Local Coherences in the Wild: An Eye-tracking Corpus Study

The phenomenon of local coherences has recently generated significant interest in the sentence processing literature. Tabor, Galantucci, & Richardson (2004) showed that participants are slower to read object-modifying reduced relative clauses (RCs) when the RC verb is part-of-speech ambiguous (tossed) than when it is unambiguous (thrown). This result is problematic for any fully incremental framework because the main-verb interpretation is incompatible with the global context and should thus be ignored by the processor. An unresolved issue with this result, however, is the extent to which such effects occur in natural reading. We test the extent to which such effects occur in natural reading, using eye-tracking data from free reading of newspaper text. Our results support both of the major theories of local coherences, and add ecological validity to the study of local coherences. 2. Corpus, Model, and Control Factors 3. Factor One: P(ti|wi) backgroun