A prediction of reduction goals in LR parsing is formalized. Based on it, an LR parser with pre-determined goals is generated. In addition, we show that the proposed LR parser can be applied to phrase-level error recovery and the computation of right context