Finding Answers from the Word of God: Domain Adaptation for Neural Networks in Biblical Question Answering

TL;DR

This paper introduces BibleQA, a new dataset for biblical question answering, and explores neural network models with transfer learning, analyzing the effects of context length and translation choice on accuracy.

Contribution

It presents a novel dataset BibleQA and evaluates neural models with transfer learning for domain-specific biblical question answering.

Findings

Transfer learning improves model accuracy.

Shorter context lengths yield better results.

Modern Bible translations enhance model performance.

Abstract

Question answering (QA) has significantly benefitted from deep learning techniques in recent years. However, domain-specific QA remains a challenge due to the significant amount of data required to train a neural network. This paper studies the answer sentence selection task in the Bible domain and answer questions by selecting relevant verses from the Bible. For this purpose, we create a new dataset BibleQA based on bible trivia questions and propose three neural network models for our task. We pre-train our models on a large-scale QA dataset, SQuAD, and investigate the effect of transferring weights on model accuracy. Furthermore, we also measure the model accuracies with different answer context lengths and different Bible translations. We affirm that transfer learning has a noticeable improvement in the model accuracy. We achieve relatively good results with shorter context lengths, whereas longer context lengths decreased model accuracy. We also find that using a more modern Bible translation in the dataset has a positive effect on the task.