According to expressivism, sentences of the form ”Φ-ing is good (bad)” express an attitude towards Φ-ing rather than a truth-evaluable proposition. Expressivism faces embedding problems: When embedded in ”If Φ-ing is good (bad), then I will tell my kids (not) to Φ”,  ”Φ-ing is good (bad)” does not express an attitude towards Φ-ing. Embedding problems also seem to arise in non-normative discourse, for example when sentences containing indexicals are embedded under modal or temporal operators. Drawing on insights from Dummett (1973) and Lewis (1980), philosophers have recently argued that such problems require a distinction between the assertoric content of a sentence and its semantic value (its role in a compositional semantics) (Rabern 2012, Ninan 2013, Yalcin 2014). This talk explores the similarities and differences between embedding problems and suggested solutions to them in normative and non-normative discourse.