Emotional Speech

Document Sample
Emotional Speech Powered By Docstoc
					Deceptive Speech




        Frank Enos • April 25, 2005
Defining Deception

   Deliberate choice to mislead a target without
    notification (Ekman‗‘01)
   Often to gain some advantage
   Excludes:
     Self-deception
     Theater, etc.
     Falsehoods due to ignorance/error
     Pathological behaviors
Why study deception?

   Law enforcement / Jurisprudence
   Intelligence / Military / Security
   Business
   Politics
   Mental health practitioners
   Social situations
     Is it ever good to lie?
Why study deception?

   What makes speech ―believable‖?
   Recognizing deception means recognizing
    intention.
   How do people spot a liar?
   How does this relate to other subjective
    phenomena in speech? E.g. emotion,
    charisma
Problems in studying deception?

   Most people are terrible at detecting
    deception — ~50% accuracy
    (Ekman & O‘sullivan 1991, etc.)
   People use subjective judgments —
    emotion, etc.
   Recognizing emotion is hard
Problems in studying deception?

   Hard to get good data
     Real world
     Laboratory
   Ethical issues
     Privacy
     Subject rights
     Claims of success
   But also ethical imperatives:
     Need for reliable methods
     Debunking faulty methods
     False confessions
Frank Tells Some Lies

Maria: I‘m buying tickets to Handel‘s Messiah for me
  and my friends — would you like to join us?
Frank: When is it?
Maria: December 19th.
Frank: Uh… the 19th…
Maria: My two friends from school are coming, and
  Robin…
Frank: I‘d love to!
How to Lie (Ekman‘’01)

   Concealment
   Falsification
   Misdirecting
   Telling the truth falsely
   Half-concealment
   Incorrect inference dodge.
Frank Tells Some Lies

Maria: I‘m buying tickets to Handel‘s Messiah for me
  and my friends — would you like to join us?
Frank: When is it?
                                    • Concealment
Maria: December 19th.
                                    • Falsification
Frank: Uh… the 19th…                • Misdirecting

Maria: My two friends from school   • Telling the truth falsely
                                    • Half-concealment
  are coming, and Robin…
                                    • Incorrect inference dodge.
Frank: I‘d love to!
Frank Tells Some Lies

Maria: I‘m buying tickets to Handel‘s Messiah for me
  and my friends — would you like to join us?
Frank: When is it?
                                    • Concealment
Maria: December 19th.
                                    • Falsification
Frank: Uh… the 19th…                • Misdirecting

Maria: My two friends from school   • Telling the truth falsely
                                    • Half-concealment
  are coming.
                                    • Incorrect inference dodge.
Frank: Oh gee, I‘m having an
  appendectomy that night.
Reasons To Lie (Frank‘’92 )

   Self-preservation
   Self-presentation
   *Gain
   Altruistic (social) lies
How Not To Lie (Ekman‘’01)

   Leakage
     Part of the truth comes out
     Liar shows inconsistent emotion
     Liar says something inconsistent with the lie

   Deception clues
     Indications that the speaker is deceiving
     Again, can be emotion
     Inconsistent story
How Not To Lie (Ekman‘’01)

   Bad lines
       Lying well is hard
       Fabrication means keeping story straight
       Concealment means remembering what is omitted
       All this creates cognitive load  harder to hide emotion
   Detection apprehension (fear)
       Target is hard to fool
       Target is suspicious
       Stakes are high
       Serious rewards and/or punishments are at stake
       Punishment for being caught is great
How Not To Lie (Ekman‘’01)

   Deception guilt (vs. shame)
     Stakes for the target are high
     Deceit is unauthorized
     Liar is not practiced at lying
     Liar and target are acquainted
     Target can‘t be faulted as mean or gullible
     Deception is unexpected by target
   Duping delight
     Target poses particular challenge
     Lie is a particular challenge
     Others can appreciate liar‘s performance
Features of Deception

   Cognitive
     Coherence, fluency
   Interpersonal
     Discourse features: DA, turn-taking, etc.
   (Some addressed by Statement Analysis)
   Emotion
Describing Emotion

   Primary emotions
     Acceptance, anger, anticipation, disgust, joy,
       fear, sadness, surprise
   One approach:
       continuous dim. model (Cowie/Lang)
   Activation – evaluation space
   Add control/agency
   Primary E‘s differ on at least 2 dimensions of this
    scale (Pereira)
Problems With
Emotion and Deception

   Relevant emotions may not differ much on
    these scales
   Othello error
     People are afraid of the police
     People are angry when wrongly accused
     People think pizza is funny
   Brokow hazard
     Failure to account for individual differences
20th Century Lie Detection

   Polygraph
      http://antipolygraph.org
      The Polygraph and Lie Detection (N.A.P. 2003)
   Voice Stress Analysis
      Microtremors 8-12Hz
      Universal Lie response
      http://www.love-detector.com/
      http://news-info.wustl.edu/news/page/normal/669.html

   Reid
      Behavioral Analysis Interview
      Interrogation
Deception Experiments (Frank‘’92)
Addresses lying as dependent variable.


