🇱🇺 Lëtzebuergesch Tools

Open source language tooling and educational resources for Luxembourgish

Mir wëlle bleiwe wat mir sinn — we want to remain what we are

Research Phase Open Source Education 🇱🇺 LLM Training
View on GitHub
6Gaps Identified
12+Existing Resources
~600k🇱🇺 Speakers
4Project Phases

E bësse Lëtzebuergesch 🇱🇺

A little Luxembourgish goes a long way.

Moien! 👋

Hello!

Äddi!

Goodbye!

Merci 💛

Thank you!

Wéi geet et?

How are you?

Jo / Nee

Yes / No

Wou ass ...?

Where is ...?

Ech verstinn net. 🤷

I don't understand.

Et geet mir gutt.

I'm doing well.

Äddi merci! 👋

Thanks, bye!

Vill Gléck! 🍀

Good luck!

Schwätzt Dir Lëtzebuergesch? 🇱🇺

Do you speak Luxembourgish?

Wannechgelift / Gär geschitt

Please / You're welcome

Existing Resources 🇱🇺

ZLS / Sproochmaschinn.lu

Official language center. STT (Whisper) and TTS (VITS2). Maintains spellchecker.lu and the LOD dictionary.

STTTTSOfficial

spellchecker.lu

HunSpell orthography checker, transferred to ZLS in 2023. Word list to be released CC0. Spelling only, no grammar.

HunSpellZLS-ownedCC0 pending

LOD (lod.lu)

Official multilingual dictionary with public API. 5 languages. Backbone dataset for ZLS TTS training.

DictionaryAPIOpen Data

LuxBank

First UD treebank for Luxembourgish (Nov 2024). Uni.lu research, not yet packaged for practical use.

UD TreebankResearch

LuxIT / LuxInstruct

Instruction tuning datasets (Oct 2025). LuxIT: 59'242 monolingual pairs. LuxInstruct: cross-lingual. Mixed fine-tuning results.

LLM TrainingDataset

LUXMT

GEMMA 3 27B fine-tuned for Luxembourgish to French/English translation (Feb 2026). Proves viable fine-tuning works.

TranslationFine-tuned

What's Missing

❌ No grammar checker

spellchecker.lu only validates orthography. Syntax, agreement, and grammatical structure go completely unchecked. The dative case, verb conjugation, and article agreement have no automated validation.

📏 No readability scorer

Teachers have no way to assess whether a text is appropriate for Cycle 1.2 or Cycle 4. No CEFR-aligned readability metric exists for Luxembourgish.

📋 No graded word lists

Luxembourg's school system has 4 cycles, but no vocabulary lists are aligned to those levels. Teachers build their own from scratch every year.

🤖 LLMs struggle with Luxembourgish

Even large models have weak grammatical understanding, especially morphology and syntax. Grammar-Book-Guided Probing (Oct 2025) found LLMs fail at minimal pair detection. Fine-tuning datasets show mixed results.

✏️ No writing support

Tools exist for checking after you write. Nothing helps you write correctly in the first place, with real-time suggestions and grammar explanations.

🏫 No educational integration

ZLS tools exist but aren't packaged for classroom use. Teachers need standalone, simple tools that work without API keys, logins, or technical setup.

Sproochentest 🇱🇺

The Sproochentest is the Luxembourgish language exam required for citizenship. Administered by INLL, it tests oral skills only. About 75€, pass rate is high with preparation.

🗣️ Speaking (A2 level)

  • 10-minute test, 2 parts
  • Part 1: Interview on a chosen topic (work, family, hobbies, food, travel, health, etc.)
  • Part 2: Describe one of three photos
  • Key grammar: dative case for spatial prepositions
  • Evaluated on vocabulary, grammar, fluency, clarity, coherence, interaction

👂 Listening (B1 level)

  • ~25-minute test, 3 audio clips
  • 📻 Radio news item
  • 💬 Everyday dialogue
  • 🎤 Interview or presentation
  • 3-7 multiple-choice questions per clip
  • Each clip played twice

🎯 What this means for our tools

A grammar checker must catch dative case errors (the most tested grammar point). A readability scorer should align with A2/B1 CEFR levels. Graded word lists should cover the 10 Sproochentest topic categories. Listening practice could use Liesmaschinn TTS. A practice mode could generate photo descriptions and evaluate them. Prepare for the Sproochentest →

Roadmap

0

Research & Contact

In Progress 🇱🇺
1

Core Tools

Planned
2

Integration

Future
3

LLM Training & Community

Future
View on GitHub Back to joelclaw.lu

Open source, open data, honest tools. 🇱🇺