Artificial Intelligence

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.
To 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.
Seek to be truthful and accountable about the use and value of Artificial Intelligence.
Carefully research and test bias within models. Attempt to discover or fine-tune models toward doctrinal neutrality and alignment with historic orthodoxy.

Risk Management: 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.

The AI Task Force includes members from our Field Operations and Information Technology teams.

This page is a public place to host research papers and outputs of the Task Force.

Downloadable Files

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.

Introduction to Large Language Models

Handwritten Text Recognition with LLMs

Use of 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.

Biblical Sentiment Analysis and Emotion Detection 2024-11-26

Downloadable Files

About This Site

Find Us

Pages