Abstract:
Aim of this master thesis is a deeper understanding of the functioning of satire in written texts – in this specific case political satire articles – of length longer than simple sentences, in order to apply this knowledge to future computational tasks. In the first chapter there is an attempt to define satire, particularly bringing into focus the specific features of this genre. This is done by analysing the relationship with previously proposed theories on humour and their linguistic entailments, so that a line could be drawn under the grey area of definitions of humour genres, at least for what does concern the use of satire. The second chapter looks at the pros and cons of computational approaches used until now for the resolution of different tasks related to the automatic processing of humorous texts. The purpose is to outline best practises, trying to understand if it is possible to use them – consequently proper rearrangements – for more complex NLP tasks on long satiric text, in a way that allows achieving meaningful results. The third chapter is dedicated to the detailed description of the experimental part of the thesis and its outcomes. On a corpus built for that purpose and composed of 112 political satire articles manual tagging has been carried out using a reduced (and modified with new criteria, where needed) version of the Appraisal Framework (Martin & White, 2005). Then a typological classification of all voices contained in the automatically collected lexicons (one for each author) has been created using three linguistic traits – namely idiomatic, metaphorical and none. This has been done in order to assign to all the entries a feature, related to the kind of use the author did of a specific item/sequence. The final step of the experiment consists in gradual attempts to tag automatically twenty new texts (ten each) by the same authors. Eventually, by virtue of the collected data and pondered interpretations, conclusions are drawn with particular focus to further research proposals.
