LATIDIA · Ciberseguridad

Auditoría de la equidad interlingüística en el modelo de lenguaje Marcas de agua

arXiv: 2608.20047v2Tipo de anuncio: replace-cross Resumen: Los esquemas de marca de agua para la salida del modelo de lenguaje grande se evalúan casi exclusivamente en texto en inglés utilizando el umbral de detección de cada esquema y un conjunto estrecho de

La noticia

arXiv:2608.20047v2 Announce Type: replace-cross Abstract: Watermarking schemes for large language model output are evaluated almost exclusively on English text using each scheme's detection threshold and a narrow set of quality measurements. Multilingual deployment exposes evaluation-design choices that are inconsequential on English but determine conclusions cross-lingually. We propose an evaluation framework with four components: detection thresholds calibrated empirically per deployment context, a threshold-independent companion measurement that distinguishes calibration failures from detection failures, three disjoint quality measurement paradigms (distributional, paired-semantic, and reference-perplexity), and a generalized-entropy decomposition of cross-language disparity over a typological family partition. Applied to six watermarking schemes, three open-weight generators, eleven languages spanning four scripts and eight typological families, and