gallery. (source · run locally with cd playground && pnpm install && pnpm dev)

Typed Korean is a TypeScript type-level library that enables the expression of complete Korean sentences through the type system. It creates a domain-specific language (DSL) based on Korean grammar rules, allowing a subset of grammatically correct natural language to be written and verified using TypeScript's compiler.

This project also explores an intermediate format for AI in language learning. For example, LLMs could return grammar analysis of Korean sentences using this format instead of JSON, enabling verification through TypeScript's type checker to improve correctness.

📖 Want to learn more? Check out our detailed blog post which explains how the TypeScript type system can be used to learn Korean grammar from the ground up. The article starts with basic programming concepts and gradually builds up to complex Korean grammatical structures like conditional sentences and interrogative phrases.

// Define the proper noun "ヒンメル"
type ヒンメル = ProperNoun<"ヒンメル">;

// Define する verb
type する = IrregularVerb & { dictionary: "する" };

// Create the そうした pattern (past form of そうする)
type そうした = DemonstrativeAction<Demonstrative & "そう", する, "Ta">;

// Create the conditional phrase "ヒンメルならそうした"
type ヒンメルならそうした = ConditionalPhrase<ヒンメル, "なら", そうした>;

// Type checking examples
const properExample: ヒンメルならそうした = "ヒンメルならそうした"; // "If it were Himmel, he would do so"
// 如果是辛美尔的话，他也会这么做的

🤖 Verb System

Verb Classes

Korean verbs are categorized into three main classes:

1. Godan Verbs (五段動詞) - Also known as "Group 1" or "u-verbs"

   • Endings: う, く, ぐ, す, つ, ぬ, ぶ, む, る
   • Examples: 話す (hanasu - to speak), 書く (kaku - to write)

2. Ichidan Verbs (一段動詞) - Also known as "Group 2" or "ru-verbs"

   • Always end with る
   • Examples: 食べる (taberu - to eat), 見る (miru - to see)

3. Irregular Verbs (不規則動詞) - Only two main verbs

   • する (suru - to do)
   • 来る (kuru - to come)

Verb Conjugation Forms

The system supports these conjugation forms:

• Dictionary (Dictionary form)
• Masu (Polite form)
• Te (Te form)
• Ta (Past form)
• Nai (Negative form)
• Potential (Potential form)
• Passive (Passive form)
• Causative (Causative form)
• Volitional (Volitional form)
• Imperative (Imperative form)
• Conditional (Conditional form)
• Hypothetical (Hypothetical form)

type 買う = GodanVerb & { stem: "買"; ending: "う" };
type 買うTe = ConjugateVerb<買う, "Te">; // 買って
type 買うTa = ConjugateVerb<買う, "Ta">; // 買った

type 食べる = IchidanVerb & { stem: "食べ"; ending: "る" };
type 食べるTe = ConjugateVerb<食べる, "Te">; // 食べて
type 食べるTa = ConjugateVerb<食べる, "Ta">; // 食べた

🎨 Adjective System

Korean adjectives are categorized into two main classes:

1. I-Adjectives (い形容詞) - End with い

   • Examples: いい (good), 楽しい (fun), 高い (expensive)

2. Na-Adjectives (な形容詞) - Require な when modifying nouns

   • Examples: 綺麗 (pretty), 静か (quiet), 好き (liked)

Adjective Conjugation Forms

The system supports these conjugation forms for adjectives:

• Basic (Basic form)
• Polite (Polite form)
• Past (Past form)
• Negative (Negative form)

type いい = IAdjective & { stem: "い"; ending: "い"; irregular: true };
type 綺麗 = NaAdjective & { stem: "綺麗" };

🔗 Copula System

The copula (だ / です) turns a 体言 (noun or na-adjective stem) into a statement — "X is …". It is not a particle: like a verb or an adjective it inflects for politeness, tense and polarity, so it is modeled as a conjugable word via ConjugateCopula<Taigen, Form> (mirroring ConjugateVerb / ConjugateAdjective).

