1 Identifying Deceptive Speech within and across Cultures Sarah Ita Levitan, Guozhen An, Michelle Levine, Andrew Rosenberg, Julia Hirschberg Computer.

Slides:



Advertisements
Similar presentations
Identifying Deceptive Speech Across Cultures (FA ) PI: Julia Hirschberg (Columbia University) Co-PI: Andrew Rosenberg (CUNY) Co-PI: Michelle.
Advertisements

Punctuation Generation Inspired Linguistic Features For Mandarin Prosodic Boundary Prediction CHEN-YU CHIANG, YIH-RU WANG AND SIN-HORNG CHEN 2012 ICASSP.
Human Speech Recognition Julia Hirschberg CS4706 (thanks to John-Paul Hosum for some slides)
Impression Management You never get a second chance to make a first impression...
Spoken Cues to Deception CS 4706 What is Deception?
Automatic Prosodic Event Detection Using Acoustic, Lexical, and Syntactic Evidence Sankaranarayanan Ananthakrishnan, Shrikanth S. Narayanan IEEE 2007 Min-Hsuan.
Emotions and Voice Quality: Experiments with Sinusoidal Modeling Authors: Carlo Drioli, Graziano Tisato, Piero Cosi, Fabio Tesser Institute of Cognitive.
Interviews September 22, Questionnaires a. What it is/when to use them Types of Questionnaires group/individual open/closed a. Face-to-face (Utah.
Audiovisual Emotional Speech of Game Playing Children: Effects of Age and Culture By Shahid, Krahmer, & Swerts Presented by Alex Park
Who Can Best Catch a Liar? A Meta-Analysis of Individual Differences in Detecting Deception Michael G. Aamodt & Heather Mitchell Radford University Radford,
Critical Thinking.
Combining Prosodic and Text Features for Segmentation of Mandarin Broadcast News Gina-Anne Levow University of Chicago SIGHAN July 25, 2004.
Comparing American and Palestinian Perceptions of Charisma Using Acoustic-Prosodic and Lexical Analysis Fadi Biadsy, Julia Hirschberg, Andrew Rosenberg,
Presented by Ravi Kiran. Julia Hirschberg Stefan Benus Jason M. Brenier Frank Enos Sarah Friedman Sarah Gilman Cynthia Girand Martin Graciarena Andreas.
Prosodic Cues to Discourse Segment Boundaries in Human-Computer Dialogue SIGDial 2004 Gina-Anne Levow April 30, 2004.
Spoken Language Processing Lab Who we are: Julia Hirschberg, Stefan Benus, Fadi Biadsy, Frank Enos, Agus Gravano, Jackson Liscombe, Sameer Maskey, Andrew.
Extracting Social Meaning Identifying Interactional Style in Spoken Conversation Jurafsky et al ‘09 Presented by Laura Willson.
Deceptive Speech Frank Enos April 19, 2006 Defining Deception Deliberate choice to mislead a target without prior notification (Ekman‘’01) Often to gain.
Deceptive Speech Frank Enos April 25, Defining Deception Deliberate choice to mislead a target without notification (Ekman‘’01) Often to gain some.
On the Correlation between Energy and Pitch Accent in Read English Speech Andrew Rosenberg, Julia Hirschberg Columbia University Interspeech /14/06.
On the Correlation between Energy and Pitch Accent in Read English Speech Andrew Rosenberg Weekly Speech Lab Talk 6/27/06.
Classification of Discourse Functions of Affirmative Words in Spoken Dialogue Julia Agustín Gravano, Stefan Benus, Julia Hirschberg Shira Mitchell, Ilia.
9/5/20051 Acoustic/Prosodic and Lexical Correlates of Charismatic Speech Andrew Rosenberg & Julia Hirschberg Columbia University Interspeech Lisbon.
10/10/20051 Acoustic/Prosodic and Lexical Correlates of Charismatic Speech Andrew Rosenberg & Julia Hirschberg Columbia University 10/10/05 - IBM.
Varying Input Segmentation for Story Boundary Detection Julia Hirschberg GALE PI Meeting March 23, 2007.
Astronaut Selection International Space Station. Astronaut Selection –~ applications reviewed every other year- Next one will begin July 2007.
Chapter 13 pt. 2: Physiology of Emotion, Detecting Lies, and Experiencing Emotion.
Detecting deception A lie: a deliberate attempt by one person to mislead another No prior warning of this intent To detect a lie, we need to understand.
Toshiba Update 04/09/2006 Data-Driven Prosody and Voice Quality Generation for Emotional Speech Zeynep Inanoglu & Steve Young Machine Intelligence Lab.
