What can Reveal 1,018 Speeches of Fidel Castro? Sergio Peignier ? , Patricia Zapata ⇧ ? CMLA, ENS Paris-Saclay, CNRS UMR 8536, Université Paris-Saclay, France ⇧ Carrera de Lingüística e Idiomas, Universidad Mayor de San Andrés, La Paz, Bolivia 8 June 2018 1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
What can Reveal 1,018 Speeches of Fidel Castro?
Sergio Peignier?, Patricia Zapata⇧
?CMLA, ENS Paris-Saclay, CNRS UMR 8536, Université Paris-Saclay, France
⇧Carrera de Lingüística e Idiomas, Universidad Mayor de San Andrés, La Paz, Bolivia
8 June 2018
1
Introduction
Fidel Castro Discourse Analysis
#Understand his Political Success
Fidel CastroMajor Cuban politician.Half-century government(1959-2008).Well known rhetoric skills.29/09/1960 UN longest speech (4h39min).
Use small (and possibly non-representative) samples#
Possibly biased results
3
Key Points
Corpus1,018 speeches
CharaudeauLinguisticDiscourseAnalysis
PrototypicDiscourses(Trees)
MachineLearningFramework
Machine Learning framework
#More representative systematic study.
4
Outline
1 Machine Learning Framework
2 Discourse Analysis Methodology
3 Results
4 Conclusion and Perspectives
5
Outline
1 Machine Learning Framework
2 Discourse Analysis Methodology
3 Results
4 Conclusion and Perspectives
6
Overall FrameworkCorpus1,018 speeches
Word-Embedding
CharaudeauLinguisticDiscourseAnalysis
SubspaceClustering
Directed Trees
Framework stepsPre-processing: data cleaning and empty words filtering.Word-Embedding ! Word Vector Representations.Subspace Clustering ! Contextually Linked Word Clusters.(Discursive Contexts)Directed Trees ! Intelligible Structure(Prototypic Discourses).Python 2.7 (re, nltk, gensim Word2Vec, SubCMedians)
7
Word-EmbeddingWord2Vec [Mikolov et al. 2013]
Distributional Hypothesis: "Words that occur in similar contextstend to have similar meanings" [Harris 1954]
Persuade through feelings and passions.Recruitment process
Lead the audience to accept the speaker’s project willingly.
Moral Values.Social Welfare.Progress (economy, technology).
Religious belonging.Ethnic belonging.Nationalism.
Rhetoric of effectsCreate feelings and passions in
the audience.HopePrideCourage
FearHatredOutrage
14
Pathos (2/2)
Triadic Scenario
Disastrous scenariopeople = victim
EnemyExternal/internal
HeroSpeaker/Ideology
Fights
Causes
Solves
15
Logos
Convince through logical reasoning and argumentation.
Simplifying strategiesSingularizationReduce the number of ideas described.EssentializationExpress a complex idea with few words.Formula = Singularization + Essentialization.Slogans.
Argumentative strategiesSimplified causal arguments.Analogy with past events/characters.Detailed descriptions ! Veracity.
16
Outline
1 Machine Learning Framework
2 Discourse Analysis Methodology
3 Results
4 Conclusion and Perspectives
17
Cluster Model Structure
Four largest clusters: 466, 93, 70, 344⇠ 60% Number of words.⇠ 50% Vocabulary.⇠ Analysis of 595 discourses out of the 1,018.Small clusters ! Specific vocabulary or general words.
18
Cluster 466Political Marxist Project
Futur PlanMarxist RevolutionDefence of the project
Identification with PeopleDetails ! Veracity
Moral Values
MarxistRevolution
FuturProject
19
Cluster 344Cold War Triadic Scenario
EnemiesOligarchy, USA
Cold War conflicts
HeroesCommunism, Cubanarmy, people
FeelingsFear, Hatred,Outrage
HerosEnemy
Negative Scenario
20
Cluster 70Social welfare, audience as beneficiary
Society as actor and beneficiarySocial welfare: Education, employment, healthFeelings ! Hope, Pride
Welfare
21
Cluster 93Economic and Technological Progress
Progress
Economic progressInfrastructures
Needs/ObligationsFeelingsHope, Pride
EthosExpert, Authority
Details ! Veracity
22
Castro Discursive Strategies
Pathos (51%)Triadic scenarioProgress Welfare NationalismHope Pride Fear Heroism