Lou Pichoun, the open-source Provençal LLM

The model

A sovereign generative AI for Provence

Lou Pichoun ("the little one" in Provençal) will be an open-source Provençal LLM, trained specifically on the Provençal language using natural language processing (NLP) methods. It will draw on a corpus of Provençal texts from literature, Félibrige almanacs, the historical press and the works of Mistral and the Félibres, gathered from major digital heritage libraries (Gallica, Internet Archive).

The approach chosen is the adaptation of an existing multilingual large model. Similar projects have already been completed for Irish, Basque, Turkish and Sami, with modest resources and convincing results. Lou Pichoun will follow this path, building on the corpus of the Provençal tradition and on a network of native speakers for validation.

Carried by Vincent Drouin, Lou Pichoun is conceived as a sovereign AI: free, open-source, controlled by the Provençal community. No prototype exists yet; the work under way focuses on building the corpus and finding scientific partners. Ultimately, Lou Pichoun will be able to run a Provençal conversational assistant capable of understanding and conversing, answering questions, translating and conjugating, assisting with writing, correction and reformulation, explaining a word or an expression, supporting creative work, and eventually reading Provençal aloud through a free text-to-speech system — all contributions to the digital preservation of the Provençal language.

Provençal resources

Resources already available

The model is still in development, but two building blocks of the project are already accessible: the dictionary for keyboards and spell-checkers, which allows you to write Provençal on computers, phones and office software without being incorrectly corrected, and which adds to your existing dictionaries without replacing them, and the conjugation of 4,804 verbs, available online.

Coming soon: an online bilingual Provençal-French dictionary.

A sovereign common good

More than a writing assistant

A model that merely produced text would be just another generalist. Lou Pichoun is designed to understand and converse in Provençal, to assist with translation, conjugation and creation, and eventually to carry a free Provençal speech synthesis. It is a tool at the service of the living language and those who keep it alive.

As such, it addresses distinct audiences, each of which finds a concrete benefit in it:

• Cultural institutions and local authorities — municipalities, regions, museums, libraries: a documentary and linguistic mediation tool for Provence.
• Tourism offices and businesses: enhance their image through Provençal culture and language.
• Artists and creators — poets, authors, musicians, filmmakers: support for composition and expression in Provençal.
• Researchers and universities: an open natural language processing (NLP) model designed for research on minority languages.
• Teachers: a pedagogical companion for the introduction and deepening of the Provençal language.
• The speaker community: everyone who speaks, reads or learns Provençal in daily life.