a study of the impact of persuasive argumentation in political debates
TRANSCRIPT
A Study of the Impact of Persuasive Argumentation in Political Debates
Dr. Amparo E. Cano-Basave and Dr.Yulan He
1
Objective
Study whether persuasive cues and persuasive argumentations can be used as predictors of speakers’ influence ranking on a political debate.
Argumentation
Claim: • Controversial statement to be judge true or false• Not accepted by an audience without additional support
Premises• Support underpinning the validity of a claim
Verbal and social activity of reason which aims to increase the acceptability of a controversial standpoint by putting forward a set of connected propositions intending to justify or refute a standpoint before a rational judge (van Eemeren et al., 1996)
Argumentation, Claim and Premise Example
“[People aren’t investing in America] [because this president has made America a less attractive place for investing and hiring than other places in the world]” (Former Governor Mitt Romney)
claim
premise
Argumentation focuses on the rational support to justify or refute a Standpoint.
Persuasive language
perceptions
Persuasion focuses on language cues aiming at: shaping, reinforcing and changing a response.
beliefsattitudesbehaviors
PersuasiveLanguage
• Emotive lexicons (e.g., atrocious, dreadful, sensational, highly effective)
• Use of alliteration (repetition of first consonants in series of words)
Persuasive Language, Example
“I’m convinced that part of the divide that we’re experiencing in the United States, which is unprecedented, it’s unnatural and it’s un-American, is because we’re divided economically, too few jobs, too few opportunities” (Former Governor Huntsman).
alliteration
Emotive language
Persuasive Argumentation?
NLP in Political textRhetoric• Detecting ideological positions (Sim et al., 2013)• Predicting voting patterns (Thomas et al., 2006; Gerrish and Blei 2011)• Characterizing Power based on linguistic features (Prabhakaran et al., 2013)
Argumentation• Argumentation extraction (Cabrio and Villata, 2012)• Recognizing arguments in online discussions (Filip Boltužić and Jan Šnajder, 2014)
• Stance classification of ideological debates (Hasan and Ng, 2013)
ContributionsWe propose to:
• Persuasive argumentation as a feature for ranking candidates’ influence in political debates
• Port annotations by means of semantic frames
DatasetsPersuasive essays Corpus (Stab & Gurevych 2014)• 90 essays comprising 1,673 sentences• Class-level argument components and argument relations
Table 1. Persuasive essay argument annotation scheme.
DatasetsPresidential Political Debates CorpusManual transcripts of debates for the Republican party presidential primary election (The American Presidency Project)1
• 20 debates (May 2011- February 2012)• 10 candidates (Avg. participation of 6.7 candidates per debate)• 30-40 hours of interaction time• On avg. 20,466 words per debate• Each transcript delimits turns between speakers and moderators • Each transcript marks-up occurrences of the audiences’ reactions (e.g.,
booing, laughter)
1 http:// www.presidency.ucsb.edu/debates.php
Semantic FramesDescription of context in which a word sense is used.FrameNet (Baker et al., 1998)2:
- Over 1000 patterns used in English (e.g., Leadership, Causality, Awareness, Hostile encounter).
2SEMAFOR (Das et al., 2010) for semantic frame extraction
Semantic Frames as Pivoting Features
PersuasiveEssays(PE)
Claim
.
.
AttackRelations
Semantic Frames (SF)
Intentionality act
Cause Change
..
Emotion Directed
Extract most representative frames for each annotation
a1
an
SFa
Claim
AttackRelations
.
.
Semantic Frames as Pivoting Features
Speaker u Entry in a Debate d
Semantic Frames (SF)
Intentionality act
Cause Change
..
Emotion Directed
Speaker u’sBag of SF
SFa
Claim
AttackRelations
.
.
fd,au
x1
xk
.
.
.
fud,claim
fud,premise
fud,attack
.
.
PD
PE
Semantic Frames an Argument Types
Table 2. Top semantic frames for each argumentation type
Argument Type
Top 5 Semantic Frames
Claim Reason, Stage of Progress, Evaluative Comparison, Competition, Cause to Change
Premise Removing, Inclusion, Killing, Cognitive Connection, Causation
For Stance Cause to Make Progress, Collaboration, Purpose, Kinship, Expensiveness
Against Stance
Intentionally Act, Importance, Capability, Leadership, Usefulness
Support Relation
Dead or Alive, Institutions, State Continue, Taking Sides, Reliance
Attack Relation
Usefulness, Likelihood, Desiring, Importance, Intentionally Act
Semantic Frames as Pivoting Features
Arg. Type
Sentence Semantic Frames
Claim If we can turn Syria and Lebanon away from Iran, we finally have the capacity to get Iran to pull back.
Cause Change, Manipulation, Capability
For Stance
And the reason is because that’s how our founding fathers saw this country set up.
Reason, Kinship, Perception Experience
Attack Rel.
But you can’t stand and say you give me everything I want or I’ll vote no.
Desiring, Posture, Capability
Influence Ranking in Political Debates
Influence index (Prabhakaran et al., 2013) calculated based on a speaker’s relative standing on polls released prior to the debate.
The influence index P of speaker u in UD is:
Where pi is the poll percentage assign to speaker u in poll i in the reference polls.
Influence Ranking Approach
Let D denote a debate with a set of speakers UD={u1,..,un} and influence indexes P(ui) for 1< i <n.
Training set for ranking:
R={(ui, γi),.., (un, γn) } where γi is the ranking of ui based on its P(ui) .
We use the Ranking SVM (Joachims, 2006) to estimate the ranking function F which outputs a score for each instance from which a global ordering of data is constructed.
Features for each speaker
External Emotion Cues
- Applause (APL)- Booing (BOO)- Laughs (LAU)- Crosstalk (CRO)
Persuasiveness
- Persuasive argumentation features (semantic frame feature vector representing each annotation scheme (fud,a) for each speaker on each debate)
- Alliteration- Emotive Language (based on emotion re
lated semantic frames, e.g., emotion_directed, emotions_by_stimulus, emotions_by_posibility).
Features for each speaker
Content Features Baselines for Influence Ranking (Prabhakaran et al., 2013)
- Unigrams (UG)- Question Deviation (QD)- Word Deviations (WD)- Mention Percentage (MP)
Ranking Task Evaluation
For each debate, compare the generated ranked list of candidates using the influence ranking approach, against a reference ranked list (gold standard) based on the poll scores for that debate.
- 5-fold cross validation - Report nDCG, nDCG-3, Kendall’s Tau and Spearman correlations.
Influence Ranking Results
Persuasive argumentation alone improves upon UG• Premise and support relation (nDCG and nDCG-3)• Attack Relation (Tau and Spearman)
Results suggest• Tendency to use well supported arguments• Tendency to attack more other candidates by presenting premises refuting a
claim.
Persuasive argumentation + WD + MP significantly outperforms all baselines• Claim• Premise• For stance • Support relation
SummaryGood predictors of influence ranking:• Relevance of “What they said”• Relevance of “Persuasiveness style of their arguments”• Relative importance given by others by means of mentions
Candidates with higher influence ranking • Tend to present more premises while clearly stating their stance (i.e., supporting a claim)
on a particular topic.
Conclusions
• Studied the impact of argumentation in speaker’s discourse and their effect in influencing an audience on supporting their candidature
• To start persuasive argumentation features, proposed a method to port annotations by means of semantic frames
• Premise and Support Relation types, better predictors of a speaker’s influence rank