Understanding and Mitigating Classification Errors Through Interpretable Token Patterns