선교 ㆍ사역 소식

제목Fluent Bullshit or Bible Translation of High-quality2024-05-31 10:09
작성자

Fluent bullshit or Bible translation of high-quality0

Bible translation is a process in which missionaries live among the language group, learn their language and culture, train local people to translate the Bible verse by verse, and together they are transformed by the Word.[1] It still takes an average of more than 20 years to translate the New Testament and at least 30 years to translate the entire Old Testament. What would happen if AI were to be utilized in this process? The obvious answer is that it could be done "faster" and at a lower cost.[2] Furthermore, according to experts, it could produce much more accurate and higher-quality translations.[3]

Recently, SIL developed a Bible translation app/program called Scripture Forge using NLLB (no language left behind) technology based on Facebook's Meta AI. (Facebook is working on a project to translate 200 minority languages to solve the problem of minority language users being unable to access much useful information online due to language barriers. The feature of this is that AI can learn the language well and produce translation results even with a small amount of language data, unlike major languages in the world with many users.)[4]

Scripture Forge, which works in conjunction with Paratext, an existing Bible translation program, can create a preliminary translation of each book of the Bible in a language for which at least one gospel has been translated, using it as data. However, SF was developed to assist in the translation-checking process. It is said to be much better than humans at checking translations within a team, checking with the community, and maintaining consistency in the translation, even back-translating from the target language to the common language for checking by Bible translation consultants.[5]

That said, AI still makes a lot of errors that need to be caught and corrected by humans. In one video I watched, the AI would say something that sounded plausible or have a conversation but then say or answer something completely wrong, which the presenter amusingly referred to as "fluent bullshit."[6] These errors are more prevalent when the AI communicates in languages with more data sources. In contrast, when there is relatively little data, local workers can train the AI well and get more accurate answers.

On the other hand, there are concerns about this process. The incarnational identification and partnership of learning the language and culture of the target people and sharing life with them is weakened, and the realizations, professions of faith, and life changes/transformations that occur as locals wrestle with the Word verse by verse can be weakened or bypassed. Granted, the role of the local translator will be more focused on checking rather than initial translation, but the process of wrestling with the Word is still there, and the time saved on initial translation can be better spent on learning, mastering, and applying the Bible correctly.[7]

But as with any new technology, there will always be unexpected side effects, so we must continue our walk of obedience, praying that we can use the opportunities and tools God has given us for good while minimizing the negative consequences.[8]

 


0  For a good summary of the active use of AI in Bible translation, see the following article by Don Barger. How AI Assists in Global Bible Translation, by Don Barger (IMB missionary)

[1] The following article is a good overview of Bible translation and the conversions, transformations, church planting, and localized movement spread that occur in the process. This will also help you understand footnote seven below.  A missiology of progress: Assessing advancement in the Bible translation movement, by Kirk J. Franklin.

[2] In his presentation, Gravelle of the Seed Company, a specialized funding organization for Bible translations, said the New Testament translation will take about five years and is expected to cost about $500,000.  AI Natural Language Processing: A Game Changer for Bible Translation, by Gilles Gravelle.

[3] I interviewed Dr. J, who has completed translating the New Testament in a sensitive country and is now working with local workers to translate the Old Testament. He said that he has used Scripture Forge to produce the drafts of the Bible books in several languages, including his own, and the results have been rated as 70-80% accurate by trained local Bible translators. Furthermore, as explained below, especially if done well at the beginning, it can help maintain consistency in the translation, which is very useful for translating the massive amount of the Old Testament.

[4] Programs like Avodah, from Gravelle's presentation, are using the NLLB open source to provide interpretation services for people who don't have access to healthcare due to language barriers. Applied to Bible translation, the AI can be trained on about 1,200 sentences, including Bible verses, and can produce a preliminary translation.

[5] According to Dr. J, when translating the entire Bible, the local translators change several times, and the volume is so large that it is difficult for humans to maintain consistency in the wording of certain words, phrases, etc. However, AI excels in this area. Therefore, AI can maintain much more consistent and quality translations, if once good data is accumulated through quality translations and checks by well-trained local translators in the early stages, and the AI learns well.

[6] In his talk, Slow Road to Artificial Intelligence, Dr. Gary Marcus, an expert in psychology and neuroscience, began by introducing false and incomplete stories about AI, and argued that we need to be more cautious about using AI for common and easy tasks, while there is still a lot of work to be done on uncommon and vital tasks.

[7] For presentations and articles that argue for more caution in using AI, see the following links.  Lost in Translation: Navigating the Intersection of AI and the Sanctity of Drafting in BT, by Larry Hayashi, Reinier De Blois; Paul Unger, Matt Merritt.

[8] When it comes to our attitude toward new technologies, Jon says the general principle of "understand, pray, and obey" applies. See the following presentation.  Obedience in this AI Moment, by Jon Hirst from SIL International