Text this: Adjacent Inputs With Different Labels and Hardness in Supervised Learning