抄録
This paper brings up various ‘branching-COMP’phenomena, and shows that the version of c-command assumed in Chomsky 1981 to treat one branching-COMP phenomenon, i.e. that-trace effects, is inadequate to deal with other branching-COMP phenomena. To overcome the inadequacy, the claim is made that the notion c-command be revised so that any element under COMP will c-command others in embedded Ss, and that a rule be provided which reduces the maximal projection S to S unless there is any element with features between it and another S-type node. For each of the claims, a good deal of evidence is provided independently of the main topic of this paper, branching-COMP phenomena. It is argued, furthermore, that the treatment of that-trace effects based on these claims is more natural than Chomsky's in the light of the generalization of environments where variables may appear.
