Pragmatic interpretations can be rapid and robust but are language-dependent inferences like scalar implicature preceded by semantic analysis? While many studies find an initial period of semantic interpretation (e.g., some interpreted as “some-and-possibly-all”), recent work has found that scalar implicatures are rapidly calculated when the quantifier was phonologically-reduced, providing a cue to the partitive construction (e.g., summa instead of some-of). This talk will examine possible reasons for this dichotomy. To distinguish between the effects of prosody and predictability on the speed of scalar implicatures, I will introduce new data from on-line and off-line interpretations in adults. I will argue that these results are most consistent with a model of comprehension involving a dynamic interplay between (1) bottom-up processes which operate over logically-ordered linguistic representations and (2) top-down processes which encode events for communication. I will also discuss how this description informs our understanding of developmental patterns of (non-)inferencing in children.