Abstract
Since complex sentences consist of more than two clauses, we have more than two subjects in one complex sentence. Therefore it is important to recognize coreference relations among these subjects for identifying the referents of these subjects. We focus here on subjects coreference relations of Japanese complex sentences conjoined by node or kara, both of them mean because in English. The reason why this is important is that in Japanese subjects are very frequently omitted, in other words come to be zero subjects. In this circumstance, to resolve zero anaphora is one of key factors for understanding or translating. For this purpose we use semantic features of predicates of main and subordinate clause respectively, and calculate the frequencies for each combination of every semantic feature of subordinate clause and those of main clause. By analyzing these frequencies we find several tendencies, for instance, if a subordinate predicate is linguistic activity like speaking, then the subject of main clause is distinct from the subject of subordinate clause in its referent. We also account for these tendencies based on linguistic and cognitive aspects of human beings.
