<p>This study examines the comprehension of characters’ emotional states in written narrative texts. Recent theories of embodied cognition suggest that emotions are grounded in somatosensory information involving ‘reexperiencing’ of actions and perceptions. In addition, views of graded embodied cognition suggest that there are levels of the embodiment of language, focusing on how perceptual and motor information interact during text comprehension. In the present experiment, one-paragraph stories were written to express six basic emotions: fear, sadness, anger, disgust, happiness, and surprise. Four stories were written for each emotion and for each story four critical sentences were composed. The critical sentences comprised a combination of emotion- and action-based components that matched or mismatched the story. Participants read the stories and the critical sentences. Their task was to respond to a question about the emotional state of the story’s main character, while we measured response latencies and errors. The results of the error rate analyses suggest that, while reading texts, sensory knowledge about characters’ emotional states is activated, but this knowledge is significantly moderated by action-based knowledge. A computational model was used to further confirm these results. The model was trained to predict emotion and action words from critical sentences using linguistic context from stories. Both, the action and emotional words activations showed a distinct effect on participants’ comprehension accuracy.</p>