Conférence plénière : Christophe Coupé (University of Hong Kong -- 25 septembre 2025)

Value of the Information & Strength of the Argument

The talk provides valuable insights into the application of LLMs to literary analysis, offering a clear overview of existing quantitative methods and their limitations. Coupé argues convincingly that LLMs can combine the scalability of distant reading with the interpretive depth of close reading, though he acknowledges the exploratory nature of his work. The argumentation is well-structured, moving from general narrative theory to specific computational techniques, and is supported by concrete examples (e.g., sentiment curves for Harry Potter). However, the talk lacks detailed empirical evidence or case studies, and some claims about LLM capabilities are presented without rigorous validation.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates scientific rigor by referencing established works (Campbell, Gutenberg, LIWC) and a 2016 study on emotional arcs, but it does not provide formal citations or links in the description. The title accurately reflects the content, and the speaker’s expertise lends credibility. The absence of detailed source documentation slightly weakens the overall rigor, but the methodological transparency and cautious tone enhance trustworthiness.

The title accurately reflects the content: a plenary lecture by Christophe Coupé at a conference on linguistics, translation, and digital humanities.

7/10

The speaker is a recognized researcher in computational linguistics and cognitive science, presenting exploratory work with appropriate caveats. The talk references established methods and studies (e.g., Hero's Journey, sentiment analysis on Gutenberg corpus) but lacks detailed citations or peer-reviewed sources in the description. The content is methodologically sound but presented as work in progress.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to the talk's focus on quantitative approaches to literary structure.
• 05:00Discussion of Hero's Journey and its universality in narratives.
• 10:00Explanation of multi-scale narrative structures and analogy with electricity consumption.
• 15:00Introduction to distant reading and computational analysis of literary works.
• 20:00Example of character network extraction and its use in author comparison.
• 25:00Sentiment analysis and emotional arcs, referencing the 2016 study on six story shapes.
• 30:00Limitations of dictionary-based and rule-based sentiment analysis.
• 35:00Introduction to large language models and their potential for narrative analysis.
• 40:00Proposal of a pipeline to measure protagonist activity/passivity using LLMs.
• 45:00Discussion of other dimensions (cognitive engagement, social relations) and future directions.

• The Hero with a Thousand Faces — Mentioned as the source of the Hero's Journey concept.
• Project Gutenberg — Mentioned as a source for public domain texts.
• LIWC (Linguistic Inquiry and Word Count) — Mentioned as a tool for text analysis.
• Study on emotional arcs (2016) — Referenced for the six story shapes.

• The Hero with a Thousand Faces — Supports the universality of narrative structures.
• Project Gutenberg — Provides the corpus for large-scale analysis.

• None — No discordant sources mentioned in the talk.