Speakers build syntactic structure so they can avoid producing ungrammatical sentences too frequently. But how speakers do so remains poorly understood. In this talk, I discuss our recent series of studies on how speakers encode structural representations, such as argument structures, filler-gap dependencies, and coreference relations, as they construct sentences. Building on experimental results, I introduce a preliminary model of structure building in speaking, which I argue can also be used to capture syntactic processes in comprehension. This model of a shared structure-building mechanism for both comprehension and production aims to contribute to the broader goal of understanding the relationship between syntactic knowledge and syntactic processes.