True or false? Assume we are using word n-grams as features to perform sentiment classification. Then higher values of $n$ will usually be less prone to overfitting (i.e., for higher values of $n$, the difference between training and testing accuracies will be smaller).