react-native-lingua

An accurate natural language detection library for React Native, powered by lingua-rs.

Features

• 🎯 Highly Accurate: Uses advanced n-gram models (1-5) for superior detection accuracy
• 📦 75 Languages: Supports detection of 75 languages
• ⚡ Fast: Built with Rust and JSI for optimal performance
• 🔒 Type-Safe: Full TypeScript support
• 📱 Cross-Platform: Works on iOS and Android
• 🌐 Offline: No internet connection required

Installation

Prerequisites

The native libraries are built automatically during installation. You'll need:

1. Rust toolchain:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

2. cargo-ndk (for Android):

cargo install cargo-ndk

3. Android NDK (for Android):

   • Recommended version: 26.1.10909125
   • Set environment variables:

   export ANDROID_SDK_ROOT=$HOME/Library/Android/sdk
   export ANDROID_HOME=$HOME/Library/Android/sdk
   export ANDROID_NDK_HOME=$ANDROID_HOME/ndk/26.1.10909125

4. Xcode (for iOS on macOS)

Install the Package

pnpm add react-native-lingua
# or
npm install react-native-lingua
# or
yarn add react-native-lingua

The native libraries will be built automatically during the native build process (Xcode/Gradle).

iOS Setup

cd ios && pod install

Android Setup

No additional steps required.

Manual Build (Optional)

If you need to rebuild the native libraries manually:

cd node_modules/react-native-lingua
npm run build:rust

Usage

Basic Language Detection

import {
  createDetectorForAllLanguages,
  createDetectorForLanguages,
  detectLanguage,
} from 'react-native-lingua';

// Create a detector for all supported languages
const detector = createDetectorForAllLanguages();

// Detect language
const language = detectLanguage(detector, 'Hello, how are you?');
console.log(language); // 'en'

// Or create a detector for specific languages (faster and more accurate)
const specificDetector = createDetectorForLanguages(['en', 'fr', 'de', 'es']);
const lang = detectLanguage(specificDetector, 'Bonjour le monde');
console.log(lang); // 'fr'

Language Confidence

Get confidence scores to understand how certain the detection is:

import {
  createDetectorForLanguages,
  computeLanguageConfidence,
  computeLanguageConfidenceValues,
} from 'react-native-lingua';

const detector = createDetectorForLanguages(['en', 'fr', 'de', 'es']);

// Get confidence for a specific language
const confidence = computeLanguageConfidence(detector, 'Hello world', 'en');
console.log(confidence); // 0.95

// Get confidence values for all languages
const confidences = computeLanguageConfidenceValues(detector, 'Hello world');

console.log(confidences);
// [
//   { language: 'en', confidence: 0.95 },
//   { language: 'fr', confidence: 0.03 },
//   { language: 'de', confidence: 0.01 },
//   { language: 'es', confidence: 0.01 }
// ]

Supported Languages

The library supports 75 languages using ISO 639-1 codes. Some examples:

const commonLanguages = [
  'en', // English
  'es', // Spanish
  'fr', // French
  'de', // German
  'it', // Italian
  'pt', // Portuguese
  'ru', // Russian
  'ja', // Japanese
  'zh', // Chinese
  'ko', // Korean
  'ar', // Arabic
  'hi', // Hindi
  // ... and 63 more
];

For a complete list of supported languages, see the lingua-rs documentation.

API Reference

createDetectorForAllLanguages()

Creates a language detector that can detect any of the 75 supported languages.

Returns: LanguageDetector

Note: Loading all languages consumes significant memory (~1 GB). Consider using createDetectorForLanguages() if you know which languages to expect.

createDetectorForAllLanguages(options?: DetectorOptions)

Creates a language detector that can detect any of the 75 supported languages.

Parameters:

• options?: DetectorOptions - Optional configuration
  • minimumRelativeDistance?: number - Minimum confidence gap between top two language candidates (0.0-0.99). Higher values make detection more conservative, returning null for ambiguous text. Omit or use -1 to disable (default).

Returns: LanguageDetector

Note: Loading all languages consumes significant memory (~1 GB). Consider using createDetectorForLanguages() if you know which languages to expect.

