This experiment employed the visual world paradigm to determine if participants could use selectional restrictions derived from the sentential verb to predict a target referent, replicating Altmann and Kamide (1999). Eye movements were monitored as participants looked at semi-realistic visual scenes and listened to associated sentences. Results showed significantly earlier fixations to the target referent when the semantics of the verb constrained the subsequent range of plausible referents than when the verb conferred no such restrictions. For example, in a scene consisting of a boy, a cake, and various inedible distractors, looks to the cake manifested themselves earlier on hearing ‘the boy will eat the cake’ than ‘the boy will move the cake’. Secondly, we looked at the phenomenon of disfluencies, specifically replacements. Ferreira et al., (2004) suggest that lexical information activated at the reparandum is not fully erased after repair and can continue to affect later processing. We therefore aimed to ascertain whether predicted themes based on the verb would linger on hearing ‘the boy will eat, err, move the cake’. We expected that if the incorrect verb ‘eat’ has an enduring effect we would witness increased looks to the cake. This was compared to a conjunction condition (‘the boy will eat and move the cake’), so any differences observed would be solely due to the effect of the repair. By the pre-noun determiner there was no evidence of the selectional restrictions of the first verb having a lasting effect. However, no firm conclusions can be drawn as the length and unnaturalness of the edit interval could be confounding the results.