Explainability in deliberation-support NLP is usually evaluated through post-hoc rationales or model-internal attribution methods, and only rarely against explicit institutional reasoning procedures. We introduce , a Romanian corpus of professional fact-checking reports that preserves the workflow of editorial epistemic arbitration, namely claim articulation, contextualisation, verification scope, evidence-based verification narrative, and calibrated conclusion. contains 789 raw reports from factual.ro and a processed benchmark release of 788 instances after removal of a singleton non-standard verdict label. Beyond six-way verdict prediction, we position as a benchmark for LLM explanation alignment, where models must generate short explanations that can be compared directly to gold institutional reasoning. We evaluate primarily with instruction-tuned LLMs, reporting full-corpus experiments for open-weight models and a matched pilot comparison with GPT-4 Turbo. The resulting evidence shows that verdict prediction and institutional explanation alignment are not the same capability: models that improve verdict accuracy do not necessarily preserve institutional calibration or produce explanations that align with professional verification narratives. These results support the central claim of the paper, namely that measures not only whether a model reaches a verdict, but also whether it does so in a manner that resembles documented public reasoning.

This paper introduces CoRoSeOf, a large corpus of Romanian social media manually annotated for sexist and offensive language. We describe the annotation process of the corpus, provide initial analyses, and baseline classification results for sexism detection on this data set. The resulting corpus contains 39 245 tweets, annotated by multiple annotators (with an agreement rate of Fleiss’κ= 0.45), following the sexist label set of a recent study. The automatic sexism detection yields scores similar to some of the earlier studies (macro averaged F1 score of 83.07% on binary classification task). We release the corpus with a permissive license.

Commonly occurring in settings such as social media platforms, code-mixed content makes the task of identifying sentiment notably more challenging and complex due to the lack of structure and noise present in the data. SemEval-2020 Task 9, SentiMix, was organized with the purpose of detecting the sentiment of a given code-mixed tweet comprising Hindi and English. We tackled this task by comparing the performance of a system, TueMix - a logistic regression algorithm trained with three feature components: TF-IDF n-grams, monolingual sentiment lexicons, and surface features - with a neural network approach. Our results showed that TueMix outperformed the neural network approach and yielded a weighted F1-score of 0.685.