Marital Agent Trait-Based Emotion System

1
Marital Agent Trait-BasedEmotion System

• System collects information about a pre-marital
  couple.
• Use questionnaires to determine
• Personality- IPIP NEO
• Relationship concerns (Gottman)
• Goals with emotion vectors
• Model couple in negotiation

2
Idea What if we programmed an agent to act like
a person in a social situation? Could a person
learn something valuable by seeing his
behavior? Could a person benefit by replaying the
situation using a new set of behaviours?
3
IPIP Personality Survey (Goldberg)
4
IPIP-NEO Narrative Report (Validated Survey)

• EXTRAVERSION 94
• Gregariousness 91
• Assertiveness 82
• Activity Level 97
• Excitement-Seeking 49
• Cheerfulness 80
• AGREEABLENESS 81
• Trust 87
• Morality 63
• Altruism 88
• Cooperation 73
• Modesty 25
• Sympathy 77

Computed from 120 questions Includes
explanation of Traits
5
IPIP-NEO Narrative Report (cont)

• NEUROTICISM 6
• Anxiety 0
• Anger 16
• Depression 16
• Self-Consciousness 39
• Immoderation 8
• Vulnerability 29
• OPENNESS TO EXPERIENCE 35
• Imagination 65
• Artistic Interests 18
• Emotionality 83
• Adventurousness 9
• Intellect 86
• Liberalism 4
• CONSCIENTIOUSNESS 84
• Self-Efficacy 96
• Orderliness 36
• Dutifulness 87
• Achievement-Striving 83

6
Interaction Style Questions John Gottmans Work

• I soften the conflict by constructively focusing
  on feelings first (i.e., "This is how I feel
  when...") and then moving on to the specific
  issue or complaint.
• I focus on one specific issue at a time and seek
  to resolve it before moving on to another issue
  so the conflict doesn't degenerate into a
  mudslinging contest.
• I focus on the present issue rather than bring up
  issues from the past as weapons to use in an
  attempt to gain power and control over the
  situation or another person.
• I think win/win and understand that if one person
  loses the argument, then both people in the
  relationship lose.
• I soothe my partner through speaking
  non-defensively, validating his or her
  perceptions and feelings, or by using humor.
• I seek to resolve the specific issue as soon as
  possible in order to avoid experiencing ongoing
  resentment, frustration, or hurt feelings.

7
Relationship Goal Questionnaire(Validity
Untested)

• I get satisfaction in making my partner happy.
• I get satisfaction in controlling our
  relationship.
• Getting my own way is important.
• Having a positive interaction is important to me.
• Being validated is important to me.

8
Program Agents with Personality and Emotion

• Bob and Alice are considering marriage.
• Evaluate their personalities
• Agent Bob and Agent Alice
• Give Agents a problem and view how they negotiate.

9
Goal Create better communication
My wife and I had words, but I never got to use
mine. -Fibber McGee
10
Goal Create more realistic expectations

• Marital happiness is a function of both
  expectation and actual relationship quality.

11
Marital Research

• How a couple differs is not so important (as
  there will always be differences).
• What is important is how they deal with those
  differences.
• This research seeks to expose differences.

12
Several studies suggest

• Researchers can predict which marriages will end
  in failure from information gathered before the
  couple marries.
• Tell people if they are at substantially greater
  risk for divorce
• Told couples argue most about children and money,
  but some believe how they argue is most important.

13
(No Transcript)
14
Planner Generates Interaction

• Simple, hierarchical ordered planner
• Implemented in Prolog
• An agent calls the planner with a list of goals
  to obtain the plans.
• The agent calls the prolog function
• plan (a, b, c, , Plans)
• 1st argument a, b and c are the goals.
• 2nd argument specifies the initial list of plans
  to start with
• 3rd argument Accumulator for the resulting plans.

15
Who determines the starting goal-list on which
the planner works to generate plans?

• The human user (useful for testing agents
  behavior on several kinds of goals)
• An automatic goal initializer
• a simple piece of prolog code
• evaluates the preferences

16

• Three phases of planner
• Environment Setting phase
• Fact retrieval phase
• Presentation planning phase

17
Dynamic re-planning

• The basic idea is to plan for many expected
  situations and when an unexpected situation
  arises, modify the database and re-plan.
• Possible extensions to planner Make it
  stochastic to simulate human indeterminism.
  Probabilities are determined by personality,
  emotion, and history.

18
Agent Database

• For each agent
• Goals
• Beliefs
• History
• Personality
• Miscellaneous

19
Decomposing goal - precondition

• showPower(Proposal, Activity, ActivityName,
  ActivityActors, ActivityType, Time),
• logic(X), X lt logicCutoff,
• polite(Y), Y lt politeCutoff,
• likes(Activity),
  enjoys(ActivityType),
• not(partnerLikes(Activity)),
• not(partnerEnjoys(ActivityType)),
• activityName(Activity,ActivityName
  ),
• activityActors(ActivityName,Activi
  tyActors),
• free(Time)
  
• ,

20
Plan - components

• propose(Activity, ActivityName,
• ActivityActors, ActivityType, Time),
• acknowledge(accept(Activity, ActivityName,
• ActivityActors, ActivityType,
  Time))
• accuse(reject(Activity, ActivityName,
• ActivityActors, ActivityType, Time))
• ).

