Rumantsch Language Tools

A collection of tools for working with the Rumantsch (Romansh) language, published as an npm package.

Requirements

Node.js 18 or later (the Proofreader uses the global fetch API), or a modern browser. The package ships both CJS and ESM builds plus TypeScript types.

Installation

pnpm add @farscrl/rumantsch-language-tools

Features

• Tokenizer — splits text into individual tokens, handling Romansh-specific abbreviations and punctuation (based on stdlib-js/nlp-tokenize)
• Proofreader — hunspell-based spellchecker supporting all Romansh idioms

Supported idioms

| Code | Idiom | |------|-------| | rm-puter | Puter | | rm-rumgr | Rumantsch Grischun | | rm-surmiran | Surmiran | | rm-sursilv | Sursilvan | | rm-sutsilv | Sutsilvan | | rm-vallader | Vallader |

These codes are exported as the Idioms type.

Usage

Tokenizer

import { Tokenizer } from "@farscrl/rumantsch-language-tools";

Tokenizer.tokenize('In test dal tokenizer.');
// ['In', 'test', 'dal', 'tokenizer']

tokenize(text, keepPunctuation?, keepWhitespace?) — both flags default to false:

Tokenizer.tokenize('In test, dal tokenizer.', true);
// ['In', 'test', ',', 'dal', 'tokenizer', '.']

Proofreader

import { Proofreader } from '@farscrl/rumantsch-language-tools';

const proofreader = await Proofreader.CreateProofreader('rm-surmiran');

await proofreader.proofreadText('Text correct');
// [] — no errors

await proofreader.proofreadText('in');
// [{ word: 'in', offset: 0, length: 2 }]

await proofreader.getSuggestions('corect');
// ['correct', ...]

CreateProofreader fetches the idiom's .aff/.dic files, compiles a WebAssembly-backed Hunspell instance, and resolves once it's ready. It throws if the dictionary fails to load. Once resolved, proofreader.isLoaded is true and proofreader.version holds the dictionary's version string.

Dictionary files are fetched from https://www.spellchecker.pledarigrond.ch by default. To use a self-hosted mirror, pass a baseUrl option:

const proofreader = await Proofreader.CreateProofreader('rm-surmiran', {
  baseUrl: 'https://your-host.example.com'
});

Call unload() when you are done to free the dictionary from memory. The instance cannot be used afterwards:

proofreader.unload();

Development

pnpm install      # install dependencies
pnpm test         # run tests
pnpm run build    # compile to lib/
pnpm run lint     # lint

Publishing a new version

There is no CI publish pipeline — releases are cut locally.

1. On an up-to-date main with a clean working tree, make sure you're logged in to npm (npm whoami).

2. Bump the version, which lints (preversion), formats and stages src/ (version), then commits, tags, and pushes both the commit and the tag (postversion):

   pnpm version patch   # or: minor / major

3. Publish to npm. prepublishOnly runs the full test suite and lint before publishing:

   pnpm publish

   If this is the first publish under a new scope, add --access public.