Advancing Language Revitalization: LLM-RBMT for Low/No-Resource Languages

  • Jared Coleman and Bhaskar Krishnamachari explore machine translation for low/no-resource languages.
  • Coleman’s background in Owens Valley Paiute drives research into language revitalization tools.
  • Their LLM-RBMT approach blends rule-based and natural language processing for accurate translations.
  • The method aids in preserving endangered languages by simplifying complex sentences.
  • Coleman’s digital tools like Kubishi support language revitalization efforts.
  • Presented at NAACL’s AmericasNLP workshop, highlighting LLMs’ role in linguistic preservation.

Main AI News:

In the latest breakthrough, Jared Coleman, recently awarded a Ph.D. in computer science, and his mentor Bhaskar Krishnamachari share a passion for languages—both human and computer-driven. Krishnamachari, raised in India with Tamil, Hindi, and English, expanded his linguistic horizons to include French and Mandarin in college. Coleman, a native English speaker, embraced Spanish in high school and mastered Portuguese with the help of his wife and college friends.

During the pandemic, Coleman delved into Owens Valley Paiute, a lesser-known language. As a member of the Big Pine Paiute Tribe, his ancestral ties to the language run deep—his father, David, hails from the tribe’s reservation in Big Pine, CA. Paiute, however, is categorized as a “no-resource language,” lacking publicly available translated sentences crucial for training machine learning models.

In their groundbreaking paper titled “LLM-Assisted Rule-Based Machine Translation for Low/No-Resource Languages,” Coleman and Krishnamachari introduce LLM-RBMT, a novel approach to aid in learning such languages. Co-authored by Khalil Iskarous, USC Dornsife associate professor of linguistics, and independent researcher Ruben Rosales, their method combines traditional rule-based translation tools with advanced natural language processing from large language models (LLMs).

Rather than directly translating Owens Valley Paiute, the LLM guides rule-based translators, ensuring accurate translations through its nuanced understanding of language intricacies. This approach, as Coleman explains, mimics the natural learning process by blending familiar and unfamiliar words, enhancing practical usability.

The tool autonomously manages much of the translation process with minimal input,” adds Krishnamachari, highlighting its adaptive capabilities.

Coleman’s contributions extend beyond academia; he spearheads Kubishi, a suite of digital tools for language revitalization, including an online dictionary and a translation system bolstered by his research.

Presented at NAACL’s AmericasNLP workshop, their findings underscore LLMs’ versatility in revitalizing endangered languages, marking a pivotal advancement in linguistic preservation.

Reflecting on his journey, Coleman acknowledges his tribe’s enduring efforts in language revitalization, viewing his research as a part of a broader initiative. As he prepares to join Loyola Marymount University as an assistant professor of computer science, Coleman sees this achievement as both personal and academic, honoring his familial legacy and paving the way for future linguistic exploration.


The development of LLM-RBMT by Jared Coleman and Bhaskar Krishnamachari marks a significant leap in linguistic preservation and technology integration. Their approach not only enhances the understanding and translation of low-resource languages but also sets a precedent for future advancements in the market of language learning and digital tools for cultural preservation. This innovation underscores the potential of AI-driven solutions to bridge cultural divides and support global efforts in linguistic diversity.