AI Natural Language Processing--A Game Changer for Minority Language Translation

Automated Bible translation has been attempted by many people over the last nearly five decades. Early attempts used rule-based learning, and that had some positive results with closely related dialect adaptations. Google’s statistical approach to translation requires a very large corpus, usually millions of words.

AI natural language processing (NLP) is different. Neural networks have the ability to learn natural language in the way humans learn language. By using multi-layered learning and a limitless ability to connect neurons (language text) and vector those neurons to form weighted networks, NLP can learn to produce very good translations in a short span of time. And it can learn from a small team of people, even an individual, rather than requiring complex rules and a large language corpus. It is well-suited for minority languages that lack translation expertise and a large text corpus.

However, adapting NLP for minority language translation also challenges some long-held assumptions about the translation process. It means perhaps the translator is not the most important part of the process. Instead a community of end users skilled at processing AI generated translations and submitting quality feedback may be the more important role. The community will be central and a human as translator a temporary role.

Quality in AI generated translation may be realized through a very different process; one that is highly iterative and ongoing as NLP gets smarter and better. What other Bible translation roles will need to be rethought in the age of AI NLP? A look at some current attempts at AI NLP Bible translation projects may help us answer this question.