Once, sentence processing research set aside prosody in order to focus on syntactic and semantic processing. Experimental sentences were mostly presented visually, often without prosodic markers such as commas. Now that we have made some progress by this ‘divide and conquer’ approach, and now that the technology for working on speech has improved, it may be time to integrate prosody into processing models. I argue here that we have no choice but to do so, because current evidence shows that even in silent reading, prosody is projected onto written sentences, and can influence the course of syntactic processing.