‘That’s what good tools do’

Left to right: Daniel Wilson of XRI Global, Jacob Bullock of SIL Global and Damian Daspit of SIL Global.

Artificial intelligence (AI) is dramatically accelerating early-stage Bible translation — especially in languages that have no written Scripture yet.

During the 2025 Missional AI Summit, a panel discussion focused on the growing partnership between SIL Global and XRI Global, a Clarksville, Tenn.-based company that develops AI-powered language technology especially for underserved language communities. The partnership is sponsored by Every Tribe Every Nation.

The conversation offered a glimpse into how AI is no longer just a futuristic concept but a practical tool already empowering translation. When SIL did a workshop last year with a translation team in Tanzania, the response surprised them.

‘We weren’t really sure, going in, what people knew about AI as a technology, how people felt about it’, said Jacob Bullock, NLP (Natural Language Processing) Developer for SIL. ‘The global church is not on-board 100 percent with the idea of using AI for ministry, for Bible translation.

‘These teams were so excited, because we welcomed them into the research. We emphasized day one: You are research partners with us. Your observations are going to steer the future adoption of this technology by other languages.’

‘Starter fertilizer’

Of the remaining languages with little or no translated Scripture, many have little to no written resources, plus inconsistent spelling or dialect norms. This lack of reliable data is a major challenge for AI, which depends on large datasets to learn how to generate text accurately.

‘Most AI models are great at English and a few other major languages’, said Daniel Wilson, founder and CEO of XRI Global. ‘But once you go beyond the top 30 or so languages, the amount of usable data online falls off a cliff.’

To solve this, XRI has been working with SIL to collect the highest-quality data possible in these under-resourced languages.

“We rapidly collect the ideal data set that is domain specific to the Bible,’  Wilson said. ‘We pass that over to our friends at SIL and Biblica and a lot of these other agencies to begin training and producing drafts for the community to check.’

The process starts with local speakers translating 8,000 carefully selected sentences using a special app. XRI then ‘cleans’ and verifies the responses to become a kind of ‘starter fertilizer’ for AI training. This curated dataset is specifically focused on Bible themes and tailored to each unique language context.

Then, with this quality data in hand, SIL's AI team uses a system they’ve developed called Slingshot. This tool can train a translation model for a specific language pair—in just two hours. In many cases, they start with a completed New Testament in a language and use it to help draft the Old Testament.

“This tool has already been used in over 400 Bible translation projects since launching in early 2024,” said Damian Daspit, senior software developer and computational linguist with SIL Global. “Many teams previously felt the Old Testament was too big a mountain to climb. But when we showed them a machine-generated draft of Genesis, their eyes lit up. They said, “We can do this.”’ 

Drafting, not finalising

The AI-generated text is not a finished translation. It’s a first draft—something for human translators to work from. That distinction is crucial.

‘The core of Bible translation is still human’, Bullock said. ‘The people are the heart of the project. AI is here to assist, not replace.’

AI-generated drafts may even strengthen collaboration within translation teams. In traditional workflows, one person drafts a passage, and others offer corrections—a dynamic that can be socially awkward or even tense in some cultures. But if the first draft comes from a computer, critique becomes easier because no one’s ego is on the line.

The process is not one-size-fits-all. Some teams prefer to edit the raw, unedited draft. Others want to compare multiple AI-generated drafts alongside source texts and commentaries. The diversity of use cases reflects what Bullock called the ‘snowflake factor’: no two translation projects are the same.

AI without Internet

Two challenges in many Bible translation contexts are a lack of reliable internet and security concerns. To solve this, XRI has developed a device called Truffle: a hand-held supercomputer that can run powerful translation models offline.

Wilson pulled the device out from under his chair to show the audience. The little white box can hold a full AI toolkit, including 70 billion parameters, or settings, that help it understand and generate human-like text. The Truffle can pair with phones, computers and monitors, all without connecting to the internet.

Daniel Wilson displays the Truffle, a hand-held supercomputer that can run powerful translation models offline.

The Power of Good Data

AI’s effectiveness depends heavily on the quality—not just quantity—of the training data. SIL’s team has found that a small amount of excellent data (like a thoroughly checked New Testament) can produce surprisingly useful drafts of other Scripture portions.

Still, getting enough good data to feed into AI models is a constant effort, and still very experimental—particularly when that data is not written.

‘We really would like to explore how we can transform oral data in these languages into a form that we can feed into the models’, Bullock said.’ And we’re making some strides forward in that. Other ministry materials produced, (like) JESUS film scripts. We have an opportunity to see how those play into the training data.’

And when spelling inconsistencies or dialect differences affect model performance, the teams innovate new benchmarks to better reflect real-life language use. “We’re not just building models; we’re building models that reflect people,” Daspit said.

Spirit-Guided and People-Centered

For those who might wonder whether AI belongs in Bible translation, the panel was clear: this is not a spiritual shortcut—it’s a practical tool. A question from the audience caught the panel’s attention: Can AI be guided by the Holy Spirit?

‘I think that ultimately God, the Holy Spirit, works through the church, through people, to translate the Bible,’ Daspit said. ‘What we’re trying to do with these models is build a model of the translation team. It’s a model of people. It’s a model of the decisions and the style, the way that the team would translate the Bible. … It’s assisting the translation team. But at the heart of the translation process and that draft, it’s the team. It’s the people in it. The technology is there to help.’

Bullock agreed, adding that rather than replacing a Spirit-led team, AI can ‘fertilize’ those efforts. ‘Very much the same way that word processors, when they came along, assisted the team in translation’, he said. ‘Just the same as some of the highly specialized exegetical resources assist the team.’

Time to Insight

AI may be exponentially faster, but it still works much the same as those earlier tools, Wilson said.

‘There are certain parts of translation that are absolute drudgery,’ he said. ‘A lot of the value created by AI is augmenting drudgery. And reducing the time to insight. That’s what good tools do.’

He compared it to using a commentary to get insight on a Scripture passage. Having that book available is faster than flying in a theologian.

‘Well, now with AI, you’re actually even getting a quicker insight, being able to ask questions directly or seeing three or four different ideas in your own language about how to translate that. That’s what AI is good at, reducing that time to insight. … The tools are progressively getting better and better and better for the tasks that we want to do.’

•••

Story: Jim Killam, Wycliffe Global Alliance

Translated with DeepL. How was the translation accuracy? Let us know at info@wycliffe.net

Video: Watch the entire presentation

Additional Mission al AI Summit videos:

• Tim Jore and Bruce Erickson from the ETEN Innovation Lab
• Ryder Wishart from the ETEN Innovation Lab
• Randall Tan from the ETEN Innovation Lab
• All Missional AI session videos

Alliance organisations are welcome to download and use images from this series.