Harnessing the Power of Artificial Intelligence to Vitalize Endangered Indigenous Languages: Technologies and Experiences

Endangered Language Models and Writing Assistants

Context: Workshops with Indigenous communities in Brazil, specifically targeting young speakers and writers of endangered languages, such as Guarani Mbya and Nheengatu. These initiatives aim to engage students from Indigenous backgrounds in co-designing and testing AI-assisted writing tools that facilitate translation, spell-checking, and word prediction in their native languages.

Implementation: Developing and fine-tuning AI models to support writing in Indigenous languages through collaborative workshops. This includes creating prototypes for writing assistance and translation that are tested and refined based on user feedback from the community, emphasizing the co-design approach to ensure relevance and usability.

Outcomes: Increased interest among young people in using their native languages for writing; improved understanding of their linguistic heritage; development of usable prototypes for language documentation and support; and enhanced engagement with technology among Indigenous students.

Challenges: Limited data resources for fine-tuning models, ethical concerns regarding data use, building community trust, ensuring technology adoption and maintenance by Indigenous communities, and addressing varying literacy levels among users.

Technical barrier

Difficulty in obtaining sufficient high-quality data for training language models due to low-resourced languages. The tools and technologies are still in their infancy and require ongoing development and testing.

Proposed Solutions: Use of fine-tuning techniques with small datasets, leveraging existing linguistic resources, ensuring high data quality through careful curation, and engaging with diverse communities to iterate on the design based on user feedback.

Ethical barrier

Distrust from Indigenous communities towards researchers and technology developers due to historical exploitation.

Proposed Solutions: Implementing community engagement strategies, ensuring transparency, obtaining informed consent, and prioritizing Indigenous data sovereignty.

Cultural/Social barrier

Historical appropriation of technology by non-Indigenous peoples has led to skepticism and challenges in adoption by Indigenous communities.

Proposed Solutions: Empower Indigenous communities through training in technology development and encourage ownership of the tools created.

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai