Why speakers produce scrambled sentences: An analysis of a spoken language corpus in Japanese

Abstract

The current research analyzes the effect of length and referential phrases on scrambled word-orders in Japanese, using the Corpus of Spontaneous Japanese (CSJ) [National Institute for Japanese Language & National Institute of Information and Communications Technology (2004)]. Three types of sentence structures are examined: Sentences with a transitive verb, a ditransitive verb and a transitive verb that takes ni-marked locatives. Within sentences in which all relevant arguments are overtly present (totaling 1,107 sentences), the ratios of scrambled order in transitive sentences, ditransitive sentences and transitive sentences with a locative ni-phrase are 6.5, 31.7 and 49.1%, respectively. This shows a great deal of variety in the frequency of scrambled sentences among sentence types. The results also confirm the effects of length and referential phrase in the production of scrambled sentences. Namely, long arguments or arguments containing a referential phrase are placed ahead in the sentences. In addition to the length effect, the study finds that the most common case of scrambling is when the canonically-preceding phrase is only one Bunsetsu (the smallest semantic unit, content words followed by a case marker or a postposition when applicable) in length. The results show that Japanese speakers tend to produce a syntactic structure that positions words of semantic or discourse prominency ahead of others, as with speakers of head-initial languages. On the other hand, they position long phrases ahead of short phrases rather than postponing like speakers of head-initial languages. This difference between the two types of language is discussed in terms of the head-directions and the flexibility of word-order.