Morphing the Monster: Using Emotionally-Intelligent Computer Characters to Produce new Versions of Mary Shelley’s Frankenstein
College of Digital Media
Citation: Clark Elliott, 2013, Morphing the Monster: Using Emotionally-Intelligent Computer Characters to Produce new Versions of Mary Shelley’s Frankenstein, Working Draft for The Chicago Colloquium on Digital Humanities & Computer Science, DePaul University, December 5-8th, Chicago, IL. http://condor.depaul.edu/elliott/ar/papers/MorphingTheMonster.html
This is a position paper. We first argue that using a highly-computable model of emotion allows us to extract an essential structure in stories which is independent of the narrative context. We then show that we can autonomously manipulate these structures with emotionally-intelligent story-morphing computer programs to produce novel emotion tapestries that are, themselves, the basis of good, new, stories in the original setting. We introduce the basics of the emotion theory relative to this work, and analyze a short section from Mary Shelley’s Frankenstein—or the Modern Prometheus to illustrate the techniques.
Various claims have been made about the essential nature of stories. For example, we have heard that there are only five (or seven, or...?) main themes in narratives. We have heard that even the simplest story must have a point, and that a point is generated by a failure of expectation. We have all encountered many theories of plot development. By contrast, the orthogonal position we take in this paper is that the simplest story is generated when something happens, and someone cares about it, and that the underlying ways in which people care is both rich and complex. Many of the important elements of plots, and all important themes, are all based on such caring, which itself yields a vast number of stories that are, irrespective of the events themselves, based on unique, but identifiable emotion patterns.
We might say, “The boy was sitting in the chair,” and it is very hard to argue that this is a story. If we say, “The boy was sitting in the chair, and felt guilty about it,” our analysis will change. The boy cares, and we might want to know why. Or we might sympathize with him because we have ourselves sat in a chair, and known that we shouldn’t. Or we are reminded of being outside the principal’s office, feeling that we shouldn’t be there. And, too, there are dozens of flavors of that guilt the boy is feeling that might be of interest to us.
We also have lots of stories where we know the outcome, and it is expected. “The boy sat in the chair and felt guilty about it. He just couldn’t stop himself from taking the candy.” There is not too much that is out of the ordinary here, but it is still a story. Yet we could say, “The boy sat in the chair, which had fourteen legs and was painted pink,” which does not meet our expectations, and is also not a story, because no one cares.
If we want to build computational models of stories, we have to start with a computational model of emotion.
In the Affective Reasoning system we view all narrative components of emotion as computable. Our model includes twenty-four emotion categories with multiple intensities and qualities within each category. For example the category anger includes annoyance, indignation, exasperation, outrage, and so on. We typically compute at least three different intensities for many of the emotions. (FOOTNOTE: We do not currently have any model of different qualities of emotion (e.g., the difference between anger as indignation and anger as rage)).
The model also includes over four-hundred channels for expressing emotions, roughly twenty related channels for each emotion. For example a somatic channel for the expression of love would include turning red, and pulse increasing.
Clark Elliott, 2013
after Ortony, et al., 1988
CATEGORY LABEL AND EMOTION TYPE
appraisal of a situation as an event
joy: pleased about an event
appraisal based on presumption of
how a situation is appraised by another
as an event
happy-for: pleased about an event desirable for another
appraisal of a
situation as a prospective
hope: pleased about a prospective desirable event
appraisal of a situation as and event
confirming or disconfirming
satisfaction: pleased about a confirmed desirable event
appraisal of a
situation as an accountable
pride: approving of one's own act
appraisal of a
situation as containing
liking: finding an object appealing
compound emotion extensions
*Non-symmetric additions necessary for some stories.
In some of our work we have used the construct of Emotionally Intelligent (computer) Agents to take the place of characters in stories [Elliott, et al. 1998]. For each such agent we can ask, how does this agent feel about the events which are unfolding? and how might this agent express those feelings?
Such agents are constructed, in computer code, of two components: a disposition which controls the way they interpret situations that unfold in a story, and a temperament which controls how they express any emotions that may arise.
Using these simple mechanisms, we can alter the disposition of the agents, such that they differently appraise situations that arise, and also alter their temperaments such that they express emotions differently. In this way, within the constraints of not altering what happens in the plot of the story, we can still greatly change the emotion structure of the story and thus change the story itself. Because the agents are internally consistent, their emotions are as well, and the new stories generated make sense within the context of the new characters being portrayed.
For example let us consider a simple example partially using what we call the fortunes-of-others emotions.
Plot: Lisa has a brother Jake who has a dog Scout. Scout gets out of the house and unbekownst to Jake eats a dead squirrel she has found at the beach. Lisa visits Jake who comments that Scout seems subdued even though it is dinnertime and she shows no interest in her dinner.
Story one: Lisa feels sorry for her brother Jake, with whom she is close. She knows that Jake is very protective of Scout and will be worrying about her, and is mad at Scout for getting out. Jake is indeed worried about Scout. He feels guilty that he let her get out. Scout does not feel well because of the rotten squirrel.
Story two: Lisa is gloating over her brother Jake, with whom she is very competitive. She feels reproach for Jake who does not know how to take care of his dog. Jake is indeed worried about Scout, but he also admires his intrepid escape-artist dog, and is proud to be her owner. Scout does not feel well because of the rotten squirrel.
Story three: Lisa is jealous of her brother Jake because Scout loves him. She is reproachful of Jake whom she feels should want to take better care of his dog. She makes a plan to scold him later. Jake is gloating over his stupid dog, because she has obviously done something wrong and is now sick because of it. Scout does not feel well because of the rotten squirrel.
