BRIMS ‘05 The “Etiquette Quotient”: An Approach to Believable Social Interaction Behaviors Christopher A. Miller Marc Chapman Peggy Wu Lewis Johnson Simulations for Cross Cultural Training Military forces increasingly need training in the cultures they deploy to (esp. for urban ops) Cross Cultural training helps, but is a huge drain on resources and time ¾ 200 soldiers trained for Iraq deployment (Mares, 2003) Soldiers willingly spend off-duty time playing PC games Solution: Make games that teach elements of culture ¾ Estimated DoD-wide training savings of >$1B/year (Chatham, 2003) Problem: Avatar behaviors are hand-scripted, hence brittle and unbelievable, impossible to change culture Solution: A general, computational social interaction model with “pluggable” culture knowledge modules 21 June 2005 2 Approach Summary We know that humans react socially (according to human conventions) to complex automation ¾ Reeves and Nass We have access to a theory/model of how humans decide what politeness behaviors to exhibit ¾ Brown and Levinson We have evidence that the model predicts human reactions to human-machine interactions So let’s develop a metric of believable politeness based on Brown and Levinson 21 June 2005 3 Reeves and Nass Reeves, Byron & Nass, Clifford (1996). The Media Equation: How People Treat Computers, Television and New Media Like Real People and Places. Cambridge University Press. Media Equation: Media = Real Life, or perhaps, People ÅÆ Media = People ÅÆ Real Life Paradigm: Well-established phenomenon from human-human interaction; substitute a computer for one actor. E.g., ¾ Evaluations more positive to actor’s face than to third party ¾ Believe and favor a flatterer, even when flattery known to be false Implication: Humans are equipped with schema for interaction with intelligent social agents; complex machines trigger schema unless we fight to counteract 21 June 2005 4 Etiquette is … “… the defined roles and acceptable behaviors or interaction moves of each participant in a common ‘social’ setting … Etiquette rules create an informal contract between participants in a social interaction allowing expectations [and interpretations] to be formed and used about the behavior of others.” --(from HCE Symposium description) “…(1) the body of prescribed social usages. (2) Any special code of behavior … : ‘In the code of military etiquette, silence and fixity are forms of deference’ (Ambrose Bierce). … Synonyms: propriety, decorum, protocol.” --American Heritage Dictionary Etiquette enables expectations about social interactions; These expectations permit judgments about politeness and its components; Expectation violations either demand (re-)interpretations or are “unbelievable” Etiquette and politeness obey culturally-generic rules, but manifest themselves in culturally specific ways. 21 June 2005 5 Brown and Levinson, 1986 Politeness strategies as universal in human-human interactions ¾ They are NECESSARY to establish intent & power relationships As means of diffusing Face Threatening Actions As linguistic hedges to Grice Degree of threat in FTA is ¾ F (PH:S, DS&H, Ri) where: ¾ PH:S = Power of Hearer over Speaker ¾ DS&H = Social distance between H&S ¾ Ri = Ranked Imposition of the act Redress ≅ Face Threat 21 June 2005 6 B&L’s Universal Politeness Strategies 1. w/o Redress, baldly Increased FTA Risk On record Do the FTA w/ Redress 2. Positive Politeness 3. Negative Politeness 4. Off record 5. Don’t do the FTA 21 June 2005 7 Negative Politeness Strategies from B&L Be direct Be direct clash Don’t presume/assume Make minimal assumptions about H’s wants 1. Be conventionally indirect 2. Question, hedge Be indirect Don’t coerce Give H option not to act Do FTAx (a) on record (b) plus redress to H’s want to be unimpinged upon Don’t assume H willing/able to do act Assume H unlikely to do act 3. Be pessimistic 4. Minimize the imposition, Rx Minimize threat Make explicit R,P,D values 5. Give deference 6. Apologize Communicate S’s want not to impinge on H 7. Impersonalize: avoid pronouns I and you Dissociate S, H from the particular infringement 8. State the FTA as a general rule 9. Nominalize Redress other wants of H, derivative from negative face 10. Go on record as incurring a debt, or as not indebting H 21 June 2005 8 Adapting B&L 1 B&L