The goal of this article is to analyze the semantic contribution of evaluative adverbs (EAs) such as unfortunately in several languages of the Romance family, namely French, Catalan, and Spanish. Following Bonami and Godard (2008), we propose to analyze EAs as items that convey projective meaning in order to explain their peculiar semantic behavior (they cannot be directly denied, do not change the truth conditions of the proposition they evaluate, and are not factive) and their unacceptability in negative assertions. Unlike what has been claimed for many other languages, French allows EAs in questions, and we show that Catalan and Spanish do too, as long as some conditions are met. We propose an account that derives their interpretation in both assertions and questions: integrated French EAs take the proposition to their right, and if they appear in a WH-question, their interpretation is similar to that of unconditionals. In contrast, nonintegrated EAs in Catalan and Spanish have scope over a set of propositions, and are acceptable in questions only if the speaker is biased toward one of the propositions in the set denoted by the question. The acceptability of EAs in such questions, rejected by previous literature, is confirmed by an experimental study.