Focus marking on a phrase like apple, as in (1), triggers the activation of contrastive alternatives to the expression in focus (e.g., expressions like pears, leeks, or cabbage; Braun & Tagliapietra, 2010).

Priming studies suggested a two-stage model of how these focus alternatives are considered in real-time comprehension: first, semantic associates are activated as in normal word recognition, and then alternatives that contrast appropriately with the focus are selected from among them (Husband & Ferreira, 2015). These priming studies did not manipulate discourse context. Reading, memory, and visual world studies, however, have shown that comprehenders also utilize discourse information and world-knowledge to identify alternatives (Sedivy et al., 2002, Fraundorf et al., 2013, Kim et al., 2015). In this talk, I will therefore consider two possible extensions of Husband & Ferreira’s (2015) model that take discourse information into account: one in which discourse information is only used at a late stage to select appropriate alternatives after semantic associates are primed, and one in which discourse information may itself be used to activate alternatives during earlier processing stages. Under the first, purely selection-based model, any effect of the contextual (in)appropriateness of alternatives should not be contemporaneous with effects of semantic association, because discourse information should be used only in the selection process, after an initial candidate set has been activated. I present the results of three Maze reading studies which suggest instead that comprehenders rapidly access discourse information to differentiate among potential alternatives, whose activation does not depend exclusively on semantic associate priming.