Glossa Node.js

High-accuracy language detection for Indian languages, built for Node.js 20+.

Glossa Node.js is dependency-free, fast to start, and suitable for production APIs, CLIs, data pipelines, search indexing, routing, moderation, analytics, and content-localization workflows.

Highlights

• Detects 14 languages: English, Hindi, Marathi, Bengali, Assamese, Tamil, Telugu, Gujarati, Kannada, Malayalam, Odia, Punjabi, Urdu, and Maithili.
• Strong script detection for Indian scripts.
• Same-script disambiguation for Hindi/Marathi and Bengali/Assamese.
• Code-mixed input handling for Latin + Indic text.
• Zero runtime dependencies.
• CommonJS package with TypeScript declarations.
• CLI included.
• Node.js 20 and above.

Installation

npm install glossa-nodejs

Quick start

const { detectLanguage, detectLanguageWithConfidence } = require('glossa-nodejs');

detectLanguage('नमस्ते आप कैसे हैं');
//=> 'hi'

detectLanguage('मी ठीक आहे तू कसा आहेस');
//=> 'mr'

detectLanguageWithConfidence('আমি বাড়ি থেকে এসেছি');
//=> { language: 'bn', confidence: 0.99 }

Supported languages

| Code | Language | Script | |---|---|---| | en | English | Latin | | hi | Hindi | Devanagari | | mr | Marathi | Devanagari | | mai | Maithili | Devanagari | | bn | Bengali | Bengali | | as | Assamese | Bengali | | ta | Tamil | Tamil | | te | Telugu | Telugu | | gu | Gujarati | Gujarati | | kn | Kannada | Kannada | | ml | Malayalam | Malayalam | | or | Odia | Odia | | pa | Punjabi | Gurmukhi | | ur | Urdu | Arabic | | unknown | Unknown / not enough signal | - |

Examples

Detect language code

const { detectLanguage } = require('glossa-nodejs');

detectLanguage('Hello, how are you today?');
//=> 'en'

detectLanguage('வணக்கம் நீங்கள் எப்படி இருக்கிறீர்கள்');
//=> 'ta'

detectLanguage('السلام علیکم آپ کیسے ہیں');
//=> 'ur'

Detect with confidence

const { detectLanguageWithConfidence } = require('glossa-nodejs');

detectLanguageWithConfidence('નમસ્તે તમે કેમ છો');
//=> { language: 'gu', confidence: 0.9 }

Confidence is a value from 0 to 1. Script-distinct languages usually return high confidence. Same-script pairs use additional grammar and n-gram signals.

Batch detection

const { detectLanguage } = require('glossa-nodejs');

const texts = [
  'Hello world',
  'नमस्ते दुनिया',
  'নমস্কাৰ মই ভালে আছো',
  'నమస్కారం మీరు ఎలా ఉన్నారు',
];

const results = texts.map((text) => ({
  text,
  language: detectLanguage(text),
}));

console.log(results);

Handle unknown and short input

const { detectLanguage, detectLanguageWithConfidence } = require('glossa-nodejs');

detectLanguage('');
//=> 'unknown'

detectLanguage('??');
//=> 'unknown'

detectLanguageWithConfidence(null);
//=> { language: 'unknown', confidence: 0 }

Restrict allowed languages

const {
  LanguageDetector,
  DetectorConfig,
  Language,
} = require('glossa-nodejs');

const config = DetectorConfig.builder()
  .allowedLanguages([Language.Hindi, Language.Marathi])
  .minConfidence(0.6)
  .build();

const detector = new LanguageDetector(config);

detector.getSrcLanguage('मी ठीक आहे');
//=> 'mr'

Same-script disambiguation

const { detectLanguage } = require('glossa-nodejs');

detectLanguage('मैं ठीक हूँ आप कैसे हैं');
//=> 'hi'

detectLanguage('मी ठीक आहे तू कसा आहेस');
//=> 'mr'

detectLanguage('আমি বাড়ি থেকে এসেছি');
//=> 'bn'

detectLanguage('নমস্কাৰ মই ভালে আছো');
//=> 'as'

Code-mixed text

const { detectLanguage } = require('glossa-nodejs');

detectLanguage('hello नमस्ते आप कैसे हैं');
//=> 'hi'

