Translation Technology: From Data to Drafting
Insights and progress from Catapult and Slingshot in 2025
Across ETEN, translation partners and vendors are advancing a range of Assisted Translation Technology (ATT) initiatives. These efforts leverage the latest technology and AI innovations to assist Bible translation projects, enabling faster, more scalable translation workflows.
Project Catapult and Project Slingshot are two examples, each playing a distinct role at different stages of the translation pipeline. The key difference lies in the availability of domain-specific Scripture data.
Slingshot builds on an existing foundation of high-quality, in-domain content, typically New Testament translations created by local Bible translation teams. This allows the team to move directly into model development using validated, context-rich material.
Catapult, by contrast, begins earlier in the pipeline, where training data must first be collected and prepared. This includes both structured data collection, such as linguistically informed sentence prompts designed to capture essential grammar and discourse patterns, and unstructured approaches, which draw from stories, conversations, or other community content. Once gathered, the data is normalized in collaboration with language experts to ensure consistency before model training begins.
In this Innovation Lab update, we’re excited to share recent progress from both projects and to highlight how God is using these initiatives to open new pathways for Scripture access.
Project Catapult
Project Catapult begins where no domain-specific Scripture data exists. It focuses on collecting and preparing language data to build the foundation needed for future AI model development.
Currently, 24 projects are active, all involving All Access Goal languages with data collection, and an additional 7 projects are in queue. Many of the language projects partner with translation vendors such as XRI to support their work. Notably, the number of projects in proofing or language model creation now surpasses those in data collection, indicating meaningful forward movement.
A key area of experimentation has been the use of Jesus Film transcripts as model training inputs. In a recent Suluh workshop led by SIL, five language teams edited or back-translated transcripts from the 1979 Jesus Film to improve sentence-level data. These transcripts offer real-world examples of language use in Scripture-centered storytelling and help sharpen the AI’s understanding of natural sentence flow. By refining seed sentences, the teams produce higher-quality data to train AI models supporting more accurate and natural-sounding drafts.
An encouraging Catapult success story, shared by Biblica’s Emerging Technology Translations team, comes from the Javanese translation team. Seeking to engage younger members of their church, they discovered during a recent training that Scripture Forge’s community check feature could be a powerful tool. By sharing review links through WhatsApp and surrounding their efforts with prayer, they successfully recruited over fifty young people to help review translation drafts, inspiring and empowering the next generation of Bible translators.
Along with SIL, Biblica also plays a key role in advancing Project Catapult. In several languages, including Alas, Gayo, Javanese, and Standard Malay, the team released completed books of the Bible onto single-language Bible apps. These were directed releases, making Scripture available to the communities for the first time. The team is also experimenting with the Lilt platform to support translation in Burmese, using suggestion-based drafting, and Pattani Malay, where the orthography is still being standardized.
Jeff Webster, Translation Consultant at the Seed Company, shared that a team working with a language group in Nepal successfully doubled their work pace by using AI-assisted drafting. After adapting the Gospel of Luke and training an AI model with it, the team edited drafts for the remaining Gospels, completing them with high-quality results.
The Innovation Lab and Project Catapult partners are now developing success metrics focused on:
Seed sentences – Ensuring training inputs are low in complexity but rich in key linguistic features.
Draft quality – Measuring acceptability, error rates, and the naturalness of AI-generated translations.
Overall acceleration – Using data to track time spent at each stage, enabling more efficient workflows.
The road ahead for Project Catapult involves overcoming barriers, scaling the impact of AI, and deepening collaboration with ETEN agencies and technology partners. Refining our approach and applying the latest innovations is laying the groundwork for a faster, more effective process that equips translation teams and serves communities around the world.
Project Slingshot
Project Slingshot builds on existing, in-domain Scripture content. It uses validated New Testament translations to train AI models that support faster, high-quality Old Testament drafting.
Scripture Forge is a key tool in this process, a collaborative platform that enables translators, consultants, and administrators to generate and refine drafts together.
Since launch, 454 translation projects have engaged with Project Slingshot onboarding teams or partners, including 130 onboarding requests so far in 2025. From February to April 2025, users submitted 455 self-service draft requests through Scripture Forge, covering 1,138 unique book drafts and 25,637 chapters. The most frequently drafted books were Genesis (47 projects), Exodus (42 projects), and Ruth (38 projects).
At a recent SIL workshop in Tanzania, 12 translation teams were onboarded, involving 27 participants, including translators, administrators, consultants, CiTs, advisors, and IT staff. This helped establish an ongoing support system for the region. The workshop also prepared three consultants from Uganda to support future onboarding, including a potential follow-up event this year.
Teams expressed a strong preference for a “both/and” approach, combining:
Drafts from 1:1 model pairings and from back translations mixed with reference texts
Partial book drafting, especially in Psalms, offering early engagement and manageable steps forward
One key barrier that emerged was the user interface (UI) language, which limited accessibility for some teams. Together, these learnings are critical for scaling AI drafting as they build on local expertise, adaptable tools, and shared ownership of the translation process.
The progress shared here is only possible because of the collaboration, innovation, and faithfulness of many. As Slingshot and Catapult continue to evolve, we invite you to stay connected, contribute feedback, and explore new ways to engage.
If you're interested in partnering, participating in testing, or learning more about upcoming opportunities, reach out to us at lab@eten.bible. You can also subscribe to our updates or visit lab.eten.bible for the latest resources and stories.