Optimization of LR(k) reduced parsers