<p>Fragmentation is a widely discussed thesis on the architecture of mental content, saying, roughly, that the content of an agent’s belief state is best understood as a set of information islands that are individually coherent and logically closed, but need not be jointly coherent and logically closed, nor uniformly accessible for guiding the agent’s actions across different deliberative contexts. Expressivism is a widely discussed thesis on the mental states conventionally expressed by certain categories of declarative discourse, saying, roughly, that prominent forms of declarative utterance should be taken to express something other than the speaker’s outright acceptance of a representational content. In this paper, I argue that specific versions of these views—Topical Fragmentation and Semantic Expressivism—present a mutually beneficial combination. In particular, I argue that combining Topical Fragmentation with Semantic Expressivism fortifies the former against (what I call) the Connective Problem, a pressing objection that lays low more familiar forms of Fragmentation. This motivates a novel semantic framework: Fragmented Semantic Expressivism, a bilateral state-based system that (i) prioritizes fragmentationist acceptance conditions over truth conditions, (ii) treats representational content as hyperintensional, and (iii) gives expressivistic acceptance conditions for the standard connectives. Finally, we discuss the distinctive advantages of this system in answering the problem of logical omniscience and Karttunen’s problem for epistemic ‘must’.</p>