The locally-least cost error repair model for LR-based parsers has heavy burden of computing the right context from the left context at parsing time in case of a syntax error. In previous works([7],[9],[14]), the right context for a state have been computed by uniting the right context of each kernel item in the state at parsing time. Jung([22]) has pointed out redundancy in computation of right context for a state by showing the right context of each kernel item in the state likely to have a common suffix set. And he has proposed a method of computing a left context independent prefix set of right context for a state at parser generation time. But his method is not complete in the sense it does not cover some states which have a left context independent prefix set and his formalisms lack some generality.
In this thesis, Jung's method is supplemented and generalized. And his method is dynamically extended to cover states which have a set of left context dependent prefix set whose size is left context independent. These results are merged to give rise to an unified algorithm using dependency between states. The proposed algorithm is implemented in the Yacc. The improvement is verified by testing the implemented system with several practical programming languages.