Abstract:
Metaphor is an ubiquitous figure of speech generally employed to describe one concept in terms of another. Within the field of study of political discourse, metaphor is additionally described as a persuasive device that is used to achieve rhetoric goals and to organize discourse structure. While the use of metaphor in political discourse has been extensively analyzed in small-scale corpus linguistics studies, there are few large-scale studies that explore metaphor patterns in political discourse due to the complexity of manual metaphor extraction. Automatic metaphor identification is however a growing topic of interest within the field of natural language processing (NLP). Adopting a computational approach to metaphor identification might provide a broader insight into the use of figurative language in political discourse. This thesis adopts a metaphor detection model that was designed by Su et al. (2020) to automatically extract metaphors in a corpus of 1721 American presidential speeches. A quantitative and qualitative analysis is performed on the metaphors extracted by the model. The results of the explorative analysis are compared to the findings of studies that do not rely on NLP for metaphor detection in order to evaluate the efficacy of the use of computational methods for corpus-based studies of this scale.
