Charismatic Speech

1
Charismatic Speech

• Andrew Rosenberg
• Spoken Language Processing
• 4/24/06

2
Overview

• Background
• Previous Work
• Speech Study
• Text Study
• Conclusion Future Work

3
Overview

• Background
• What is charisma?
• Does charismatic speech exist?
• Charismatic Speech vs. Emotional Speech
• Why study charismatic speech?
• Previous Work
• Speech Study
• Text Study
• Conclusion Future Work

4
Background - What is charisma? (What do I mean
by charisma?)

• Not closed door charisma.
• Rather, political (or religious) charisma
• The ability to attract, and retain followers by
  virtue of personality as opposed to tradition or
  laws. (Weber)
• E.g. Ghandi, Hitler, Che Guevara.
• Charismatic speech Speech that encourages
  listeners to perceive the speaker as
  charismatic.

5
Background - Is there such a thing as charismatic
speech?

• Pro
• Potential charismatic leaders must communicate
  with would-be followers.
• Charismatic leaders have historically had a
  particular gift at public speaking
• Hitler, MLK Jr., Castro.
• Con
• Charisma as a relationship between leader and
  followers.
• The mythologizing of a charismatic leader extends
  beyond public address.

6
Background - Charismatic speech vs. Emotional
speech

• Similarities
• Paralinguistic phenomena.
• Not represented the traditional
  syntax-semantics-pragmatics paradigm.
• Can be studied in the same way via perceptual
  studies
• Differences
• Charisma is not a speaker state.
• Social context of charisma.
• Personal attitudes towards charisma.

7
Background - Why study charismatic speech?

• General scientific interest.
• Feedback system for politicians and academic
  instructors.
• Identification of potential charismatic leaders
• Automatic generation of charismatic-like speech

8
Overview

• Background
• Previous Work
• C. Tuppen, Dimensions of Communicator
  Credibility An oblique solution.
• A. Hamilton B. Stewart, Extending an
  Information Processing Model of Language
  Intensity Effects
• Speech Study
• Text Study
• Conclusion Future Work

9
Previous Work - Tuppen

• Christopher Tuppen, Dimensions of communicator
  credibility An oblique solution, Speech
  Monographs(41), 1974.
• 101 subjects read a booklet containing ten
  character sketches.
• Student, professor, ad exec, farmer, unethical
  businessman, doctor, ret. Army officer, man of
  religion, hippie, tv personality.
• Topics how much sleep you need, marijuana and
  health, duration of US envolvement in SE Asia,
  and tuition at State Colleges.
• The subjects rated each communicator on 64
  scales.
• 28 bipolar adjective, 36 seven-point Likert
  scales.

10
Previous Work - Tuppen (2)

• The subject ratings were grouped using cluster
  analysis
• Cluster 1 Trustworthiness
• Trustworthy, honest, safe, dependable, reputable,
  etc.
• Cluster 2 Expertise
• Qualified, skilled, informed, experienced, etc.
• Cluster 3 Dynamism
• Bold, active, aggressive, strong, emphatic, etc.

11
Previous Work - Tuppen (3)

• Cluster 4 Co-orientation
• Created a favorable impression, stood for a group
  whose interests coincided with the rater,
  represented acceptable values, was someone to
  whom the rater would like to listen.
• Cluster 5 Charisma
• Convincing, reasonable, right, logical,
  believable, intelligent, whose opinion is
  respected, whose background is admired, in whom
  the reader has confidence.

12
Previous Work - Hamilton Stewart (1)

• M. Hamilton B. Stewart, Extending an
  Information Processing Model of Language
  Intensity Effects, Communication Quarterly
  (412), 1993
• How forceful should my language be in order to
  maximize my social influence?
• I.e., what is the relationship between language
  intensity and persuasion.

13
Previous Work -Hamilton Stewart (2)

• Intensity is expressed by manipulating two
  language features emotionality and specificity.
• Emotionality degree of affect present in the
  language. Ranges from stolid displays to
  histrionics.
• Specificity degree to which precise reference is
  made to attitude objects.
• Attitude change is a product of message
  discrepancy, perceived source credibility and
  message strength.

a - attitude, f - force, s - source credibility d
- discrepancy, c - counterargument ? - impact
parameter
14
Previous Work -Hamilton Stewart (3)

• 518 subjects presented with a persuasive
  message with manipulated intensity.
• The messages language was evaluated on 11 terms
  using a 7-point bipolar adjective scale.
• Intense, strong, active, extreme, forceful,
  emotional, vivid,vigorous, powerful, assertive,
  potent
• Perceived source competence, trustworthiness and
  dynamism were assessed.

15
Previous Work -Hamilton Stewart (4)

• Correlations between subject ratings and
  manipulated features were calculated using a
  causal modeling program, PATH.

Extremity of position
.42
charisma sequence
-.32
Manipulated intensity
Perceived intensity
Source dynamism
Source competence
Source trustworthiness
.73
.52
.78
.64
-.18
16
Overview

• Background
• Previous Work
• Speech Study
• Questions
• Description
• Results
• Text Study
• Conclusion Future Work

17
Speech Study - Questions

• Do subjects agree about what is charismatic?
• What do subjects mean by charismatic?
• What makes speech charismatic?

