Intelligent Decision Support for Article Translation in Multilingual Newsrooms

Zhan Liu; Adrien Bertaud; Anne Darbellay; Nicole Glassey Balet

doi:10.1145/3748522.3779719

What is it about?

In multilingual countries like Switzerland, news editors must decide every day which German-language stories are worth translating into French for readers in other regions. These choices are often made quickly, based on gut feeling. We built an AI tool that helps editors make these decisions. It reads each article and predicts whether it should be translated, while explaining why — for example, because the story mentions a French-speaking region, a well-known person, or has strong emotional appeal. Trained on nearly 16,000 real news articles, the tool was correct about 85% of the time. Importantly, editors stay in full control: the system only offers suggestions and learns from their feedback over time. The goal is not to replace journalists, but to support them and help news flow more easily across language communities.

Photo by Markus Winkler on Unsplash

Why is it important?

Most AI tools for newsrooms are "black boxes" that just push recommendations without explaining themselves, and they usually work in only one language. Our approach is different in two ways. First, it is built specifically for multilingual decisions, deciding what to translate across language communities, a problem that has been largely overlooked. Second, it is transparent and keeps editors in charge: it explains the reasons behind each suggestion and learns from editor feedback, so the tool adapts to how a real newsroom actually works. As AI becomes part of journalism, this kind of human-centered, explainable system shows how the technology can support editors rather than replace them, and help news reach audiences across language barriers, strengthening informed public debate in multilingual societies.

Perspectives

What I found most rewarding about this project was working directly with newsroom editors rather than building a tool in isolation. Their honest feedback, including their skepticism about AI, shaped the system far more than any algorithm did. It reminded me that the goal of AI in journalism should not be efficiency for its own sake, but supporting the human judgment that gives news its value.
Zhan Liu
HES-SO Valais-Wallis

This page is a summary of: Intelligent Decision Support for Article Translation in Multilingual Newsrooms, March 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3748522.3779719.
You can read the full text:

Read

Resources

Open Access version
Intelligent Decision Support for Article Translation in Multilingual Newsrooms
We present an intelligent decision support system, powered by AI-driven prediction, designed to assist multilingual newsrooms in selecting articles for cross-regional translation as part of the digital transformation of journalism. The task is to predict whether a German-language article should be translated into French for cross-regional publication. Trained on 15,933 German-language articles from a major Swiss publisher, the system uses a hybrid architecture combining multilingual BERT embeddings with 41 engineered features and incorporates real-time editorial feedback through an active learning loop. It achieves an accuracy of 85.0%. During deployment, F1 improved from 77.5% to 81.2% after four weeks of feedback-driven exemplar refresh. Ablation studies indicate that sentiment polarity, regional relevance, and person-type named entities are the most influential features. The interface highlights key factors, ensuring transparency and consistency with editorial practice. By pairing hybrid NLP with human-in-the-loop prompting, the approach operationalizes intelligent translation triage in a live newsroom while preserving human control over final decisions.

Contributors

The following have contributed to this page

Zhan Liu
HES-SO Valais-Wallis

How AI helps newsrooms decide which articles to translate

What is it about?

Why is it important?

Perspectives

Resources