Parafoveal preview benefit in reading is not cumulative across multiple saccades.

Two empirical predictions can be generated from recent parallel processing models of eye movement control in reading concerning the phenomenon of parafoveal preview benefit. These predictions derive from the assumption that lexical activation accumulates across more than one saccade. A large corpus of eye movement data was used to determine whether parafoveal preview benefit of a target word is modulated by (1) its distance from the penultimate fixation, and (2) the amount of time it spends in the 'perceptual span'. There was an unexpected inverse relationship between first fixation duration and distance: fixations were longer the closer the penultimate fixation was to the target word, and a positive relationship between fixation duration and time: fixations were longer the greater the summed durations of the ultimate and penultimate fixations. These findings represent a challenge for computational models of eye movements in reading.