Evaluating LLMs for Quotation Attribution in Literary Texts: A Case Study of LLaMa3

TL;DR

This paper evaluates Llama-3's ability to attribute quotations to speakers in novels, demonstrating it surpasses previous models and establishing a new state-of-the-art in literary quotation attribution.

Contribution

It is the first comprehensive evaluation of Llama-3 for quotation attribution in literature, showing its superior performance over prior models and validating that memorization does not account for the results.

Findings

Llama-3 outperforms ChatGPT and baselines in quotation attribution.

Memorization does not explain the performance gains.

Llama-3 achieves state-of-the-art results in English literary quotation attribution.

Abstract

Large Language Models (LLMs) have shown promising results in a variety of literary tasks, often using complex memorized details of narration and fictional characters. In this work, we evaluate the ability of Llama-3 at attributing utterances of direct-speech to their speaker in novels. The LLM shows impressive results on a corpus of 28 novels, surpassing published results with ChatGPT and encoder-based baselines by a large margin. We then validate these results by assessing the impact of book memorization and annotation contamination. We found that these types of memorization do not explain the large performance gain, making Llama-3 the new state-of-the-art for quotation attribution in English literature. We release publicly our code and data.