Full language example set

const { detectLanguage } = require('glossa-nodejs');

const examples = {
  en: 'The weather is beautiful today',
  hi: 'मैं ठीक हूँ आप कैसे हैं',
  mr: 'मी ठीक आहे तू कसा आहेस',
  bn: 'আমি বাড়ি থেকে এসেছি',
  as: 'নমস্কাৰ মই ভালে আছো',
  ta: 'வணக்கம் நீங்கள் எப்படி இருக்கிறீர்கள்',
  te: 'నమస్కారం మీరు ఎలా ఉన్నారు',
  gu: 'નમસ્તે તમે કેમ છો',
  kn: 'ನಮಸ್ಕಾರ ನೀವು ಹೇಗಿದ್ದೀರಿ',
  ml: 'നമസ്കാരം നിങ്ങൾ എങ്ങനെയുണ്ട്',
  or: 'ନମସ୍କାର ଆପଣ କେମିତି ଅଛନ୍ତି',
  pa: 'ਸਤਿ ਸ੍ਰੀ ਅਕਾਲ ਤੁਸੀਂ ਕਿਵੇਂ ਹੋ',
  ur: 'السلام علیکم آپ کیسے ہیں',
};

for (const [expected, text] of Object.entries(examples)) {
  console.log(expected, detectLanguage(text));
}

API reference

detectLanguage(text)

Returns an ISO-style language code, or unknown.

detectLanguage(text: string): string

detectLanguageWithConfidence(text)

Returns a language code and confidence score.

detectLanguageWithConfidence(text: string): {
  language: string;
  confidence: number;
}

detectLanguageWithModel(text, model)

Runs detection with a specific model type.

const { detectLanguageWithModel, ModelType } = require('glossa-nodejs');

detectLanguageWithModel('Hello world', ModelType.ScriptOnly);
//=> 'en'

Available model values:

| Model | Notes | |---|---| | ModelType.Whatlang | Default lightweight provider | | ModelType.Ensemble | Ensemble wrapper over available providers | | ModelType.ScriptOnly | Script-based detection only | | ModelType.Cld2 | Accepted for compatibility; uses built-in fallback | | ModelType.Cld3 | Accepted for compatibility; external provider not bundled | | ModelType.Lingua | Accepted for compatibility; external provider not bundled | | ModelType.FastText | Accepted for compatibility; external model not bundled |

LanguageDetector

Reusable detector class for custom configuration.

const detector = new LanguageDetector(DetectorConfig.default());

detector.detect('Hello world');
//=> 'English'

detector.detectWithConfidence('Hello world');
//=> ['English', 0.9]

detector.getSrcLanguage('Hello world');
//=> 'en'

DetectorConfig

const config = DetectorConfig.builder()
  .model(ModelType.Whatlang)
  .minConfidence(0.5)
  .minTextLength(3)
  .enableDisambiguation(true)
  .disambiguationThreshold(0.7)
  .allowedLanguages([])
  .build();

minTextLength is measured in UTF-8 bytes. An empty allowedLanguages list means all supported languages are allowed.

Exports

const {
  VERSION,
  detectLanguage,
  detectLanguageWithConfidence,
  detectLanguageWithModel,
  LanguageDetector,
  DetectorConfig,
  DetectorConfigBuilder,
  ModelType,
  DetectionResult,
  Language,
  Script,
  code,
  fromCode,
  languageScript,
  allLanguages,
  modelName,
  modelFromString,
} = require('glossa-nodejs');

CLI

Install globally:

npm install -g glossa-nodejs

Or run with npx:

npx glossa-nodejs "नमस्ते आप कैसे हैं"
# hi

The command name is glossa.

glossa "नमस्ते आप कैसे हैं"
# hi

glossa --confidence "Hello world"
# en    0.9000

glossa --format json "আমি বাড়ি থেকে এসেছি"
# {"text":"আমি বাড়ি থেকে এসেছি","language":"bn"}

printf "வணக்கம்\nالسلام علیکم\n" | glossa --lines
# ta
# ur

CLI options:

| Option | Description | |---|---| | -c, --confidence | Include confidence score | | -f, --format <fmt> | text or json | | -m, --model <model> | whatlang, ensemble, script_only, script-only, script, or cld2 | | -l, --lines | Read stdin and detect one input line at a time | | -h, --help | Show help |

TypeScript

Type declarations are bundled.

import { detectLanguage, detectLanguageWithConfidence } from 'glossa-nodejs';
import type { LanguageCode } from 'glossa-nodejs';

const language: LanguageCode = detectLanguage('નમસ્તે તમે કેમ છો');
const result = detectLanguageWithConfidence('Hello world');

Accuracy and limitations

Glossa Node.js is strongest for script-distinct Indian languages and includes deterministic disambiguation for common same-script pairs.

Maithili shares Devanagari with Hindi and Marathi. Without a larger external model, short Maithili text may be classified as Hindi or Marathi. For user-facing flows, treat mai as best-effort unless text is long and distinctive.

The test suite includes 1,161 multilingual cases and currently passes a 93%+ accuracy gate on public test data.

Production notes

• Reuse LanguageDetector for high-throughput services with custom config.
• Use detectLanguage for simple one-off calls.
• Inputs that are empty, too short, punctuation-only, or unsupported return unknown.
• No network calls are made at runtime.
• No model download is required.

Development

npm test
npm run lint
npm pack --dry-run

Author

Suraj Kumar Jha

License

MIT