Paolo Rosso
Building evaluation campaigns that work for Iberian-Romance languages — Spanish, Catalan, Portuguese — instead of porting English-centric methodology and accepting the resulting blind spots.
Rosso has spent two decades organizing evaluation campaigns from inside the Spanish-language NLP community: IberLEF for the Iberian languages, PAN for plagiarism and authorship problems across languages, and a long line of shared tasks on author profiling, irony detection, hate-speech analysis, and disinformation. The substantive value is that the tasks were designed with native-speaker sensitivity to what linguistic and cultural features actually carry signal — irony in Spanish is not the irony of English-language datasets, and hate-speech categories shift with the regulatory and cultural context. For ai100, which evaluates an engine's behavior in Spanish as one of its five language audits, this is the closest precedent for locale-native evaluation done with multi-year discipline.