Authorship Attribution in Multilingual Machine-Generated Texts

2025-08-03 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

A new study introduces the challenge of Multilingual Authorship Attribution (AA) for texts generated by Large Language Models (LLMs), moving beyond traditional binary classification to identify specific LLM generators or human authors across diverse languages. Current AA efforts are largely confined to monolingual settings, primarily English. Researchers investigated the cross-lingual transferability of existing monolingual AA methods across 18 languages, encompassing various language families and writing scripts, and 8 generators (7 LLMs and the human-authored class). The findings, accepted at ACL 2026, indicate that while some monolingual AA techniques can be adapted for multilingual use, substantial limitations persist, particularly in transferring attribution capabilities across different language families. This underscores the inherent complexity of multilingual AA and the urgent need for more robust methodologies to address real-world scenarios effectively.

Key takeaway

For NLP Engineers developing MGT detection systems, recognize that current authorship attribution methods struggle significantly in multilingual contexts. If your applications involve diverse languages, you must move beyond monolingual approaches. Prioritize developing or integrating robust, truly multilingual AA models. Adapting existing methods offers limited success and may lead to inaccurate generator identification across varied language families.

Key insights

Multilingual authorship attribution for LLM-generated text is complex, with current monolingual methods showing limited cross-lingual transferability across diverse language families.

Principles

• LLM fluency complicates MGT detection.
• Fine-grained AA is needed for LLM diversity.
• Cross-lingual transfer remains challenging.

Method

The study investigated monolingual AA method suitability for multilingual settings by testing cross-lingual transferability and generator impact across 18 languages and 8 generators (7 LLMs + human).

In practice

• Adapt monolingual AA for new languages.
• Test AA methods across language families.
• Develop robust multilingual AA models.

Topics

• Authorship Attribution
• Machine-Generated Text
• Large Language Models
• Multilingual NLP
• Cross-Lingual Transfer
• ACL 2026

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Cross-Lingual Empirical Evaluation of Large Language Models for Arabic Medical Tasks — English-centric LLMs exhibit significant performance gaps and reliability issues in Arabic medical tasks.
• Assessing Capabilities of Large Language Models in Social Media Analytics: A Multi-task Quest — Modern LLMs demonstrate varied capabilities across social media analytics tasks, with GPT-4 excelling in authorship verification and Gemini in attribute inference.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• A Dataset for Probing Translationese Preferences in English-to-Swedish Translation — Language models often prefer literal "translationese" over idiomatic phrasing, especially when exposed to the source language.
• Bridging Cultural Gaps in Automated Translation of Brazilian Expressions: A Study on Cultural Adaptation — Automated translation systems struggle with cultural adaptation, even with advanced LLMs, requiring human oversight.
• Understanding LLM Behavior in Multi-Target Cross-Lingual Summarization — LLMs perform cross-lingual summarization and translation jointly in later layers, not sequentially.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.