Lingua Nostra: free AI chatbot for the Provençal language

The project

From digital flourishing to an artificial intelligence specialised in Provençal

Lingua Nostra wishes to contribute to the digital flourishing of Provençal, that is, to its presence on the Internet and in digital tools. We already publish a few resources for the Provençal-speaking community, and we are gradually gathering texts written in Provençal. We will then develop platforms in Provençal and natural language processing tools. This work prepares what lies at the heart of the project: a free and sovereign artificial intelligence specialised in Provençal, which we will train with the help of speakers.

Read the full project

Where things stand

The first pieces of work

Writing and reading aids are already available on the site: a conjugation tool, a dictionary, expressions and proverbs. They will help to create future automatic correctors and tools for recognizing and understanding the language, useful to the community as well as for processing the literary corpus. We are also beginning to collect all available writings, which may serve tomorrow to train the artificial intelligence model specialised in Provençal.

Indeed, training an artificial intelligence for Provençal requires building a quality corpus and obtaining the necessary permissions.

See the roadmap

Why

Irish

Basque

Catalan

A language with little presence in the digital world

Provençal is an endangered language. It has little presence in the digital world: no machine translation and no correctors, no help with writing or reading, no dictionaries built into keyboards, and very few spaces to read it, write it and communicate in it. Large language models, for their part, know it poorly: they write it imperfectly, mixed with French and Romance dialects, even with invented words and fanciful spelling. Yet a few minoritised languages have already entered the era of artificial intelligence and language processing: Irish, Basque with Latxa, and Northern Sami.

Following this path, we intend to train a generative artificial intelligence and make it available to those who want to write, create, publish and pass on Provençal, and thereby help to rebalance the place of the language in the digital space. In practical terms, this would mean an online chatbot, API access for developers, and the model made available to all.

A global movement

Several similar projects exist

As mentioned above, a few low-resource languages have already entered the era of artificial intelligence: Basque, Irish and Sami.

Provençal, for its part, has too few digital tools. Lingua Nostra therefore intends to contribute, for Provençal, to this international effort to preserve intangible heritage and linguistic diversity in the digital world.

Why Provençal

Why start with Provençal

An urgent need for preservation. Provençal, the historical language of Provence, is today endangered, both in its use and in its spelling, which is particularly threatened. We therefore want to contribute to its preservation and to its presence in the digital world.
Support for reclamation would help keep the language in use and passed on. Digital tools will therefore help those who use Provençal to write it, read it and pass it on.
Closeness to the language. The project leader speaks Provençal, which gives particular agility in carrying out the project (pre-correction, contacts, decisions) and speeds it up.
Today's large models write Provençal imperfectly, mixed with French and Romance dialects, even with invented words, while still claiming, wrongly, to have mastered it.

How

The envisaged method, in three steps

01

The corpus

Listing, digitising, collecting and publishing Provençal texts (journals, dictionaries, literature), then processing them. Target: between 50 and 100 million words.

02

Model adaptation

Building natural language processing tools: recognising Provençal texts, splitting words, analysing forms, then creating a first grammar corrector. These tools will help to prepare the training texts.

03

Community validation

Preparing the artificial intelligence: we will start from a large multilingual model that has already been trained and adapt it to Provençal in three stages: pre-training on corpora in Provençal, fine-tuning with instructions corrected by speakers, then alignment with their preferences. All three stages will be carried out in collaboration with speakers.

Feasibility

Precedents that have succeeded

Adapting a large AI model to a low-resource language has already been done. Several teams have achieved it, sometimes with very little data and modest hardware.

LlamaTurk

Turkish · 2024

Adapted from roughly 274 million words on near-consumer hardware. 89% accuracy in comprehension.

Qomhra

Irish Gaelic · 2025

Adaptation to a Celtic language structurally distant from English, validated by native speakers.

Northern Sami

Sami · 2024

Adapted with only 22 million words. Usable results despite an extremely small corpus.

Project	Language	Volume	Result
Northern Sami	Sami	~22 million words	usable despite a tiny corpus
Qomhra	Irish	~44 hours of compute (2 GPUs)	validated by native speakers
LlamaTurk	Turkish	~274 million words	89% accuracy in comprehension
Lou Pichoun	Provençal	50 to 100 million words targeted	to come

The challenge

Language sovereignty

A language poorly known to today's artificial intelligence risks fading from tomorrow's digital space.

Sovereignty is the ability of a given community to have free tools for processing its heritage language, without outsourcing to private companies. Lingua Nostra wishes for tools and a language model that the community can control: open and self-hosted models, independence from proprietary solutions.

By "free", we mean publishing the software, the data and the model under a free licence, opening the corpus where rights allow, and involving the language community in governance.

« Se tèn sa lengo, tèn la clau
que di cadeno lou deliéuro. »

"Who holds his language holds the key that frees him from his chains."