Shriberg, Stolcke, Ang: Prosody for Emotion Detection DARPA ROAR Workshop 11/30/01 1 Liz Shriberg* Andreas Stolcke* Jeremy Ang + * SRI International International.
Acoustic and Linguistic Characterization of Spontaneous Speech Masanobu Nakamura, Koji Iwano, and Sadaoki Furui Department of Computer Science Tokyo Institute.
ActivePsych: Classroom Activities Project / Copyright © 2007 by Worth Publishers Emotional Expressivity Assessing outward display of emotions by themselves.
Schizophrenia and Depression – Evidence in Speech Prosody Student: Yonatan Vaizman Advisor: Prof. Daphna Weinshall Joint work with Roie Kliper and Dr.
THE BIG FIVE David Normansell.
A Study in Cross-Cultural Interpretations of Back-Channeling Behavior Yaffa Al Bayyari Nigel Ward The University of Texas at El Paso Department of Computer.
On Speaker-Specific Prosodic Models for Automatic Dialog Act Segmentation of Multi-Party Meetings Jáchym Kolář 1,2 Elizabeth Shriberg 1,3 Yang Liu 1,4.
1 Deceptive Speech CS4706 Julia Hirschberg 2 Everyday Lies Ordinary people tell an average of 2 lies per day –Your new hair-cut looks great. –I’m sorry.
CS 6998 Computational Approach to Emotional Speech Instructor: Prof. Julia Hirschberg Columbia University 12/21/2009 Julia’s Little Helper : A Real-time.
Evaluating prosody prediction in synthesis with respect to Modern Greek prenuclear accents Elisabeth Chorianopoulou MSc in Speech and Language Processing.
Turn-taking Discourse and Dialogue CS 359 November 6, 2001.
Graham Davies Week 5 Detecting Deception in Witnesses and Suspects.
1 Robust Endpoint Detection and Energy Normalization for Real-Time Speech and Speaker Recognition Qi Li, Senior Member, IEEE, Jinsong Zheng, Augustine.
1 Computation Approaches to Emotional Speech Julia Hirschberg
Predicting Student Emotions in Computer-Human Tutoring Dialogues Diane J. Litman&Kate Forbes-Riley University of Pittsburgh Department of Computer Science.
Recognizing Discourse Structure: Speech Discourse & Dialogue CMSC October 11, 2006.
Voice Activity Detection based on OptimallyWeighted Combination of Multiple Features Yusuke Kida and Tatsuya Kawahara School of Informatics, Kyoto University,
Predicting Voice Elicited Emotions
1/17/20161 Emotion in Meetings: Business and Personal Julia Hirschberg CS 4995/6998.
Brian Lukoff Stanford University October 13, 2006.
Control of prosodic features under perturbation in collaboration with Frank Guenther Dept. of Cognitive and Neural Systems, BU Carrie Niziolek [carrien]
Phone-Level Pronunciation Scoring and Assessment for Interactive Language Learning Speech Communication, 2000 Authors: S. M. Witt, S. J. Young Presenter:
Acoustic Cues to Emotional Speech Julia Hirschberg (joint work with Jennifer Venditti and Jackson Liscombe) Columbia University 26 June 2003.
Presented By Meet Shah. Goal  Automatically predicting the respondent’s reactions (accept or reject) to offers during face to face negotiation by analyzing.
On the role of context and prosody in the interpretation of ‘okay’ Julia Agustín Gravano, Stefan Benus, Julia Hirschberg Héctor Chávez, and Lauren Wilcox.
1 Deceptive Speech: Acoustic, Prosodic Lexical, Gender, Ethnicity, and Personality Factors Julia Hirschberg Computer Science Columbia University 14 June.
Dean Luo, Wentao Gu, Ruxin Luo and Lixin Wang
Intonation and Computation: Personality
Hyperscanning Deception: EEG Analysis of the Lying Game
Intonation and Computation: Deception
Studying Intonation Julia Hirschberg CS /21/2018.
Automatic Fluency Assessment
Comparing American and Palestinian Perceptions of Charisma Using Acoustic-Prosodic and Lexical Analysis Fadi Biadsy, Julia Hirschberg, Andrew Rosenberg,
Fadi Biadsy. , Andrew Rosenberg. , Rolf Carlson†, Julia Hirschberg
Liz Shriberg* Andreas Stolcke* Jeremy Ang+ * SRI International
Towards Automatic Fluency Assessment
Emotional Speech Julia Hirschberg CS /16/2019.
Low Level Cues to Emotion
Acoustic-Prosodic and Lexical Entrainment in Deceptive Dialogue
Guest Lecture: Advanced Topics in Spoken Language Processing
Automatic Prosodic Event Detection
Presentation transcript:

