Artificial Intelligence
Artificial Intelligence has the potential to greatly enhance organizational processes. Wycliffe Associates is carefully investigating how AI can assist mother-tongue translators in their Bible translation work, and they anticipate significant developments in the near future.
Overview
Wycliffe Associates is carefully and thoughtfully investigating ways to use Artificial Intelligence to enhance Church-Owned Bible Translation. To that end, WA has created an AI Task Force to lead this work.
Purposes
- Fully and responsibly explore the potential impact of Artificial Intelligence for Wycliffe Associates and Church-Owned Bible Translation.
- Consider the ethics and viability of current commercial and open-source AI, Machine Learning, and Natural Language Processing models.
- Faithfully take risks in the name of novel approaches to Bible Translation, with the desire to see massive improvement in translation quality and consistency.
Ethical Guidelines
As a Christian ministry, we desire to operate legally and ethically, while respecting data privacy by considering the source of training data. We seek to be truthful and accountable about the use and value of Artificial Intelligence. We will carefully research and test bias within models. In addition, we will attempt to discover or fine-tune models toward doctrinal neutrality and alignment with historic orthodoxy.
Risk Management
Our task force will consider potential risks associated with the use of AI, including technological, reputational, and operational risks, and propose strategies for mitigating these risks.
Additional Reading
Introduction to Large Language Models
This whitepaper is a brief introduction to using a type of machine learning called Large Language Models (LLMs). It is not intended to be comprehensive and does not represent a recommendation that readers deploy commercial versions of this technology in Bible Translation. It is intended for purely educational purposes.
Prepared by Dr Travis Wright for the AI Task Force. Version date: 8.16.24.
Handwritten Text Recognition with LLMs
Results of preliminary Handwritten Text Recognition (HTR) research for potential use by Bible Translators.
Prepared by Joe Sarabia and Travis Wright for Wycliffe Associates. Version date: 11.26.24.
Using Sentiment Analysis for BT
This study investigates the efficacy of fine-tuned BERT models in detecting emotional content within Biblical texts, focusing on narratives such as the Prodigal Son and Psalms. The research evaluates the ability of state-of-the-art emotion detection systems to capture nuanced emotional dimensions, including relief and joy, across a fixed corpus with controlled variables. The results highlight substantial limitations in the models' accuracy and consistency, underscoring challenges in applying emotion detection to specialized and high-context domains like Bible translation. These findings contribute to the ongoing discourse on the applicability of natural language processing techniques in complex editorial tasks.
Written by Dr. Drew Curley, Joe Sarabia, Thomas Hajny, et. al.
Introduction to RAG
The paper provides an introduction to Retrieval-Augmented Generation (RAG), an AI approach that enhances language model responses by retrieving relevant external information, making it particularly useful for providing domain-specific or up-to-date knowledge. It explores the potential of RAG in religious contexts, highlighting its ability to generate more accurate and trustworthy responses by drawing from authoritative theological sources, thus addressing concerns about bias in AI-generated content. The study also describes an experiment using multilingual biblical texts and vector databases to improve retrieval efficiency, concluding that while RAG enhances response quality, its probabilistic nature limits its reliability, making it useful only in contexts where perfect accuracy is not required.
A Bible in Every Language
Explore our growing library of Bible translations and resources. If you don’t see what you’re looking for, or if you want other kinds of information, please contact us here.