Frédéric Mistral, I Troubaire catalan, 1861

Lou Pichoun, Lingua Nostra's Provençal language model

Lou Pichoun

Lou Pichoun ("the little one"), a generative artificial intelligence, specialised in Provençal, is in preparation. It will be able to speak and write Provençal in its two main varieties, Rhodanian and Maritime, in its historical and heritage spelling. We will publish the model under a free licence, so that anyone can host it. In the long term, a free speech synthesis will allow it to read aloud; speech recognition could also be designed. As already mentioned, developing natural language processing tools and processing the available literary corpus in Provençal are preliminary steps to the training of this future artificial intelligence, and they will directly benefit the community of speakers and learners who care about reclaiming their intangible heritage.

Discover Lou Pichoun in detail →

What will be possible

Translating, conjugating, and writing in Provençal

These tools and this model will open up uses for those who want to live the language day to day.

Institutions

Enabling institutions and local authorities to write and communicate in Provençal, and to translate their communications into Provençal.

Artists

Helping artists and craftspeople create and share work in Provençal more easily.

General public

Writing correctly in Provençal in everyday life: naming a boat or a home, drafting a note or a sign. Answering a message, learning, translating, understanding...

Language development

Helping cultural associations proofread a newsletter or a poster. Self-correcting spelling, creating, communicating.

Digital presence

Helping to produce more Provençal content online, with a concern for the highest possible quality, without outsourcing to American or Chinese companies, nor even to Mistral, for free, open and sovereign access.

Long-term vision

Beyond Provençal

All our attention is therefore on Provençal. That said, the project's method, tools and results will serve other minoritised language communities in France, as well as the scientific community.

Provençal Montréal Sociology AI

Who leads the project

Vincent Drouin

Born in Nice, he grew up in La Seyne-sur-Mer, now lives in Montréal, Canada, and plans to spend at least twelve months in Provence over the next three years. A sociologist by training, he holds a master's degree in sociology from Université Paul-Valéry Montpellier 3, with a thesis on collective memory and the Provençal imaginary.

His professional experience spans fifteen years of web marketing and development, three of them developing automation and generative artificial intelligence solutions for small businesses. He will carry out the project's development, study, research and management himself.

Academic framework

Doctoral research

The project is set in the prospect of a doctorate, devoted to artificial intelligence applied to Provençal, which we will pursue with specialist researchers.

We could explore several interdisciplinary questions: how could a free language model, adapted to Provençal, contribute to the digital vitality of the language, and what effect would it have on its digital flourishing? What methods would make it possible to adapt it to specific tasks (translation, writing, teaching) with a limited corpus? Can a model be trained to master two closely related dialect varieties? What levels of corpus quality and quantity give the best possible results? What cognitive biases emerge from training on texts that mostly belong to a bygone era? What biases or alignment will emerge from training on texts in Provençal: will the model be conservative, liberal, federalist? As you will understand, the questions are many, whether technical or ethical.

Get involved

A place for everyone

Lingua Nostra speaks to different audiences: institutions and local authorities, businesses, the tourism sector, cultural associations, teachers, artists and craftspeople.

What we are looking for today

Public partners: regions, municipalities, departments, French and European cultural programmes
Private partners: businesses, individual donors, patrons and foundations
Civil society partners: Provençal associations and networks of speakers
Partner universities and laboratories

How to contribute

support the project, as an institution, an association or an individual
share Provençal texts, with the agreement of their rights holders
become a scientific partner
become a volunteer speaker: proofread the outputs and judge the future model's answers
spread the word in your community

Waiting list

Join the waiting list

Leave your address to follow the project and be notified of its key milestones.

Would you like to support the project, become a partner, or contribute scientifically? Everything is detailed on the full project page.

By leaving your address, you agree to receive news about the project. Unsubscribe at any time.

Do you care about these issues, or can you contribute to or support the project? Write to us.

contact@linguanostra.org

Frequently asked questions

Does a chatbot or AI that speaks Provençal already exist?

Generally available large models write Provençal imperfectly, mixed with French and Romance dialects, even with invented words. Lingua Nostra is preparing Lou Pichoun, an artificial intelligence that will be able to speak and write Provençal in its two main varieties. Its training has not begun. Sign up for the waiting list to follow the project's progress.

What is Lingua Nostra?

The project of digital flourishing of Provençal, in three parts: digital flourishing, with resources published for the Provençal-speaking community, natural language processing tools and an artificial intelligence in preparation. The first two parts prepare the training of the artificial intelligence.

Will it be possible to translate or conjugate in Provençal with artificial intelligence?

The site already offers a Provençal-French dictionary and conjugation. The artificial intelligence, in preparation, will in the long term make it possible to translate, answer questions, help with drafting and correction, and explain a word or a turn of phrase. A free speech synthesis will also allow it to read aloud. These features will arrive progressively.

How can I be notified of the launch?

Sign up for the waiting list: you will receive project announcements.

Join the waiting list