Texts have been classified by topic, genre, sentiment orientation (positive or negative opinions), or even by the gender of the authors. This talk will focus on classifying texts by the emotion expressed by the authors. Results on several datasets will be presented. A global dataset is then used for training, in order to obtain a more general classifier. Results of a hierarchical classification approach will also be discussed. A bootstrapping method for extracting paraphrases for emotion depression will be introduced. For each emotion class, the method starts with seed words from WordNet Affect, and it learns equivalent expressions from aligned similar contexts. The paraphrases are evaluated by human judges. At the end, the idea of automatically generating texts that express a given emotion will be discussed. The generated texts can be used in various applications, such as intelligent tutoring systems or digital games where written messages are exchanged, in order to create the impression that the characters have personality.