Example:

// Default behavior
const detector = createDetectorForAllLanguages();

// With minimum relative distance to avoid false positives
const detector = createDetectorForAllLanguages({
  minimumRelativeDistance: 0.7,
});

createDetectorForLanguages(languages: string[], options?: DetectorOptions)

Creates a language detector for specific languages.

Parameters:

• languages: string[] - Array of ISO 639-1 language codes
• options?: DetectorOptions - Optional configuration (see above)

Returns: LanguageDetector

Example:

const detector = createDetectorForLanguages(['en', 'fr', 'de']);

// With minimum relative distance for short/ambiguous text
const detector = createDetectorForLanguages(['it', 'pt', 'es'], {
  minimumRelativeDistance: 0.8,
});

detectLanguage(detector: LanguageDetector, text: string)

Detects the language of the given text.

Parameters:

• detector: LanguageDetector - Language detector instance
• text: string - Text to analyze

Returns: string | null - ISO 639-1 language code or null if detection failed

Example:

const lang = detectLanguage(detector, 'Hello world'); // 'en'

computeLanguageConfidence(detector: LanguageDetector, text: string, languageCode: string)

Computes the confidence value for a specific language.

Parameters:

• detector: LanguageDetector - Language detector instance
• text: string - Text to analyze
• languageCode: string - ISO 639-1 language code

Returns: number - Confidence value between 0.0 and 1.0

Example:

const confidence = computeLanguageConfidence(detector, 'Hello world', 'en'); // 0.95

computeLanguageConfidenceValues(detector: LanguageDetector, text: string)

Computes confidence values for all languages in the detector.

Parameters:

• detector: LanguageDetector - Language detector instance
• text: string - Text to analyze

Returns: LanguageConfidence[] - Array of { language: string, confidence: number } sorted by confidence (descending)

Example:

const confidences = computeLanguageConfidenceValues(detector, 'Hello');
// [{ language: 'en', confidence: 0.95 }, ...]

Handling Ambiguous Text with minimumRelativeDistance

Language detection uses statistical models that can struggle with short text, especially between closely related languages (e.g., Italian vs Portuguese). The minimumRelativeDistance option helps control this behavior.

Problem: Short phrases like "Ciao" or "Amor" exist in multiple languages and may be misclassified.

Solution: Set a minimum relative distance threshold to require higher confidence before returning a result.

import {
  createDetectorForLanguages,
  detectLanguage,
} from 'react-native-lingua';

// Without threshold - may guess wrong for short text
const detector = createDetectorForLanguages(['it', 'pt']);
detectLanguage(detector, 'Ciao'); // Might return 'pt' instead of 'it'

// With threshold - returns null for ambiguous text
const detector = createDetectorForLanguages(['it', 'pt'], {
  minimumRelativeDistance: 0.7,
});
detectLanguage(detector, 'Ciao'); // Returns null (ambiguous)
detectLanguage(detector, 'Ciao, come stai oggi?'); // Returns 'it' (confident)

Recommended values:

Value	Behavior	Use Case
-1 or omitted	Disabled (default)	General text, longer sentences
0.0 - 0.3	Lenient	Mixed content, some ambiguity acceptable
0.4 - 0.6	Moderate	Short messages, related languages
0.7 - 0.9	Strict	Very short text, high accuracy required

Note: The value represents the minimum logarithmized probability distance between the top two candidates. Higher values mean stricter detection. Values above 0.99 are not allowed by the underlying library.

Performance Tips

1. Use specific languages: Create detectors with only the languages you need for better performance and accuracy
2. Reuse detectors: Create detector instances once and reuse them
3. Memory usage: The detector with all languages uses ~1 GB of memory. Use specific language sets to reduce memory footprint

How It Works

This library wraps the Rust lingua-rs library using:

1. Rust FFI: C-compatible FFI bindings for the lingua library
2. C++ JSI: JavaScript Interface for direct communication with JavaScript
3. React Native TurboModule: Modern architecture for optimal performance

The detection uses both rule-based and statistical Naive Bayes methods with n-grams of sizes 1-5, providing high accuracy even for short text snippets.

Contributing

See the contributing guide to learn how to contribute to the repository and the development workflow.

License

MIT

---

Made with create-react-native-library