In Lafayette, Louisiana, a team of linguists and local researchers is employing artificial intelligence to help preserve Louisiana French, an endangered dialect intrinsic to the region’s cultural heritage. The initiative is spearheaded by Joshua Caffery, director of the Center for Louisiana Studies at the University of Louisiana, who aims to integrate the language into modern digital tools to ensure its survival.

The effort emerged from challenges posed by mainstream voice assistants like Amazon’s Alexa, which failed to recognize iconic Cajun musician Dewey Balfa, instead directing users to unrelated modern pop music. Such incidents highlight the disconnect between automated systems and regional dialects, underscoring the urgency to develop language models that accommodate underrepresented languages.

Louisiana French, deeply entwined with the Bayou’s identity and history, faces threats from generational language loss, pressures of anglicization, and limited representation in technology. Once the predominant language in South Louisiana, its use declined significantly following the 1921 state constitution designating English as the official language, and institutional discouragement of French in schools. According to recent estimates, fewer than 120,000 speakers remain.

Caffery and his team are addressing these challenges by training an automatic speech recognition model on a large archive of oral histories, songs, and interviews, much of which has been preserved on legacy formats including wax cylinders and reel-to-reel tapes. The Center for Louisiana Studies houses more than 12,000 hours of audio that document Cajun and Creole folklore, provided in part by cultural activist Barry Jean Ancelet.

The development process involves linguists Amanda LaFleur and Colby LeJeune, who transcribe audio files to create a "ground-truth" dataset for machine learning. Research scientist Peyton Leathem-Boe is advancing the project through an open-source model the team has named “tataville,” after a creature from Louisiana French folklore.

Testing the model with recordings such as the nursery rhyme “Trois Jolis Tambours” demonstrated that the tailored AI system significantly outperforms generic speech recognition tools, which often misinterpret the dialect as unrelated languages like Hindi or Arabic. While not yet flawless, the specialized model effectively recognizes unique grammatical features and vocabulary specific to Louisiana French, which are absent from mainstream data sources.

Experts note broader implications beyond cultural preservation. Christine Mallinson, a professor of language and culture, emphasized that automated speech recognition accuracy affects critical areas such as employment and healthcare, where language bias in AI can exacerbate social inequities. Technologies that better understand and represent diverse linguistic varieties can reduce such disparities.

Efforts to use AI for language revitalization are also evident worldwide, with institutions applying similar techniques to Scottish Gaelic, Many, and various African languages. Caffery envisions that the Louisiana French model will not only aid preservation but also inspire contemporary artistic expression. As a musician himself, he hopes it will enable singers and creators fluent in the dialect to engage more deeply with their linguistic heritage.

Ultimately, this project aims to integrate Louisiana French into the digital sphere in a way that respects community ownership and cultural continuity, ensuring that the language remains a living, evolving part of the region’s identity rather than a relic of the past.