21
Express interaction patterns as regular
expression

• propose (reject cope) (accept react)?
• zero or more occurrences
• ? zero or one occurrences

22
Express interaction as stochastic context free
grammar

• Used as a generator.
• Grammar to control options
• Stochastic to give probability to actions.
• Probability depends on history, personality,
  interaction patterns.
• Dynamically evaluated

23
A very simple plan might look like

• // Environment setting phase
• setHonesty(minHonestyValue),
• setAffection(minAffectionValue),
• setPride(maxPrideValue),
• // Fact Retrieval phase
• getFreeTimes(self, FreeTimes),
• getFreeTimes(partner, FreeTimes)
• getActivity(Activity, FreeTimes),
• not PartnerLikes(Activity),
• // Presentation planning phase
• propose(Activity),
• acknowledge(accept(propose(Activity)),
• accuse(reject(propose(Activity)),
• replan(other).

24
Interaction Patterns

• speaker/listener (take roles)
• criticism
• defensiveness
• contempt
• stonewalling (listener withdrawal emotionally and
  perhaps physically)
• kitchen sink (prior complaints brought up)

25
Emotions to expression

• Emotions passed as an internal form.
• complete range of values
• no need to parse expression for meaning
• can filter so not transparent
• Expressions are generated for GUI
• Difficulty in mapping large number of emotions
  into expression.

26
Mapping rejection phrases

• Sorry.
• I cant.
• I cant. Maybe some other time
• Id rather not.
• No.
• I wont
• Absolutely not.
• That is ridiculous. I wont consider it.

27
Mapping - motivations

• Consider various reasons for saying no
• Conflict.
• Possible conflict
• No interest in going to event.
• Too busy.
• Anger over other rejections
• Feel person is inconsiderate

28
Mapping personalities

• How does personality affect answer?
• aggressive
• trusting
• cooperative
• cautious
• depressed
• anxious
• angry

29
In theory, mapping
Personality
Result
History
output expression
Plan
Very complex mapping
30
Regular expression

• Rejection reject explanation judgment
  soften? counter-proposal?
• one or more
• zero or more
• ? zero or one
• Repetition determined by parameters

31
Explanation

• because of something you have done (Im mad at
  you, I dont want to spend time with you, I would
  rather be with my friends/family)
• because it is (ridiculous, dumb, self-centered,
  unworkable)
• because of some situation.
• because I have a conflict
• because of prior history
• I am so sorry.
• Maybe another time
• I am just too busy for you.
• I am just too busy. I should make time for my
  good friends.

32
Judgment

• You have such good ideas.
• You are so thoughtful to have asked.
• That does sound fun.
• You always want to do things I dont like.
• You never consider my feelings.
• Why did you think I would want to do that?

33
Softening

• You have so many good traits
• Ive heard really good things about you
• I remember when we had a good time together.

34
Counter proposal

• Maybe another time/day
• Maybe another activity
• Maybe we should do some other thing I know you
  like.
• Maybe we should do something we both like.
• Maybe we should do something only I like.

35
Grammar may be viewed as a stochastic finite
state machine
Explanation
reject
Judgment
reject
soften
counter

• Rejection reject explanation judgment
  soften? counter-proposal?

36
Thus, responses might vary from

• No. Maybe another time
• I am so sorry. You have such good ideas. Maybe
  we should go bowling. Maybe we should go golfing.
  Maybe we should go tomorrow. Maybe we should go
  Friday.
• No. No. No. I wont. I am too busy for you. You
  never consider my feelings.

37
Modeling Emotions

• Emotions are important in giving Disney
  characters the illusion of life.
• Believability vs realism may be better to use
  simplified, exaggerated characters.

38
How to Combine Emotions

• Winner take all ignore all but the highest
  intensity emotion
• Additive but may be confusing to model joy and
  sadness simultaneously
• Logarithmic log(2emotion1 2emotion2)
• Focus kicking example

39
How created?

• Emotions are tied to goals (through personality
  survey). When a goal is achieved, attached
  emotions are generated.
• Factors surprise, importance of goal, difference
  in emotion felt with success or failure of same
  goal. (e.g., goal to have companion)

40
Goals

• Intensity
• Chance of succeeding
• Emotions generated when fail
• Emotions generated when chance of succeeding
  increases/decreases.

41
What kind of transformations?

• Decay all at same rate?
• Combine
• Filter
• Idea create an algebra of emotions through
  matrix manipulation

42
What effects emotions?

• Personality each personality type will express
  emotions in its own way.
• relationships affect what emotions are felt and
  how strongly
• memory previous experiences (Were you angry
  when the first telemarketer called?)

43
Challenges

• Cardboard personalities?
• How create grammar?
• IPIP survey validated
• Gottman research well-respected, but is it valid
  for self-reporting?
• Goal data unclear what goals to even ask about
• Held to a higher standard not just
  entertaining.
• How do we test it? (subjective tests?)

44

• After a quarrel, a husband said to his wife, You
  know, I was a fool when I married you. The wife
  replied, Yes dear, but I was in love and didnt
  notice.

• Much testing is needed.