Fast and Accurate Quotation Attribution in Literary Texts

Researchers introduced a fast encoder-based “joint scoring” method for attributing quotes to speakers in literary texts.

The model resolves multiple quote-speaker links within a shared context window and reached 94.5% accuracy on the Project Dialogism Novel Corpus.

The team also released ModernBookNLP, a BookNLP fork that includes the new attribution system.

The approach delivers state-of-the-art performance with much lower compute cost than LLM-based systems, making large-scale quote attribution more practical.