Community Intelligence Meets AI: Can Local Literacy Programs Improve AI-Generated Scripture?

Artificial Intelligence is increasingly portrayed as a tool for drafting Scripture more quickly and with great accuracy, although this depends on the quality of training data. Nevertheless, AI relies on a substantial existing corpus to learn the patterns needed to generate new text. This is especially challenging in minority language contexts where existing literature may be nonexistent or limited to a handful of translated materials that may not represent natural text across a variety of genres. Meanwhile, the focus on translating Scripture into as many languages as possible as quickly as possible has, in our experience, often eclipsed discussion of the role of Scripture engagement.

We contend that community-based literacy programs can provide the “Community Intelligence” to inform AI models, giving community members valuable practice in reading their language as well as developing the editorial skills needed to process AI-generated drafts.

To test this hypothesis, we will use a machine translation tool to produce AI-generated drafts of several Old and New Testament books in the Bisu language of northern Thailand, using different corpora:
1) Materials created by local authors for the literacy program (including big books, small books, folktales, and cultural stories)
2) The Bisu New Testament (2015):
a. Some books of the Gospels
b. All New Testament books
3) Combination of:
a. 1) and 2a)
b. 1) and 2b)

The resultant drafts will be evaluated by an experienced MTT with over 15 years of Bible translation experience and a handful of Bisu believers who have only recently begun learning to read their language. Translation quality assessment tools will be used to examine how community materials influence AI drafts. This presentation will report on our findings.