1 Identifying Deceptive Speech within and across Cultures Sarah Ita Levitan, Guozhen An, Michelle Levine, Andrew Rosenberg, Julia Hirschberg Computer Science Columbia University

2 Collaborators Zoe Baker-Peng, Lingshi Huang, Leighanne Hsu, Melissa Kaufman-Gomez, Yvonne Missry, Elizabeth Petitti, Sarah Roth, Molly Scott, Jennifer Senior, Grace Ulinski, Christine Wang, Mandi Wang

3 Our Definition of Deception Deliberate choice to mislead –Without prior notification –To gain some advantage or to avoid some penalty Not: –Self-deception, delusion, pathological behavior –Theater –Falsehoods due to ignorance/error Everyday (white) Lies hard to detect But Serious Lies?

4 Why are ‘serious’ lies difficult? Hypotheses: –Our cognitive load is increased when lying because… Must keep story straight Must remember what we’ve said and what we haven’t said –Our fear of detection is increased if… We believe our target is hard to fool or suspicious Stakes are high: serious rewards and/or punishments Makes it hard for us to control indicators of deception

5 Cues to Deception: Current Proposals –Body posture and gestures (Burgoon et al‘94) Complete shifts in posture, touching one’s face,… –Microexpressions (Ekman‘76, Frank‘03) Fleeting traces of fear, elation,… –Biometric factors (Horvath‘73) Increased blood pressure, perspiration, respiration…other correlates of stress Odor –Changes in brain activity: true vs. false stories –Variation in what is said and how (Adams‘96, Pennebaker et al‘01, Streeter et al‘77)

6 Current Approaches to Deception Detection Training humans –John Reid & Associates Behavioral Analysis: Interview and Interrogation Laboratory studies: Production and Perception `Automatic’ methods –Polygraph –Nemesysco and the Love DetectorLove Detector –No evidence that any of these work….but publishing this statement can be dangerous! (Anders Eriksson and Francisco La Cerda)but publishing this statement can be dangerous!

7 What’s Missing? More objective, experimentally verified studies of cues to deception which predict better than humans or polygraphs Method: –Identify acoustic, prosodic, and lexical cues that can be extracted automatically as well as simple personality features –Examine statistical correlations with deception –Use Machine Learning techniques to train models to classify deceptive vs. non-deceptive speech

8 Our Previous Work on Deception Created Columbia/SRI/Colorado Deception Corpus –Within-subject recordings of deceptive and non-deceptive speech 32 adult native American English speakers 25-50m interviews by trained interviewer (Reid technique): 15.2h of speech, 7h from subjects Subjects given tasks and incentivized to lie about performance Ground truth identified by subjects

9 Acoustic/Prosodic Features Duration features –Phone / Vowel / Syllable Durations –Normalized by Phone/Vowel Means, Speaker Speaking rate features (vowels/time) Pause features (cf Benus et al ‘06) –Speech to pause ratio, number of long pauses –Maximum pause length Energy features (RMS energy) Pitch features –Pitch stylization (Sonmez et al. ‘98) –Model of F0 to estimate speaker range –Pitch ranges, slopes, locations of interest Spectral tilt features

10 Lexical Features Presence and # of filled pauses Is this a question? A question following a question Presence of pronouns (by person, case and number) A specific denial? Presence and # of cue phrases Presence of self repairs Presence of contractions Presence of positive/negative emotion words Verb tense Presence of ‘yes’, ‘no’, ‘not’, negative contractions Presence of ‘absolutely’, ‘really’ Presence of hedges Complexity: syls/words Number of repeated words Punctuation type Length of unit (in sec and words) # words/unit length # of laughs # of audible breaths # of other speaker noise # of mispronounced words # of unintelligible words

11 Subject-Dependent Features % units with cue phrases % units with filled pauses % units with laughter Lies/truths with filled pauses ratio Lies/truths with cue phrases ratio Lies/truths with laughter ratio Gender

12

13 Results 88 features, normalized within-speaker –Discrete: Lexical, discourse, pause –Continuous features: Acoustic, prosodic, paralinguistic, lexical Best Performance: Best 39 features + c4.5 ML –Accuracy: 70.00% –TRUTH F-measure: Lexical, subject-dependent & speaker normalized features best predictors Interesting individual differences

