belmorph

Беларуская

Zero-dependency Belarusian morphological analyzer

A fast, lightweight library for analyzing Belarusian words and their morphological properties.

Demo

Features

• Morphological analysis: Parse words to determine part of speech, case, gender, number, tense, aspect, animacy, comparison degree, and other grammatical properties
• Inflection: Generate different grammatical forms of words
• Lexeme generation: Get all possible forms of a word
• Morphological prediction: Unknown words are analyzed using suffix-based patterns
• Zero dependencies: No external runtime dependencies
• TypeScript support: Full type definitions included
• Efficient storage: Uses DAWG (Directed Acyclic Word Graph) for fast lookups
• Browser compatible: Works in Node.js, browsers, and Deno — no native dependencies
• Dictionary covers 6,574 inflection paradigms derived from GrammarDB

Installation

npm install belmorph

Quick Start

Node.js — filesystem (bundled dictionary)

import { MorphAnalyzer } from 'belmorph';
import { loadDict } from 'belmorph/node';

const morph = new MorphAnalyzer(loadDict()); // uses bundled dict/
const results = morph.parse('горад');

console.log(results[0].lemma);    // 'горад'
console.log(results[0].tags.pos); // 'N' (noun)

// Inflect to different forms
results[0].inflect({ case: 'I', number: 'P' })?.word; // 'гарадамі'

// Full names and short codes are interchangeable
results[0].inflect({ case: 'instrumental', number: 'plural' })?.word; // 'гарадамі'

// Get all forms
const lexeme = results[0].lexeme;
console.log(lexeme.map(r => r.word));

Browser / Deno / Node.js — HTTP

import { MorphAnalyzer } from 'belmorph';

// Serve the dict/ folder as static files and point to it:
const morph = await MorphAnalyzer.init('/dict/');
// or from a CDN:
const morph = await MorphAnalyzer.init('https://cdn.example.com/dict/');

const results = morph.parse('горад');

When a word is not found in the dictionary, the analyzer falls back to suffix-based prediction — those results have predicted: true and may be less accurate.

API

Loading

Method	Environment	Description
new MorphAnalyzer(dict)	all	Construct from a pre-loaded DictData object
MorphAnalyzer.init(baseUrl?)	all	Async factory — fetches and decompresses dict files over HTTP. Default base URL: '/dict/'
loadDict(dir?) from belmorph/node	Node.js only	Synchronous filesystem load. Defaults to the bundled dict/

ParseResult

MorphAnalyzer.parse(word) returns an array of ParseResult objects. Each result has:

• word — the inflected word form
• lemma — the dictionary form
• tags — grammatical properties (Grammeme interface, see src/tags.ts)
• predicted — whether the analysis was predicted
• inflect(target) — returns the form matching the given grammemes, or null
• pluralize(count, targetCase = 'N') — returns the form agreeing with count, following Belarusian numeral agreement rules. targetCase is the case the whole numeral phrase is governed by — nominative by default, or an oblique case (e.g. dative when governed by a verb or preposition)
• lexeme — all forms of the word

const book = morph.parse('кніга')[0];
book.pluralize(1)!.word;  // 'кніга' (1 кніга)
book.pluralize(2)!.word;  // 'кнігі'  (2 кнігі)
book.pluralize(5)!.word;  // 'кніг'   (5 кніг)

// Oblique government, e.g. "давяраю пяці кнігам" (dative, governed by the verb)
book.pluralize(5, 'D')!.word; // 'кнігам'

Grammeme values follow GrammarDB codes. Both short codes ('I') and full English names ('instrumental') are accepted by inflect().

Building the Dictionary

The library requires a pre-built dictionary. To build it:

git submodule update --init
npm run build:dict

This will create the necessary dictionary files in the dict/ directory.

Testing

npm test           # Run tests once
npm run test:watch # Run tests in watch mode

Tokenizer

Belmorph includes a streaming tokenizer for Belarusian text. It splits text into tokens (words, numbers, punctuation, spaces, hashtags, mentions, emails, and links).

import { Tokenizer, only, TokenType } from 'belmorph/tokens';

// One-shot tokenization
const tokens = Tokenizer.tokenize('Прывітанне, свет! #мова');
console.log(tokens.map(t => t.toString()));

// Streaming (for large documents)
const tz = new Tokenizer();
tz.append('Прывіта');
tz.append('нне, свет!');
const tokens = tz.done();

// Filter by type
const words = only(tokens, TokenType.WORD);
console.log(words.map(w => w.toString())); // ['Прывітанне', 'свет']

Token types

Type	Description
WORD	Cyrillic or Latin word (subtype: CYRIL, LATIN, MIXED)
NUMBER	Integer or decimal number
PUNCT	Punctuation mark
SPACE	Whitespace (spaces, tabs)
NEWLINE	Line break
EMAIL	Email address
LINK	URL (with protocol or starting with www)
HASHTAG	#hashtag
MENTION	@mention

Tokenizer options

const tz = new Tokenizer({
  hashtags: true,   // recognize #hashtags
  mentions: true,   // recognize @mentions
  emails: true,     // recognize emails
  links: true,      // recognize URLs
});

Text Analysis (TextAnalyzer)

TextAnalyzer combines the tokenizer and morphological analyzer for full text parsing.

import { TextAnalyzer, MorphAnalyzer } from 'belmorph';

const morph = await MorphAnalyzer.init('/dict/');
const analyzer = new TextAnalyzer(morph);

const result = analyzer.analyze('Горад прыгожы!');
// [{ token: Token("Горад"), parse: { lemma: "горад", tags: { pos: "N", ... } } },
//  { token: Token(" "),      parse: null },
//  { token: Token("прыгожы"), parse: { lemma: "прыгожы", tags: { pos: "A", ... } } },
//  { token: Token("!"),      parse: null }]

// Words only
const words = analyzer.words('кніга і тавар');
words.map(w => w.parse!.lemma); // ['кніга', 'тавар']

License

This project is dual-licensed:

• Source Code: MIT License
• Dictionary Data: CC BY-SA 4.0 (derived from GrammarDB)

See the LICENSE file for full details.

Credits

• Dictionary data from GrammarDB by Aleś Bułojčyk and contributors.
• Inspired by pymorphy2 and other morphological analyzers.