LanguageDetector Polyfill

A polyfill for the LanguageDetector Web API that uses Google's CLD3 (Compact Language Detector v3) neural network model via WebAssembly cld3 binary for accurate language detection.

Features

• Browser-agnostic implementation of the LanguageDetector Web API specification
• Uses Google's CLD3 neural network model (same as Chrome's built-in detector)
• WebAssembly-based for high performance
• Supports 104 languages (see CLD3 supported languages)
• TypeScript support with full type definitions

Installation

npm install languagedetector-polyfill

Usage

Auto install

import { installPolyfill } from 'languagedetector-polyfill';

// Install globally (only if native API is not available)
installPolyfill();

// Now use the standard Web API
const detector = await LanguageDetector.create();
const results = await detector.detect('Hello world');

Script Tag (UMD)

For classic <script> tags without module bundlers:

<script src="https://unpkg.com/languagedetector-polyfill"></script>
<script>
  (async () => {
    const { installPolyfill } = LanguageDetectorPolyfill;
    installPolyfill();

    const detector = await LanguageDetector.create();
    const results = await detector.detect('Bonjour le monde!');
    console.log(results[0].detectedLanguage); // "fr"
  })();
</script>

Or via jsdelivr:

<script src="https://cdn.jsdelivr.net/npm/languagedetector-polyfill"></script>

Script Tag (ESM)

For <script type="module">, use the explicit ESM path:

<script type="module">
  import { installPolyfill } from 'https://unpkg.com/languagedetector-polyfill/dist/languagedetector-polyfill.js';
  installPolyfill();

  // Now LanguageDetector is available globally
  const detector = await LanguageDetector.create();
  const results = await detector.detect('Bonjour le monde!');
  console.log(results[0].detectedLanguage); // "fr"
</script>

Additionally, LanguageDetector ESM could be used as a fallback for the native LanguageDetector Web API

import { LanguageDetector as LanguageDetectorFallback } from 'languagedetector-polyfill';

let detector = await window.LanguageDetector.create();

try {
  detector.detect('Bonjour le monde!');
} catch (e) {
  detector = await LanguageDetectorFallback.create();
}

// Detect language
const results = await detector.detect('Bonjour le monde!');
console.log(results[0].detectedLanguage); // "fr"
console.log(results[0].confidence);       // 0.99

// Clean up when done
detector.destroy();

Limitations

Expected Languages

Important: Per the Web API specification, expectedInputLanguages is used as an optimization hint during model initialization. The CLD3 WASM module loads all languages at once, so this expectedInputLanguages API is provided for compatibility but it doesn't affect detection.

const detector = await LanguageDetector.create({
  expectedInputLanguages: ['en', 'es', 'fr', 'de']
});

API Reference

Follows LanguageDetector Web API

LanguageDetector.availability(options?)

Check if the detector is available for a given configuration.

Parameters:

• options.expectedInputLanguages (string[]): Expected input language codes

Returns: Promise<'available' | 'downloadable' | 'downloading' | 'unavailable'>

LanguageDetector.create(options?)

Create a new LanguageDetector instance.

Parameters:

• options.expectedInputLanguages (string[]): Optimization hint for expected input languages (provided for compatibility only, it doesn't affect detection)
• options.signal (AbortSignal): Cancel creation
• options.monitor (function): Download progress callback

Returns: Promise<LanguageDetector>

detector.detect(text)

Detect the language(s) of the provided text.

Parameters:

• text (string): Text to analyze

Returns: Promise<LanguageDetectionResult[]>

interface LanguageDetectionResult {
  detectedLanguage: string;  // BCP 47 language code
  confidence: number;        // 0-1 confidence score
}

detector.measureInputUsage(text)

Measure how much input quota would be used. As cld3-asm does not expose similar API so it simply returns the length of text.

Returns: Promise<number>

detector.destroy()

Release resources. Call when done using the detector.

detector.inputQuota (readonly)

Available input quota for detection operations. As cld3-asm does not expose similar API this property returns the remaining number of characters left after subtracting all the characters passed to detector.detect(text). Initialized with 10000, minimum value is 0.

detector.expectedInputLanguages (readonly)

Array of expected input language codes.

LanguageDetector.dispose()

Static method to cleanup CLD3 resources globally.

Supported Languages

CLD3 supports 104 languages with BCP-47 language codes. For the complete list, see CLD3 Supported Languages.

How It Works

This polyfill uses cld3-asm, a WebAssembly port of Google's CLD3:

1. Neural Network Model: CLD3 uses a neural network trained on text from the web to identify languages

2. On-Device Processing: All detection runs locally in WebAssembly - no data is sent to any server

3. High Accuracy: The same model used in Chrome's built-in LanguageDetector API

4. Lazy Loading: The WASM binary (emscripten-generated cld3.js) (~1MB) is loaded on first use

Comparison with Native API

Feature	Native API	Polyfill
Model	CLD3 (on-device)	CLD3 (WASM)
Privacy	Full	Full (no network)
Accuracy	Very High	same
Size	~1MB (system)	6.45 kB polyfill + 1,057.17 kB lazy-loaded WASM binary OR 1,046.90 kB UDM bundle
Languages	104	104

Browser Support

• Firefox 90+
• Safari 14+
• Edge 90+
• Chrome 90+ (please prefer using native API in v138+)

No Node.js Support

Roadmap

• Replace outdated cld3-asm/emscripten-wasm-loader with modern ESM-compatible WASM build

Development

# Install dependencies
npm install

# Run tests
npm test

# Build
npm run build

# Watch mode
npm run dev

License

• languagedetector-polyfill: MIT
• cld3-asm: MIT
• cld3: original license

Credits

• CLD3 - Google's Compact Language Detector v3 (CLD3)
• cld3-asm - WebAssembly port of CLD3