Abstract
In a “subject-oriented sentence” such as “Taroo-ga Hanako-ni Tokyo-e iku koto-o hakuzyoosita (Taroo confessed to Hanako to go to Tokyo)”, the person who will go to Tokyo is “Taroo”. On the other hand, in an “object-oriented sentence” such as “Taroo-ga Hanako-ni Tokyo-e iku koto-o meireisita (Taroo ordered to Hanako to go to Tokyo)”, “Hanako” will go to Tokyo. In the above examples, the main clause verb's location at the end of sentence causes a tentative ambiguity which is not resolved until this verb becomes available to the human sentence processing mechanism (i.e., a “parser”). How does the parser deal with this informationally ambiguous situation? More specifically, is the processing procedure “top-down” or “bottom-up”? We report the results of an experiment using a dichotic-listening method, examining the real-time processing of sentences. A sentence was presented to one ear of a subject, and after the offset of the sentence, a possible antecedent for the empty subject (i.e., either the subject or the object of the main clause) was given to the other ear. There were 6 test points with 300 msec intervals. The subject's task was to answer whether the given antecedent would really go to Tokyo or not, by pressing the yes- or no-key as quickly as possible. In the case of ‘yes’ responses, at 0 msec ISI, the mean RT in the subject-oriented sentences was significantly shorter than that in the object-oriented sentences. In the case of ‘no’ responses, only the effect of ISI was statistically significant, showing that the RTs became shorter, and overlapped with those of ‘yes’ responses beyond 900 msec ISI. These results suggest that the human parser assumes that the main clause subject is the default antecedent for the empty subject of a subordinate clause. This initial “top-down” assumption seems to be suppressed by a “bottom-up” process by 900 msec after the end of the sentence.
