Comprehension of text develops over time, such that a single word may lead to the activation of different concepts—and thus different expectations—at different times after it is encountered. Since the pioneering work of Neely (1977), much evidence has been found that multiple senses of a word are activated shortly after it is encountered, but that senses that are contextually irrelevant are later inhibited. This has implications for the processing of anomalies in sentences, including malapropisms like switching “antidote” for “anecdote”: if the anomaly is primed by the semantic context (e.g., it is preceded by the word “doctor”) shortly before it is encountered, it should be less likely to be noticed by the reader and is less likely to impede reading (and comprehension) of the sentence. We present results from a reading-time study using sentences in which malapropisms are presented in either neutral or semantically congruent contexts, with the semantic prime at varying distances from the anomaly (e.g., “doctor”, above). These results are interpreted within a dynamic sentence comprehension model based on the attractor model of Tabor, Juliano, & Tanenhaus (1997).