日本語における「ガ格連続文」の処理について

Abstract

A construction like "NP-ga NP-ga Predicate" in Japanese invites an ambiguity concerning the status of the second NP-ga. Since the nominative marker, -ga, can indicate either a subject or an object, the second NP can be the subject of a following clause or can be the object of a single clause. The former case would be exemplified as Taro-ga [Hanako-ga waratta] to itta. (Taro said that Hanako laughed) and the latter case as [Taro-ga Hanako-ga sukida]. (Taro loves Hanako). The results of four experiments (two off-line and two on-line) were reported. The off-line experiments showed that the sentence type with the second NP as "object" is preferred. On the other hand, the on-line experiments revealed that the sentence type with the second NP as "subject" is preferred. We feel that this discrepancy can be explained by postulating two assumptions. (i) There is a difference of mechanism between off-line and on-line processing. (ii) In on-line processing, the parser adopts the "selective delayed parsing." On encountering an ambiguous element (NP-ga can be either a subject or an object), the "tree-designer" requires the "tree-builder" to suspend the routine work of incremental structure building. Finally, we mentioned that the effect of animacy has not been highly considered in this experimental paradigm. Further investigation is necessary for more understanding of the parsing mechanism of Japanese.