Arai, M.
Seijo University
Previous studies demonstrated that the greater an ambiguity effect with a garden-path sentence is, the larger the change in comprehenders? future expectation about the structure, providing support for error-based structural learning (Fine et al., 2013). Our study examined how this error-based learning operates in L2 sentence processing. Participants (N=28) were randomly assigned to one of two groups and asked to read relative clause sentences among fillers in two experimental sessions using a self-paced reading task. In the first ?exposure? session, one group saw reduced relative clauses (RRCs) and the other group did unreduced relative clauses (URCs) along with fillers. Our results showed that temporarily ambiguous RRCs caused an ambiguity effect reflecting prediction error due to initial misanalysis and that the reading times in the disambiguating region of RRCs were longest at the beginning and gradually decreased through the session although those for URCs did not change across items. In the next ?test? session, both groups read both types of relative clauses. We found that the difference in reading times at disambiguation between RRCs and URCs was significantly smaller in the group who saw RRCs in the exposure stage. Our results revealed an important role of prediction error in L2 processing.