14 Evaluation: Compared to Human Deception Detection Most people are very poor at detecting deception –~50% accuracy (Ekman & O’Sullivan ‘91, Aamodt ‘06) –People use unreliable cues, even with training Our study –32 Judges, rating 2 interviews –Received ‘training’ on one subject. Pre- and post-test questionnaires Personality Inventory

15 A Meta-Study of Human Deception Detection (Aamodt & Mitchell 2004) Group#Studies#SubjectsAccuracy % Criminals Secret service Psychologists Judges Cops Federal officers Students1228, Detectives Parole officers

16 What Makes Some People Better Judges? Costa & McCrae (1992) NEO-FFI Personality Measures –Extroversion (Surgency). Includes traits such as talkative, energetic, and assertive. –Agreeableness. Includes traits like sympathetic, kind, and affectionate. –Conscientiousness. Tendency to be organized, thorough, and planful. –Neuroticism (opp. of Emotional Stability). Characterized by traits like tense, moody, and anxious. –Openness to Experience (aka Intellect or Intellect/Imagination). Includes having wide interests, and being imaginative and insightful.

17 By Judge 58.2% Acc. By Interviewee 58.2% Acc.

18 Neuroticism, Openness & Agreeableness Correlate with Judge’s Performance On Judging Global lies.

Current Study: Hypotheses Peronality factors can help to predict differences in deceptive behavior Subjects who deceive better can also detect deception better Cultural differences and gender play a role in deceptive behavior and in deception detection abilities New task: Studies of pairs of American English and Mandarin Chinese native speakers, speaking English, interviewing each other 19

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 20

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 21

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 22

Biographical Questionnaire

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 24

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 25

NEO-FFI O C E A N openness to experience conscientiousness extraversion agreeableness neuroticism 26

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 27

Experiment 28

Experiment Background survey Biographical Questionnaire Baseline sample NEO-FFI Lying game Survey 29

Scoring and Motivation Success –Ability to lie -> as interviewee, number of lies believed true by interviewer –Ability to detect lies -> as interviewer, number of correct guesses for truth and lie Note: $1 added or subtracted for each right or wrong decision 30

Example: “Where were you born?” True or False? 31

Example: “Where were you born?” False! 32

Current Corpus ● 126 subject pairs ● ~120 hours of speech ● Pair types: 33 EnglishChinese English/ Chinese

Annotation Transcribed using Amazon Mechanical Turk –5 Turkers per utterance, combined using Rover techniques Automatically segmented using Praat into Inter- Pausal Units (IPUs) at 50ms silence Automatically aligned with speech using aligner built with Kaldi 34

Rover Example ROVER Score = ( /3+2/ /3+1+1)/10=0.9 Index itsreallyfunumIgoliketoaplace 2itsreallyfunigolikeaplace 3it’sreallyfunumIgoliketoaplace Rover output itsreallyfunumIgoliketoaplace score1112/

Statistical Results: Deception Detection People’s ability to detect deception correlates with their ability to deceive r(252) = 0.13, p = 0.04 Holds across all subjects but –Strongest for females r(126) = 0.26, p = –No difference for English and Chinese females 36

Personality & Ability to Deceive Extraversion significantly negatively correlated –English/Male r(68) = -0.25, p = 0.04 Tendencies: –Chinese/female extroversion positively correlated with ability to deceive – American/female conscientiousness negatively correlated with ability to deceive 37

Personality & Deception Detection No effect of personality factors at all –Contra earlier findings for English speakers (Enos et al ’06) 38

Confidence in Judgments Ability to detect deception negatively correlates with confidence in judgments r(250) = -0.14, p = 0.03 –Holds across all subjects but strongest for females r(126) = -0.24, p = 0.01 Neuroticism negatively correlates with confidence for female/Chinese subjects r(68) = -0.27, p = 0.02 Openness to experience negatively correlates for females r(126) = -0.21, p = 0.02 –Strongest for Female/Chinese r(68) = -0.29, p =

Classification Results Features: –Acoustic Features –Gender –Ethnicity –Personality scores 40

ClassPrecisionRecallF1 T F Wtd Avg J48 with Speaker Normalized Data J48 with Raw and Speaker Normalized Data ClassPrecisionRecallF1 T F Wtd Avg Initial Classification Results

Future Work Many more classification experiments –Lexical and subject-dependent features –More acoustic-prosodic features: voice quality, speaking rate, other duration features –Examining entrainment as a factor Delivering our data to Air Force researchers in the fall 42

Publications: – S. I. Levitan, M. Levine, J. Hirschberg, N. Cestero, G. Ahn, A. Rosenberg, "Individual Differences in Deception and Deception Detection," Cognitive 2015, Nice. (Best Paper Award) Many presentations 43

Thank you!