   Type and form of lie             Motive for Lying
     Concealment                      Self-preservation
     Falsification                    Self-presentation
     Misdirecting                     Gain
     Telling the truth falsely        Altruistic (social) lies
     Half-concealment
     Incorrect inference
      dodge.
Deception Experiments (Frank‘’92)
Addresses lying as dependent variable.


   Scenario
     *Topic of the lie: opinion; state; event.
     Stakes for lying / stakes for telling the truth.
     Interval between event and subject‘s account.
   Interpersonal structure
       Characteristics of the liar
       Characteristics of the target
       Presence or absence of a ―coach‖
       Presence or absence of others
The Good Old Days

   Mehrabian 1971:
      Nonverbal Betrayal of Feeling
Bulk of extant deception research…

   Not focused on verifying 20th century
    techniques
   Done by psychologists
   Considers primarily facial and physical cues
   ―Speech is hard‖
   Little focus on automatic detection of
    deception
Modeling Deception in Speech

   Lexical
   Prosodic/Acoustic
   Discourse
Deception in Speech (Depaulo ’03)

   Positive Correlates
     Interrupted/repeated words
     References to ―external‖ events
     Verbal/vocal uncertainty
     Vocal tension
     F0
Deception in Speech (Depaulo ’03)

   Negative Correlates
     Subject stays on topic
     Admitted uncertainties
     Verbal/vocal immediacy
     Admitted lack of memory
     Spontaneous corrections
Problems, revisited

   Differences due to:
     Gender
     Social Status
     Language
     Culture
Columbia/SRI/Colorado Corpus

   With Julia Hirschberg, Stefan Benus,
    Sarah Friedman, Sarah Gilman, and
    colleagues from SRI/ICSI and U. C. Boulder
   Goals
     Examine feasibility of automatic deception
      detection using speech
     Discover or verify acoustic/prosodic, lexical,
      and discourse correlates of deception
     Model a ―non-guilt‖ scenario
     Create a ―clean‖ corpus
Columbia/SRI/Colorado Corpus

   Inflated-performance scenario
   Motivation: financial gain
    and self-presentation
   32 Subjects: 16 women, 16 men
   Native speakers of Standard American English
   Subjects told study seeks to identify people who
    match profile based on ―25 Top Entrepreneurs‖
Columbia/SRI/Colorado Corpus

   Subjects take test in six categories:
     Interactive, music, survival, food,
       NYC geography, civics
   Questions manipulated 
     2 too high; 2 too low; 2 match
   Subjects told study also seeks people who can
    convince interviewer they match profile
     Self-presentation + reward
   Subjects undergo recorded interview in booth
     Indicate veracity of factual content of each utterance using
       pedals
CSC Corpus: Data

   15.2 hrs. of interviews; 7 hrs subject speech
   Lexically transcribed & automatically aligned 
                lexical/discourse features
   Lie conditions: Big Lie / Little Lie
   Segmentations (LT/LL):
       slash units (5709/3782), phrases (11,612/7108),
       turns (2230/1573)
   Acoustic features (± recognizer output)
CSC Corpus: Results

   Classification
    (Ripper rule induction, randomized 5-fold cv)
     Slash Units / Little Lies — Baseline 39.8% err
        Lexical & acoustic: 37.2 %; + subject dependent: 33.6%
     Phrases / Little Lies — Baseline 38.2% err
        Lexical & acoustic 34.0%; + subject dependent: 27.9%

   Other findings
       Positive emotion words deception (LIWC)
       Pleasantness  deception (DAL)
       Filled pauses  truth
       Some pitch correlation — varies with subject
Our Future Work

   Individual differences
     Wizards of deception
   Mark Frank Mock Theft Paradigm
   New paradigm
     Shorter
     Addition of personality test
     Higher stakes?

				
DOCUMENT INFO