Segmentation is a cornerstone of language processing across levels of linguistic analysis, and yet, standard models of linguistic memory leave the role of higher-order segments in online comprehension understudied. This dissertation advances the Context-Sensitive Encoding (CSE) hypothesis: that implicit prosodic boundaries (Bader, 1998; Fodor, 1998, 2002) serve to partition sentences into distinct encoding contexts via a temporal context mechanism (Howard & Kahana, 2002) that shifts a gradually evolving contextual representation bound to item encodings at unambiguously marked prosodic boundary positions. In a series of reading and recognition memory studies, we demonstrate the role of CSE using three segmented sentence structures as test cases: appositive relative clauses, which have been shown to display bypassing of sentence-medial segments during online comprehension (Dillon et al., 2017; Kim & Xiang, 2023, 2024), and two types of focus-sensitive coordination that can prosodically separate their coordinates ("not only...but also" and "...as well as..."). The studies establish two consequences of CSE during sentence processing: previous contextual states may be reinstated at later points (a mechanism we term Reinstantiation), and in limited cases, the contents of a targeted segment may be accessed to the exclusion of other sentence content following cue-based retrieval (termed Context-Sensitive Retrieval). The account proposed here ultimately argues that bypassing stems from the interaction between (i) shifting the encoding context at prosodic boundary positions and (ii) anticipating upcoming subject-verb dependencies. We argue that this interaction can entirely account for an effect that has previously been attributed to idiosyncratic discourse properties of appositives.