Eye-tracking evidence for active gap-filling regardless of dependency length

Previous work on real-time sentence processing has established that comprehenders build and interpret filler-gap dependencies without waiting for unambiguous evidence about the actual location of the gap ("active gap-filling") as long as such dependencies are grammatically licensed. However, this generalisation was called into question by recent findings in a self-paced reading experiment by Wagers and Phillips, which may be taken to show that comprehenders do not interpret the filler at the posited gap when the dependency spans a longer distance. In the present study, we aimed to replicate these findings in an eye-tracking experiment with better controlled materials and increased statistical power. Crucially, we found clear evidence for active gap-filling across all levels of dependency length. This diverges from Wagers and Phillips's findings but is in line with the long-standing generalisation that comprehenders build and interpret filler-gap dependencies predictively as long as they are grammatically licensed. We found that the effect became smaller in the long dependency conditions in the post-critical region, which suggests the weaker effect in the long dependency conditions may have been undetected in Wagers and Phillips's study due to insufficient statistical power and/or the use of a self-paced reading paradigm.