Copula Conjugation Forms

• Plain → だ (plain) ・ Polite → です (polite)
• Past → だった ・ PolitePast → でした
• Negative → ではない ・ PoliteNegative → ではありません
• NegativePast → ではなかった ・ PoliteNegativePast → ではありませんでした
• CasualNegative → じゃない ・ CasualPoliteNegative → じゃありません
• Written → である（formal written）・ Te → で（connective）

The copula has no generic attributive form: な is licensed only for な-adjectives (静かな町), never for plain nouns (×医者な — a noun takes の), so it lives in the adjective system, not here.

type 医者だ = ConjugateCopula<"医者", "Plain">; // 医者だ
type 医者ではありません = ConjugateCopula<"医者", "PoliteNegative">; // 医者ではありません

📚 Phrase and Sentence Composition

The system now supports:

• Adjectives and verbs with particles
• Connecting phrases with Korean punctuation
• Basic sentence structures
• Conditional expressions with particles like なら
• Demonstrative forms with actions

Example: Connecting simple adjective and imperative verb phrases

// I-adjective "ii" (good) with irregular conjugation
// Then add particle "yo" to basic form of "ii" -> "ii yo"
type いい = IAdjective & { stem: "い"; ending: "い"; irregular: true };
type いいよ = PhraseWithParticle<ConjugateAdjective<いい, "Basic">, "よ">;

// Irregular verb "kuru" (to come)
// Then add particle "yo" to imperative form of "kuru" -> "koi yo"
type 来る = IrregularVerb & { dictionary: "来る" };
type 来いよ = PhraseWithParticle<ConjugateVerb<来る, "Imperative">, "よ">;

// Connect both phrases -> "ii yo, koi yo"
type いいよ来いよ = ConnectedPhrases<いいよ, 来いよ>;

// Type checking examples
const correctPhrase1: いいよ = "いいよ"; // "It's good!" (114)
const correctPhrase2: 来いよ = "来いよ"; // "Come here!" (514)
const correctFullPhrase: いいよ来いよ = "いいよ、来いよ"; // "It's good, come here!"

Example: More flexible component-based sentence construction

type SentenceParts = [
  AdverbPart<"なんで">, // "Why" - question adverb
  IntensifierPart<"そんなに">, // "So much" - intensifier
  VerbPart<慣れる, "Te">, // "Get used to" in te-form
  ContractedPart<"ん">, // Contraction of "の" - colloquial nominalizer
  CopulaPart<"Plain">, // Copula "is" (だ) — a conjugable copula, not a particle
  ParticlePart<"よ"> // Emphatic sentence-ending particle
];

// Combines all parts into a single string
type JoinedSentence = JoinPhrasePartsValue<SentenceParts>;
const joinedSentence: JoinedSentence = "なんでそんなに慣れてんだよ"; // "Why are you so used to it?!"
// 你为什么这么熟练啊？

⚙️ Technical Implementation

The system uses TypeScript's template literal types, conditional types, and mapped types to create a purely type-level representation of Korean grammatical rules.

Key components:

• Type definitions for grammatical elements
• Rule mapping via conditional types
• String literal manipulation for form generation
• Type inference for grammatical validation

💡 Why Typed Korean?

• Educational tool - Learn Korean grammar through code
• AI-assisted learning - Provide structured formats for language analysis
• Grammar verification - Express and verify Korean grammar in code
• Integration potential - Basis for typed Korean language tools

⚠️ Limitations

• This is a type-level system only - it doesn't provide runtime functionality
• The system handles standard forms but doesn't account for linguistic nuances
• Some rare or archaic language patterns may not be accurately represented

This project is still in very early stages and heavily relies on LLM-generated grammar rules, which may occasionally contain hallucinations or inaccuracies. If you find any issue during actual use, please help by confirming and providing feedback.

🛠️ Development

If you're interested in contributing to or experimenting with Typed Korean:

1. Ensure you have Node.js and pnpm installed
2. Clone the repository
3. Install dependencies: pnpm install
4. Run the tests: pnpm test

The tests validate that the type system functions correctly and all grammatical rules are properly implemented.

We welcome contributions! Feel free to open issues for bugs or feature requests, or submit pull requests with improvements.

📬 Contact

For sponsorship opportunities, research collaborations, or commercial inquiries, please reach out to contact@typedgrammar.com.

⚖️ License

MIT

Copyright (c) 2025-present, Yifeng Wang