Measuring Embedding Sensitivity to Authorial Style in French: Comparing Literary Texts with Language Model Rewritings
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916000868335616 |
|---|---|
| author | Icard, Benjamin Sainero, Lila Breton, Alice Zve, Evangelia Ganascia, Jean-Gabriel |
| author_facet | Icard, Benjamin Sainero, Lila Breton, Alice Zve, Evangelia Ganascia, Jean-Gabriel |
| contents | Large language models (LLMs) can convincingly imitate human writing styles, yet it remains unclear how much stylistic information is encoded in embeddings from any language model and retained after LLM rewriting. We investigate these questions in French, using a controlled literary dataset to quantify the effect of stylistic variation via changes in embedding dispersion. We observe that embeddings reliably capture authorial stylistic features and that these signals persist after rewriting, while also exhibiting LLM-specific patterns. These analytical results offer promising directions for authorship imitation detection in the era of language models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_10606 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Measuring Embedding Sensitivity to Authorial Style in French: Comparing Literary Texts with Language Model Rewritings Icard, Benjamin Sainero, Lila Breton, Alice Zve, Evangelia Ganascia, Jean-Gabriel Computation and Language Artificial Intelligence Large language models (LLMs) can convincingly imitate human writing styles, yet it remains unclear how much stylistic information is encoded in embeddings from any language model and retained after LLM rewriting. We investigate these questions in French, using a controlled literary dataset to quantify the effect of stylistic variation via changes in embedding dispersion. We observe that embeddings reliably capture authorial stylistic features and that these signals persist after rewriting, while also exhibiting LLM-specific patterns. These analytical results offer promising directions for authorship imitation detection in the era of language models. |
| title | Measuring Embedding Sensitivity to Authorial Style in French: Comparing Literary Texts with Language Model Rewritings |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2605.10606 |