1. Bob Marinier John Laird University of Michigan Electrical Engineering and Computer Science August 2, 2007

2. Introduction Want to build computational models of emotion, mood and feeling Computational models need to fill in a lot of blanks that are not addressed by most theories How are emotion, mood and feeling represented? How do they interact? On what principles do we answer these questions? Existing computational models lack principled models of the relationship between emotion, mood, and feeling 2

3. Introduction Want to build computational models of emotion, mood, and feeling Need specific algorithms and data structures for representing and manipulating these Existing computational models propose reasonable solutions But little attempt to define reasonable We present a more comprehensive theory of the integration of emotion, mood, and feeling We present explicit criteria for evaluating models of integration Human data would be best, but isnt available We propose functional, simple criteria We apply these criteria by building on existing models and suggesting actual functions 3

4. Appraisal theories Idea: Humans evaluate a situation with respect to their goals along a number of innate dimensions Novelty, Goal Relevance, Causality, Conduciveness Appraisals trigger emotional responses Mapping between appraisal values and emotions is fixed 4

5. Appraisals to emotions Scherer 2001JoyFearAnger SuddennessHigh/mediumHigh UnpredictabilityHigh Intrinsic pleasantnessLow Goal/need relevanceHigh Cause: agentOther/natureOther Cause: motiveChance/intentionalIntentional Outcome probabilityVery highHighVery high Discrepancy from expectationHigh ConducivenessVery highLow ControlHigh PowerVery lowHigh 5

6. Relationship between emotion, mood, and feeling Emotion: Result of appraisals Is about the current situation Mood: Average of recent emotions Provides historical context Feeling: Emotion + Mood What agent actually perceives 6

7. Representation of emotion, mood, and feeling Use a frame that contains the current value of each appraisal dimension (Gratch & Marsella 2004) Since appraisal-to-emotion mapping is fixed, this frame can represent the emotion For simplicity, use appraisal frames to represent emotion, mood, and feeling 7

8. Appraisal frame representation Values are represented on one of two scales [0,1] : Dimension has endpoints that correspond to low and high intensity E.g., Suddenness [-1,1] : Dimension has endpoints that correspond to high intensity, with midpoint of low intensity E.g., Conduciveness 8

9. Appraisal frame representation Scherer 2001Range Suddenness[0,1] Unpredictability[0,1] Intrinsic pleasantness[-1,1] Goal/need relevance[0,1] Cause: agent (self)[0,1] Cause: agent (other)[0,1] Cause: agent (nature)[0,1] Cause: motive (intentional)[0,1] Cause: motive (negligence)[0,1] Cause: motive (chance)[0,1] Outcome probability[0,1] Discrepancy from expectation[0,1] Conduciveness[-1,1] Control[-1,1] Power[-1,1] 9

10. Cognition Emotion Mood Feeling Combination Function Pull (10% per cycle) Decay (1% per cycle) Active Appraisals Perceived Feeling Interaction between emotion, mood, and feeling 10

11. Criteria for combining emotion and mood Neal Reilly (1996, 2006) developed the basis for many of these criteria Assumption: Dimension Independence Can compute combination of each dimension in the frame separately v feeling = C(v mood, v emotion ) Output must fall in [0,1] or [-1,1] range 11

12. Criteria for combining emotion and mood Distinguishability of inputs Dont want a large range of inputs to map to a small range of outputs The agent wouldnt be able to distinguish between the inputs, and thus couldnt form diverse responses 12

13. Criteria for combining emotion and mood Limited range: Avoid going out of scale as much as possible Averaging doesnt make sense Example: If mood is one of mild conduciveness, and emotion is of strong conduciveness, feeling should be of stronger, not weaker conduciveness Output should be between the input with the maximum magnitude and the sum of the inputs 13

14. Criteria for combining emotion and mood Non-linear Consider these examples C(0.5, 0.5) = ? C(0.8, 0.9) = ? C(0.5, 0.9) = ? If sum the inputs, then first two are not distinguishable If max the inputs, then the last two are not distinguishable Relationship may be logarithmic 14

15. Criteria for combining emotion and mood Symmetry Emotion and mood contribute equally to feeling We have no reason to assume the function is symmetrical, but it seems like a reasonable place to start Symmetry around 0 C(x, 0) = C(0, x) = x C(0, 0) = 0 Symmetry of opposite values C(x, -x) = 0 Symmetry of all values C(x, y) = C(y, x) 15

16. Combination function Good Limited range Non-linear Problems Not centered at zero: C(0,0) = 0.069 Doesnt work with negative values (not symmetrical) 16

17. Combination function Good Limited range Non-linear Centered at zero Problems Doesnt work with negative values (not symmetrical) 17

18. Combination function Good Limited range Non-linear Symmetrical Problems Distinguishability of inputs C(-0.1, 0.9) = 0.899979 C(-0.5, 0.9) = 0.898164 18

19. Combination function Good Distinguishability of inputs C(-0.1, 0.9) = 0.854532 C(-0.5, 0.9) = 0.585615 Limited range Non-linear Symmetrical 19

20. Example 20 EmotionMoodFeeling Suddenness [0,1]0.235 Unpredictability [0,1].250.400.419 Intrinsic-pleasantness [-1,1]0-.235 Goal-relevance [0,1].750.222.750 Causal-agent (self) [0,1]000 Causal-agent (other) [0,1]000 Causal-agent (nature) [0,1]1.6601 Causal-motive (intentional) [0,1]000 Causal-motive (chance) [0,1]1.6601 Causal-motive (negligence) [0,1]000 Outcome-probability [0,1].750.516.759 Discrepancy [0,1].250.326.362 Conduciveness [-1,1].500-.269.290 Control [-1,1].500-.141.402 Power [-1,1].500-.141.402 Labelela-joyanx-worela-joy

21. Feeling intensity Often useful to compress feeling frame into a single intensity value 21

22. Feeling intensity criteria Limited range: Should map onto [0,1] No dominant appraisal: No single value should drown out all the others Cant just multiply values, because if any are 0, then intensity is 0 Realization principle: Expected events should be less intense than unexpected events 22

23. Intensity function Realization principle: Surprise factor OP = Outcome Probability DE = Discrepancy from Expectation 23 OP=lowOP=high DE=lowSF=highSF=low DE=highSF=lowSF=high

24. Intensity function No dominant appraisal Just average the rest of the appraisals together 24

25. Intensity function Normalize ranges to same size Treat values as magnitudes 25

26. Example 26 EmotionMoodFeeling Suddenness [0,1]0.235 Unpredictability [0,1].250.400.419 Intrinsic-pleasantness [-1,1]0-.235 Goal-relevance [0,1].750.222.750 Causal-agent (self) [0,1]000 Causal-agent (other) [0,1]000 Causal-agent (nature) [0,1]1.6601 Causal-motive (intentional) [0,1]000 Causal-motive (chance) [0,1]1.6601 Causal-motive (negligence) [0,1]000 Outcome-probability [0,1].750.516.759 Discrepancy [0,1].250.326.362 Conduciveness [-1,1].500-.269.290 Control [-1,1].500-.141.402 Power [-1,1].500-.141.402 Labelela-joyanx-worela-joy Intensity.127

27. Conclusion Contributions Proposed concrete distinction between emotion, mood and feeling Proposed common representation for these, including value ranges Listed criteria for models of mood-emotion combinations Listed criteria for models of feeling intensity Proposed functions that fulfill those criteria Future work Discover more criteria and alternative functions Demonstrate that usage of these functions confers a functional advantage Human data 27