18
Speech Study - Description

• Subjects Friends and colleagues, no incentive
• Interface Presentation of 45 short speech
  segments (2-30secs) via a web form
• Dependent variables 5-point Likert scale ratings
  of agreement on 26 statements.
• Duration avg. 1.5 hrs, min 45m, max 3hrs

19
Speech Study - Description

• Interface
• http//www1.cs.columbia.edu/amaxwell/survey/

20
Speech Study - Description

• Materials 45 tokens of American political speech
• Speakers 9 Candidates for Democratic Partys
  nomination for President
• Clark, Dean, Edwards, Gephardt, Kerry, Kucinich,
  Lieberman, Mosley Braun, Sharpton
• Topics Postwar Iraq, Healthcare, Bushs Tax
  plan, Reason for Running, Content-Neutral

21
Speech Study - Description

• Example Tokens
• 1.
• 2.
• 3.
• 4.

22
Speech Study - Results

• Inter-subject agreement
• Using the weighted kappa statistic with quadratic
  weighting, mean kappa was 0.213
• Inter-subject agreement by token
• No significant differences across all tokens
• Inter-subject agreement by statement
• The individual statements demonstrate
  significantly different agreements

23
Speech Study - Results

• Most consistent statements
• Charisma 0.224 (8th)

• Least consistent statements

The speaker is accusatory 0.512
The speaker is passionate 0.458
The speaker is intense 0.431
The speaker is angry 0.404
The speaker is enthusiastic 0.362
The speaker is trustworthy 0.037
The speaker is reasonable 0.070
The speaker is believable 0.074
The speaker is desperate 0.076
The speaker is ordinary 0.115
24
Speech Study - Results

• Statement Co-occurrence
• Using the kappa statistic determined which pairs
  of statements were most closely correlated with
  the charismatic statement.

The speaker is enthusiastic 0.606
The speaker is charming 0.602
The speaker is persuasive 0.561
The speaker is boring -0.513
The speaker is passionate 0.512
The speaker is convincing 0.503
25
Speech Study - Results

• Speaker Influence
• There is a significant difference between
  speakers (p1.75e-2)
• Most charismatic
• Rep. Edwards (3.73)
• Rev. Sharpton (3.40)
• Gov. Dean (3.32)
• Least charismatic
• Sen. Lieberman (2.38)
• Rep. Kucinich (2.73)
• Rep. Gephardt (2.77)

26
Speech Study - Results

• Genre Influence
• The tokens were taken from debates, interviews,
  stump speeches, and a campaign ad
• Stump speeches were the most charismatic. (3.28)
• Interviews the least. (2.90)
• Topic Influence
• No significant influence.

27
Speech Study Results

• Speaker Recognition
• Subjects were asked to identify which, if any,
  speakers they recognized at the end of the study
• Mean 3.25
• Subjects rated recognized speakers (3.28)
  significantly more charismatic than those they
  did not (2.99).

28
Speech Study - Results

• Acoustic/Prosodic Properties
• Min, max, mean, std. dev. F0 and intensity
• Phrase dynamics
• Length (seconds)
• Phrase final behavior rising, falling, plateau
• ToBI Pitch accent type.
• Lexical Properties
• Function/Content word ratio
• Pronoun density
• Lexical complexity
• Length (words, syllables)
• Repetition of words
• Number of disfluencies

29
Speech Study - Results

• Properties highly correlated with ratings of
  charisma
• Length. More content, more charismatic.
• Min, max, mean std. dev. of F0 over male speakers
• zscore of mean F0 (calculated over speaker)
• Higher in pitch range, more charismatic
• Mean intensity
• Fewer rising contours (L-H, H-H)
• Fewer L and LH pitch accents

30
Speech Study Results

• Faster speaking rate (syllables per second)
• Mean and standard deviation of normalized phrase
  intensity
• Standard deviation of normalized maximum pitch
• First person, but not second person, pronoun
  density
• Lexical complexity (mean syllables per word)
• More repeated words
• Fewer disfluencies

31
Overview

• Background
• Previous Work
• Speech Study
• Text Study
• Questions
• Description
• Results
• Comparisons to Speech results
• Conclusion Future Work

32
Text Survey - Questions

• When reading a transcript of speech, do subjects
  rate charisma consistently?
• What do subjects mean by charisma?
• Do they mean the same thing when referring to
  text and speech?
• How does what is said influence subject ratings
  of charisma?

