Conference Agenda

This contribution proposes a multifaceted evaluation framework for assessing the performance of Large Language Models (LLMs) in encoding historical letters according to the TEI Guidelines, using the Joseph von Hammer-Purgstall correspondence edition as a case study.

We investigate the extent to which LLMs are able to learn a redefinition of a concept from literary studies, focalization, and apply it adequately to text examples. It shows that, with one exception, there are no statistically significant differences between the LLM output for prompts with and without the redefinition.

Aiming to study cross-cultural narrative patterns, this research develops a computational framework for extracting character features from English and Chinese fanfiction. By evaluating traditional Bert-based models and LLMs on tasks including character recognition, coreference resolution, dialogue and trait extraction, it provides insights into NLP tools' performance in characterization analysis.

We developed a system to automatically assign word sense tags to all content words in a substantial historical Japanese corpus, comprising over 20 million words. Our approach leverages a system based on Bidirectional Encoder Representations from Transformers (BERT), achieving an accuracy of 88.57%.

In this work, we propose a framework for extracting, annotating, attributing and visualizing dialogue characters in classic Chinese novels. We leverage interactive workflows to incorporate expert’s knowledge in the dialogue character extraction and attribution process.