Measuring Embedding Sensitivity to Authorial Style in French: Comparing Literary Texts with Language Model Rewritings

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Research Methodology & Innovation · Depth: Expert, quick

Summary

A study investigated the capacity of language model embeddings to capture authorial writing styles and the retention of this stylistic information after large language model (LLM) rewriting. Focusing on French literary texts, researchers utilized a controlled dataset to quantify stylistic variation through changes in embedding dispersion. The findings demonstrate that embeddings reliably encode authorial stylistic features. Crucially, these stylistic signals persist even after texts undergo LLM rewriting, although they also exhibit patterns specific to the LLM used for the rewriting process. These analytical observations suggest new avenues for developing methods to detect authorship imitation, particularly as LLMs become more sophisticated in emulating human writing.

Key takeaway

For NLP Engineers developing authorship attribution or detection systems, this research confirms that language model embeddings retain crucial stylistic signals even after LLM rewriting. You should consider incorporating embedding dispersion analysis into your models to identify original authorial style and differentiate it from LLM-generated stylistic patterns. This approach could enhance the robustness of your imitation detection tools.

Key insights

Embeddings reliably capture authorial style in French texts, persisting even after LLM rewriting, with LLM-specific patterns.

Principles

• Embeddings encode authorial stylistic features.
• Stylistic signals persist post-LLM rewriting.
• LLM rewriting introduces specific stylistic patterns.

Method

Quantified stylistic variation in French literary texts by measuring changes in embedding dispersion using a controlled dataset.

In practice

• Develop authorship imitation detection.
• Analyze LLM-specific stylistic patterns.

Topics

• Language Model Embeddings
• Authorial Style
• Authorship Attribution
• LLM Rewriting
• French Literature
• Stylometry

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• I don't understand AI. How does it work? — LLMs predict text by mapping words to semantic vectors and learning statistical relationships from vast training data.
• Differentially-Private Text Rewriting reshapes Linguistic Style — DP text rewriting alters linguistic style, homogenizing discourse towards a non-involved register.
• How Far Can Classical NLP Go? From Bag-of-Words to Stacking on Spooky Author Identification — Classical NLP, with careful feature engineering and stacking, excels at stylistic authorship attribution.
• Solving the Embedding Step in NLP Projects — Effective embeddings are crucial for NLP model generalization and reliable downstream task performance.
• Self-Preference Is Weak or Absent in Verifiable Instruction-Following Revision: A Four-Model Test Under Genuine Authorship — LLMs show no detectable self-preference when revising their own text based on verifiable instruction-following.
• Which sentence is doing the most work in your favourite novel ? I tried to find out. — Machine learning can identify semantically distinctive sentences in texts, revealing structural or narrative pivots.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.