Emotion classification in text is typically performed with neural network\nmodels which learn to associate linguistic units with emotions. While this\noften leads to good predictive performance, it does only help to a limited\ndegree to understand how emotions are communicated in various domains. The\nemotion component process model (CPM) by Scherer (2005) is an interesting\napproach to explain emotion communication. It states that emotions are a\ncoordinated process of various subcomponents, in reaction to an event, namely\nthe subjective feeling, the cognitive appraisal, the expression, a\nphysiological bodily reaction, and a motivational action tendency. We\nhypothesize that these components are associated with linguistic realizations:\nan emotion can be expressed by describing a physiological bodily reaction ("he\nwas trembling"), or the expression ("she smiled"), etc. We annotate existing\nliterature and Twitter emotion corpora with emotion component classes and find\nthat emotions on Twitter are predominantly expressed by event descriptions or\nsubjective reports of the feeling, while in literature, authors prefer to\ndescribe what characters do, and leave the interpretation to the reader. We\nfurther include the CPM in a multitask learning model and find that this\nsupports the emotion categorization. The annotated corpora are available at\nhttps://www.ims.uni-stuttgart.de/data/emotion.\n