PREDICTIVE INFERENCES OCCUR ONLINE, BUT WITH DELAY - CONVERGENCE OF NAMING AND READING TIMES

Naming and word reading times were used to investigate the time course of predictive inferences. Priming contexts were followed to disambigu ating sentences in which a target word either confirmed or disconfirme d the consequence suggested by the context. In Experiment 1, sentences were presented word by word at a predetermined pace; readers pronounc ed the target word, which appeared either 500 or 1250 ms after the ons et of the last word in the priming context. In Experiment 2, the parti cipants read sentences one to four words at a time using the self-pace d moving-window technique; reading times for the target word, the post target region, and the last word in the disambiguating sentence were c ollected. There was facilitation (a) in naming the confirming target w ord when it was primed in the 1250-ms stimulus onset asynchrony (SOA), but not in the 500-ms SOA; and (b) in reading the confirming target w ord plus the posttarget region (spill-over effect), but not in the tar get word itself. These results suggest that predictive inferences occu r on-line, but require time to be drawn and that they are initially en coded to some degree, but completed later.