Researchers at Brigham Young University are leveraging artificial intelligence to build translation tools for thousands of languages currently excluded from mainstream technology. The PathSay Project, run through BYU's MATRIX lab, pairs computer science students with international participants in BYU-Pathway Worldwide to gather speech and text data for low-resource languages.
The initiative addresses a critical gap in AI development. Most translation systems focus on high-resource languages like English, Mandarin, and Spanish, which have massive datasets available for training algorithms. Languages spoken by smaller populations lack the digital infrastructure needed to train effective AI models. Without adequate data, these languages remain "invisible" to modern translation technology, limiting access for millions of speakers seeking to preserve their linguistic heritage or participate in the digital economy.
PathSay collects both audio recordings and written text from native speakers of underrepresented languages. This crowdsourced approach enables researchers to build datasets for training machine translation models tailored to communities often overlooked by tech giants. The project combines student involvement with practical language preservation work, creating a two-way benefit: computer science students gain real-world experience in AI development while communities gain tools to document and share their languages.
The work responds to a documented crisis in linguistic diversity. Linguists estimate that one language disappears every two weeks globally, often without digital records. AI-powered translation tools could help stabilize endangered languages by making them more accessible to younger generations and facilitating cross-cultural communication.
BYU's approach differs from typical corporate translation development. Rather than extracting data from existing online sources, PathSay partners directly with speakers, ensuring more authentic and culturally appropriate language samples. This method also builds local capacity and maintains community control over linguistic data.
Challenges remain substantial. Creating functional translation systems requires thousands of hours of quality data. Scaling the PathSay model across thousands of languages demands sustained funding and coordinated international effort.