The Processing Cost of Scrambling and Topicalization in Japanese

This article presents two reading comprehension experiments, using the sentence correctness decision task, that explore the causes of processing cost of Japanese sentences with S(NOM)O(ACC)V, S(TOP)O(ACC)V, O(ACC)S(NOM)V, and O(TOP)S(NOM)V word orders. The first experiment was conducted in order to see if either syntax or frequency plays a significant role in the processing of these sentences. The results of the first experiment have shown that both the structure-building process and frequency directly affect processing load. We observed that there was no difference in processing cost between S(NOM)O(ACC)V and S(TOP)O(ACC)V, both of which are easier to process than O(ACC)S(NOM)V, which is in turn easier to process than O(TOP)S(NOM)V: S(NOM)O(ACC)V = S(TOP)O(ACC)V < O(ACC)S(NOM)V < O(TOP)S(NOM)V. This result is the mixture of the two positions. Specifically, the structure building cost of S(TOP)O(ACC)V was neutralized by its high frequency. The aim of the second experiment was to investigate the interaction between syntactic structure, frequency, and information structure. The results showed that the processing cost of O(ACC)S(NOM)V was facilitated by given-new ordering, but S(NOM)O(ACC)V, S(TOP)O(ACC)V, and O(TOP)S(NOM)V were not. Thus, we can conclude that information structure also influences processing cost. In addition, the distribution of informational effects can be accounted for by Kuno's (1987, p. 212) Markedness Principle for Discourse Rule Violations: S(NOM)O(ACC)V and S(TOP)O(ACC)V are unmarked/canonical options, and as such are not penalized even when they violate given-new ordering, O(ACC)S(NOM)V is penalized when it does not maintain given-new ordering because it is a marked/non-canonical option, and O(TOP)S(NOM)V is penalized even when it obeys given-new ordering possibly because more specific contexts are needed. Another reason for the increased processing cost of O(TOP)S(NOM)V is a garden path effect; upon encountering O(TOP) of O(TOP)S(NOM)V, the parser preferentially (mis)interpreted it as S(TOP) due to a subject-before-object preference. The revision of the interpretation may be the cause of the high processing cost observed in O(TOP)S(NOM)V.