EU legislation is equally authentic and legally binding in all 24 official languages, rendering high-quality translation a legal obligation rather than a mere choice. Therefore, high-quality language technology supporting translation processes in all EU languages are essential for language professionals at the European Parliament (EP). This paper investigates whether domain-specific fine-tuning of an open-weight Large Language Model (LLM) yields consistently larger quality gains on legislative text compared to generic text, in all 23 EU target languages from English. We evaluate ten experimental conditions: base model, in-domain and cross-domain fine-tuning, sequential generic-then-legislative fine-tuning, and zero-shot Claude Sonnet 4.6 as a proprietary reference. We analyse BLEU, chrF, TER, and COMET metrics on nearly 700,000 segments. Results confirm the hypothesis for all 23 languages: legislative fine-tuning enhances BLEU by +12.30 compared to +7.10 for generic fine-tuning, demonstrating a consistent advantage of +5.20 BLEU in all the metrics. The fine-tuned EuroLLM-22B decisively outperforms Claude Sonnet 4.6, Anthropic’s latest frontier model, on both domains, highlighting that targeted adaptation of a smaller open-weight model can surpass a state-of-the-art proprietary system. Cross-domain transfer within institutional domain is positive for all languages, with no catastrophic forgetting. Low-resource languages such as Irish and Maltese benefit the most from fine-tuning, while a divergence between BLEU and COMET rankings for some languages underlines the need of evaluation metrics alongside traditional measures.

The work of Emergency Management (EM) agencies requires timely collection of relevant data to inform decision-making for operations and public communication before, during, and after a disaster. However, the limited human resources available to deploy for field data collection is a persistent problem for EM agencies. Thus, over the last decade, many of these agencies have started leveraging social media as a supplemental data source and a new venue to engage with the public. Such uses present both opportunities and challenges. While prior research has analyzed the potential benefits and attitudes of practitioners and the public when leveraging social media during disasters, a gap exists in the critical analysis of the actual practices and uses of social media among EM agencies, across both geographical regions and phases of the EM lifecycle - typically mitigation, preparedness, response, and recovery. In this paper, we conduct a mixed-method analysis to update and fill this gap on how EM practitioners in the U.S. and Europe use social media, building on a survey study of about 150 professionals and a follow-up interview study with 11 participants. The results indicate that using social media is no longer a non-traditional practice in operational and informational processes for the decision-making of EM agencies working at both the local level (e.g., county or town) and non-local level (e.g., state/province, federal/national) for emergency management. Especially, the practitioners affiliated with agencies working at the local level have a very high perceived value of social media for situational awareness (e.g., analyzing disaster extent and impact) and public communication (e.g., disseminating timely information and correcting errors in crisis coverage). Further, practitioners now engage with the public during the preparedness phase to mobilize them during the response phase. We present a model to understand the current practices of communication between agencies and the public, as well as among practitioners while leveraging social media. We also discuss novel challenges, including public fragmentation caused by the increasing use of multiple social media platforms, information integrity, and social listening expectations. We conclude with the policy, technological, and socio-technical needs to design future social media analytics systems to support the work of EM agencies in such communication.

In the immediate aftermath of natural or man-made disasters, social media plays an essential role in assessing the impact of the event. The images from social media demonstrated the potential to accelerate the response to a crisis. However, finding the exact location of relevant social media images remains a problem for both humans and computer systems.

Before urban flooding actually happens, weather forecasts with varying degrees of precision are available to emergency managers. In the aftermath of the event, authoritative information including Earth Observation (EO) data can be used to estimate precisely the flood extent, possibly after several hours. This study aims to determine how social media information can reduce the inherent uncertainty of the information in the immediate aftermath of an urban flood event. Specifically, the study investigates how to collect relevant social media images and to interpolate such data in order to create a map. The premise of the study is that social media platforms, when combined with digital surface models, can provide control points for creating a reliable near real-time estimate of the flood extent. In the study, we compared a flood extent map derived from social media with that derived from authoritative altimetry data during one of the worst floods to hit Venice, which occurred in November 2019. The results of the experiments show a good overall accuracy using several digital surface models. Given the global coverage of such models and the low resources required, we think the methodology proposed could be beneficial for emergency managers. Specifically, we describe how a flood extent map can be made available within 24 h, or even less, after urban flooding strikes a densely inhabited area, where data generated by the public are available.