33
Text Survey - Description

• Subjects 24 paid participants found
• http//newyork.craigslist.org
• Talent gigs section
• Interface Presentation of 60 short transcripts
  (words) via a web form
• Dependent variables 5-point Likert scale ratings
  of agreement on 26 statements.
• Duration avg. 1.5 hrs, min 45m, max 3hrs

34
Text Survey - Description

• Interface
• http//www1.cs.columbia.edu/amaxwell/textsurvey/A
  /

35
Text Study - Descrption

• Materials 60 of 90 tokens of American political
  speech
• The 90 transcripts were the 45 used in the speech
  study, and 45 longer paragraphs
• Each subject was presented with all 45 short
  (mean 28 words) and a semi-random set of 15 long
  transcripts (mean 130 words)
• Speakers Same as Speech Study
• Topics Same as Speech Study

36
Text Study - Description

• Examples
• Token 1

Were driving seniors out of medicare into HMOs. Every provision that wouldve brought down the cost of prescription drugs, the drug companies were against em all. They all came out.
37
Text Study - Description

• Examples
• Token 2.

and Id like to begin by, saying that I hope that, this afternoons talk will be an opportunity to challenge some underlying assumptions that we have about the world cause thats why Im uh running for President.
38
Text Study - Description

• Examples
• Token 3

stabilize iraq because we occupy it. Yet he will not talk about the deficits in the fifty states we already occupy.
39
Text Study - Description

• Examples
• Token 4

by two thousand five and then let their parents on a sliding scale based on income buy into medicaid at a price much below what theyd have to pay in the market.
40
Text Study - Description

• Some tokens are rated very similarly whether
  presented as speech or a transcript.
• Example 1 always charismatic
• Example 2 always uncharismatic
• Others are rated very differently
• Example 3 more charismatic in speech
• Example 4 in text

41
Text Study - Results

• Inter-subject agreement
• Using the weighted kappa statistic with quadratic
  weighting, mean kappa was 0.149
• Inter-subject agreement by token
• No significant differences across all tokens
• Inter-subject agreement by statement
• The individual statements demonstrate
  significantly different agreements

42
Text Study - Results

• Most consistent statements
• Charisma 0.134 (18th)

• Least consistent statements

The speaker is accusatory 0.280
The speaker is angry 0.263
The speakers message is clear 0.206
The speaker is friendly 0.197
The speaker is knowledgeable 0.193
The speaker is spontaneous 0.039
The speaker is ordinary 0.048
The speaker is boring 0.050
The speaker is desperate 0.064
The speaker is enthusiastic 0.093
43
Text Study - Results

• Charismatic statement cooccurrence
• Using the kappa statistic determined which pairs
  of statements were most closely correlated with
  the charismatic statement.

The speaker is charming 0.576
The speaker is enthusiastic 0.511
The speaker is persuasive 0.503
The speaker is powerful 0.485
The speaker is convincing 0.483
The speaker is passionate 0.446
44
Text Study - Results

• Those statements that positively cooccur with the
  charismatic are identical in the speech and text
  study
• Charming, enthusiastic, persuasive, convincing,
  passionate

45
Text Study - Results

• Speaker Influence
• There is a significant difference between
  speakers (p1.67e-10)
• Most Charismatic
• Gen. Clark (3.61)
• Sen. Kerry (3.56)
• Gov. Dean (3.54)
• Least Charismatic
• Sen. Lieberman (3.03)
• Rep. Kucinich (3.12)
• Amb. Mosley-Braun (3.23)

46
Text Study - Results

• Genre Influence
• Looking at only original speech tokens, genre
  demonstrates a significant influence on charisma
  (p9.18e-14)
• Stump (3.34) and debate (3.32) above mean (3.15)
• Interview below mean (2.85)

47
Text Study - Results

• Speaker Recognition
• No speaker recognized by every subject, no
  subject recognized every speaker (mean1.22)
• Subjects rated recognized speakers (3.48)
  significantly more charismatic than those they
  did not (3.22).

48
Text Study - Results

• Correlation of lexical properties with ratings of
  charisma
• Function/Content word ratio
• Positively correlated (p.0058)
• Pronoun density
• First person very significant (p1.4e-4) but
  negatively correlated.
• Lexical complexity (mean syllables per word)
• uncorrelated
• Length
• No correlation, however, the amount of time a
  subject spent on a particular token positively
  correlated (p0.046)
• Repetition
• Weak positive correlation (p0.0757)
• Number of Disfluencies
• Strongly negatively correlated (p1.46e-7)

49
Overview

• Background
• Previous Work
• Speech Study
• Text Study
• Conclusion
• Future Work

50
Conclusion

• Enthusiasm, passion, charm, persuasion and being
  convincing used to describe someone who they
  find charismatic.
• Personal speech is considered more charismatic
  when heard, but not when read.
• Emotion is largely insignificant to judgments of
  charisma.
• The lexical and acoustic/prosodic properties
  reflect the presence of enthusiasm and passion

51
Conclusion

• Broadly, this type of approach can be applied to
  any paralinguistic phenomena.
• Make no assumptions about the phenomena a priori
• Have subjects evaluate examples that are presumed
  to demonstrate the phenomena
• Analyze the examples, using subject ratings as
  dependent variable.

52
Conclusion - Future Work

• Further analysis of speech vs. transcription
  results
• TTS modification study.
• By modifying prosody of tokens can we make
  Lieberman charismatic? Sharpton uncharismatic?
• Repetition of the both studies with Palestinian
  Arabic political speech tokens.
• What are the similarities and differences between
  American and Palestinian notions of charisma?
• What lexical and acoustic/prosodic properties are
  displayed by charismatic Palestinian speech?