In this talk we discuss the challenge of sentiment classification: automatically determine the semantic orientation (polarity) of a subjective piece of text, such as customer reviews, opinion surveys, news reports, etc. While machine learning with basic word-based features is strong enough for simple topic classification tasks, it appears to be limited when applied to the more demanding task of sentiment classification. Adding a linguistic dimension proves to make a real difference. We discuss the types of enhancements which are found to contribute most.