are attempting to explain/account for Action Production They note that P,D, and R affect redressive strategy usage– and that these factors are culturally informed They note that redressive strategies themselves are culturally-specific (though there are universal abstractions) We infer a “Character” variable– the predisposition of an individual to subvert his/her own face to others D(S,H) P(H,S) Rx Culture-specific Markers C(S) Ax Culture-specific redressive strategies 21 June 2005 9 Adapting B&L 2 Social Distance Power Difference Imposition D(S,H) Character P(H,S) Rx C(S) Face Threat (Wx) Elements of an interaction Redressive Actions (Ax) Face Threat (Wx) For Normal/Believable/Unremarkable Interactions: Wx = Ax 21 June 2005 10 Adapting B&L 3 We need to account for interpretations of believability of observed redressive strategies Context History Assumed D(S,H), P(H,S), Rx, C(S) Culture-specific Markers Expected Redress in Ax Culture-specific redressive strategies “Believable” Revisions to D(S,H), P(H,S), Rx, C(S) “Unbelievable” Observed Redress in Ax Match? Goal of Phase I: an Etiquette Quotient ¾ ¾ ¾ A Believability Metric based on the degree to which Face Threat = Redressive Actions Assumption: “Believable” means expected redress in context Unexpected Redress may mean “unbelievable” or reinterpretation of context 21 June 2005 11 The Grand Vision Believability Metric could be used for simple scoring/evaluation Much more interesting/useful to use for avatar behavior recognition and generation “Culture Modules” to swap in and out to give avatars culture-specific etiquette sensitivities and reactions Familiarity Level (D) Power Level (P) Character Selected Politeness/ Redress Level Selected Actions for Redress Imposition Level (R) Culture-specific Knowledge Base 21 June 2005 12 Independent LifeStyle Assistant (ILSA) A NIST ATP Program 21 June 2005 Alternate Reminder Wording Face Threat ≈ Impoliteness Alternate presentations for a Med-Advisor 5. You’ve missed a dose of medication. Take your medication now. 3. Your health is important. It looks like you’ve missed a dose of medication you wanted me to check on. Why don’t you take your medication now. 2. I’m sorry, but Med-Advisor hasn’t detected you taking your medication scheduled for <time>. If you haven’t taken it, could you please take it now? 1. This is Med-Advisor calling to remind you that your health is important. 4. You’ve missed a dose of medication that was scheduled for <time>. Bald Pos. Polite Neg. Polite Off Record Used (Pos. Polite/Bald) 21 June 2005 14 Experiment Conditions Subjects: ¾ ¾ ¾ Elder’s with no Med-Advisor experience Nominals asked about Med-Advisor Med-Advisor engineers 21 June 2005 15 Perceived Impoliteness Impolitness Ratings 6.00 5.00 4.00 Nominals-Tech Engineers 3.00 Elders B&L's Prediction 2.00 1.00 0.00 A. Bald B. Pos. Polite C. Neg Polite D. Off Record E. Used (Pos + Bald) 21 June 2005 16 Perceived Inappropriateness Inappropriateness Ratings 6.00 5.00 4.00 Nominals-Tech Engineers 3.00 Elders 2.00 B&L's Predictions 1.00 0.00 A. Bald B. Pos. Polite C. Neg Polite D. Off Record E. Used (Pos + Bald) 21 June 2005 17 Phase I Plans 1. Identify high priority contexts/interactions for training Scenarios and Dimension variations 2. Develop predictive model for scoring interaction believability Formalize B&L for observations 3. Implement Model 4. Design experiment, including experimental stimuli using Avatars ¾ Micro interactions in TLTS 5. Conduct Preliminary Experiment 6. Analyze and Report 21 June 2005 18 USC’s Tactical Language Training System 21 June 2005 19 Payoffs and Future Work A completed and verified model would provide: ¾ ¾ ¾ ¾ Metric for scoring believability of actions between two social actors Ability to predict believability of potential actions Ability to adapt an avatar’s behavior to be appropriate to social interaction context (P,D & R initially and over time and context) Starting place to create culture-specific believability shaping and scoring algorithms Resulting in a computational, rather than scripted approach to generating and tracking avatar social interaction behaviors ¾ ¾ ¾ More rapid, less costly behavior generation Therefore, more diverse avatars And more robust, less brittle interaction behavior 21 June 2005 20
© Copyright 2026 Paperzz