Sigmoidal interview question

How do you approach preprocessing text data for NLP applications? Which techniques have you found most effective?