Multichannel Convolutional Transformer and Intertextuality: A Latin Case Study

The detection of intertextuality is at the heart of many linguistic studies. A lot of efforts are currently underway to provide tools for analyzing relationships between authors. Most of them use standard statistics to compare textual data and find traces of text reuse from one author to another. The main objective of this work is to provide a new approach based on deep learning architectures and corpus-driven analysis. Building on previous contributions, we propose a hybrid architecture called multichannel convolutional transformer (MCT). Using this method, we develop a new tool for intertextuality detection based on authorship attribution. We have empirically demonstrated its efficiency using a Latin corpus. We conclude that our model can highlight complex linguistic patterns as features responsible for the classification decision. We consider these patterns as new categories of intertextuality traces, complementary to the existing ones.