Year: 2025
Type: article
Abstract: In the Christian tradition, fourteen letters of the New Testament have been attributed to the Apostle Paul. However, for seven of these letters—1 and 2 Timothy, Titus, Ephesians, Colossians, 2 Thessalonians, and Hebrews —the attribution to Paul has been the subject of scholarly debate. This study aims to develop a bidirectional long short-term memory (BiLSTM) network to classify chunks of these disputed letters, each approximately 100 words long, as either authored by Paul or not. Two model variants—a plaintext variant and a lemmatized text variant—were trained on undisputed Pauline letters and ‘impostor letters’ which serve as negative examples of Paul’s writing. The plaintext variant achieved 84% accuracy and the lemmatized text variant 83% accuracy. Both variants classify the majority of text chunks from Colossians and 2 Thessalonians as Pauline and the majority of chunks from Hebrews and 1 Timothy as non-Pauline, although caution is warranted in drawing strong conclusions from these results. For the remaining disputed Pauline letters—Titus, 2 Timothy, and Ephesians—the majority classification varies between the model variants, further emphasizing the need for caution. Nevertheless, this study introduces a deep learning approach to the authorship verification problem of the disputed Pauline letters, potentially serving as a model for future research.
Source: HIPHIL Novum
Authors: Rianne De Heide, Evy Beijen
Institutions: University of Twente, Vrije Universiteit Amsterdam
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: Vrije Universiteit Amsterdam
OpenAlex ID: W4411079952