Students and faculty at Grove City College are using AI to speed up and improve biblical translations to spread the Word in a polyglot world.
For the past few years, Computer Science professors Britton Wolfe and Brian Dickinson and their students have been working with SIL Global, a faith-based nonprofit that uses machine learning to develop language solutions to improve lives.
One might think that the best-selling book of all time had been translated in every language known to man, but that’s not the case.
“The Bible has been translated into most widely spoken languages and can reach most people, however, the full Bible is available in only about 10% of all languages. One of the primary applications of our work is completing the Bible translations in languages where it is partially complete,” Dickinson said.
High-quality translation tools don’t exist for hundreds of languages, impacting the ability of millions – particularly in Africa and Asia – to access not just the Bible, but digital tools necessary for communicating and participating fully in today’s world.
To aid SIL, student-faculty research conducted at Grove City College is focused on making algorithms that it uses to automate biblical translation work better and more efficiently. In addition to better translations, the coding updates can reduce demand on computing power, a key benefit to missionaries and others in the field who rely on personal laptops to do the work.
Dickinson said the work focuses on simplifying the conversion of language characters into common Roman (or Latin) script. “In our Romanization project, we are working to bridge the gap between state of the art AI translation models like NLLB-200 and languages with non-Latin scripts on which it may not perform as well,” Dickinson said.
The goal is to create completely or nearly-completely reversible romanizations of non-Latin texts, which should allow existing translation models to do a better job creating initial translations of the Old Testament in languages that already have a New Testament, he said.
The “big idea,” Dickinson said, is teaching a computer to translate based on a portion of the Biblical text that already exists in a non-Latin script. “Then the computer could make a rough translation of thus far untranslated portions of scripture. That would simplify the daunting task of translating the entire Old Testament into a task of correcting the computer's mistakes,” he said.
Better AI translation tools can reduce the number of mistakes the computer makes, helping to further reduce the burden on human translators.
While work on the Romanization project is ongoing, revised code developed at Grove City College to speed up word alignment was submitted to SIL for implementation this past summer. Dickinson said SIL was satisfied with the work and hoped to get it into their model soon.
Dr. Brian Dellinger, chair and professor of Computer Science, hailed the work of the professors and several former and current students who have worked on the project – Jonathan Allarassem’24, Allison Harnly, Alex Hemmerlin, Keith Graybill Jr., Hannah Proctor, and John Bauer.
“I think the work they’re doing is a great model of what we want our students to do: to be thoughtful about how their skills as computer scientists can contribute uniquely to the Kingdom of God,” he said.
For more about Computer Science at Grove City College, visit gcc.edu/csci.