Crafting a data set to benchmark some tweet language prediction algorithms. CLD is at 8.8% while an aspell dict based predictor makes 5.5%.