Story four: Lisa does not feel much of anything—it is not her problem. Jake is furious at his dog Scout for getting out of the house and getting into trouble. He is speaking in a really loud voice. Scout is unhappy that Jake is mad at her. She is afraid that Jake will punish her. She is happy about having eaten the dead squirrel which was the high point of her day. Her stomach hurts but she doesn’t care much about that.
Story five: Lisa does not feel much of anything—it is not her problem. Jake is furious at his dog Scout for getting out of the house and getting into trouble. He is trembling and red in the face, but not saying anything. Scout has mixed emotions. She feels guilty that she ate the dead squirrel and she is afraid that Jake will start yelling at her, but she is also happy about having eaten the dead squirrel which was the high point of her day.
In this way using only the crudest of the manipulations that can be controlled by the Affective Reasoner mechanisms, we are nonetheless able to generate scores of stories.
Let us now examine the mechanisms used in this simple example in a little more detail.
In table 1, we see that there are four main divisions of the twenty-five emotion categories. First is the large set of emotions that arise because of the goals of agents—what agents want and don’t want. Second is the set of emotions based on the principles of the agents—relevant to actions agents believe should and should not be performed. Third is the set of small set of emotions based on what agents like and don’t like—their preferences. Fourth is the set of emotions based on combinations of other categories that subsume their constituent parts.
Now, on the one hand, to control how agents respond to the circumstances—the emotion eliciting conditions -- that unfold in a story, we have to change the way they appraise those circumstances. These appraisals are part of what we build into an agent’s stable disposition.
In the first case Lisa appraises her brother Jake’s distress as blocking one of her own fortunes-of-others goals (Lisa’s desire for Jake’s ongoing well-being): when Jake is sad, she feels sorry for him. Lisa is angry (a compound emotion) at Scout because Scout has both violated Lisa’s principle that dogs should not run away and get in trouble, and also blocked her goal for Jake’s well-being. Jake apparaises the situation as indicative of a possible future goal of his own—keeping his dog healthy—being blocked, although this is currently uncertain (worry). He also, independently appraises the situation as an instance of him violating his own principle of keeping Scout safe (guilt). Scout’s own health well-being goal is being blocked by an uncomfortable stomach (distress).
In the second case these appraisals change, and so do the resulting emotion states. For example, in the second story Lisa’s friendship relationship with Jake has changed to become (in this situation) adversarial. So now when Jake’s ongoing well-being has taken a downturn, Lisa feels good about it.
All of the many changes in the emotions that arise in the various simple stories about Lisa, Jake, and Scout, stem only from such changes in the ways that agents appraise the otherwise identical events that that occur in the plot.
Among the twenty-odd action channels that differentially control the expression of any particular emotion we have a spectrum of paths such as somatic, behavioral directed toward an inanimate [or animate] object, communicative [non-]verbal responses, evaluative self-directed attributions, ..., repression, suppression, reappraisal of the situation, reappraisal of one’s self, other-directed emotion modulation, ..., full plan initiation, and so on, from the simplest body responses, to the most complex intellectual responses.
So now, on the other hand, to control how agents express their emotions we have to change what we build into their temperaments. We achieve this by changing the action-expression channels that are activated at any given moment for an agent, which in turn, taken together, yield the agent’s temperament.
For example, in story four Jake expresses his anger by shouting, a communicative verbal repsonse, or possibly (because he is talking to a dog) a behavioral response directed toward an animate object indicating that for this temperament, those channels are activated. By contrast, in story five, Jake is trembling and red in the face which are somatic responses.
Having introduced the basics, we can now look at some additional ways in which we can control the emotion structure of stories.
Relationships: First, as hinted above, we have four relationships that we model, between agents: friendship, adversarial, cognitive-unit, and no-relationship. Friendship is intended to collect together all relationships wherein e.g. agent Lisa will feel good when good things happen to agent Jake, and bad when bad things happen to Jake. Adversarial is the opposite: Lisa will feel bad when good things happen to Jake, and good when bad things happen to Jake. Cognitive-unit is when Lisa feels exactly what Jake feels, as though she were in his shoes. (For example, a mother may feel frightened with her son when he is forgetting his lines during the school play.) These relationships are unilateral and even when bi-lateral might not be symmetric.
It is possible to change the relationships that agents have with one-another, and in this way affect the fortunes-of-others emotions that will arise in the system (sorry-for, happy-for, gloating, resentment).
Emotion intensity variables: Next we can change the emotion-intensity-variables which contribute to the particular (intensity of) emotion that is generated within any one of the emotion categories, and also, thus, subsequently any change in action-responses that are dependent on emotion intensity.
In the Affective Reasoner we can manipulate up to twenty-four different variables that affect how strongly situations elicit emotional responses in the agents, divided into three categories [Elliott & Siegle, 1993].
The first category of such variables, the simulation-event variables, are those that are contained in the (simulation of the) situation itself. For story-morphing there are constraints on the usefulness of this set of variables because they are bound to the external plot and description of the story itself, which changes we want to keep to a minimum. That is, these are always actual plot changes, albeit possibly representing purely local changes that do not affect the remainder of the plot. For example, if we change the amount of money a patron leaves as a tip, the waiter might also appraise the event of getting the money differently—which is what we want with story-morphing—but we have to be very careful that such a change does not alter the plot in ways requiring real-world knowledge to control, which is explicitly beyond the scope of the AR’s capabilities. Nonetheless some changes are possible within these constraints.
The values in the simulation-event variables change independently of an agent’s interpretation of them, and one change in a single place can conceivably affect all of the agents in the system at once. Among these variables are how much a goal is realized or blocked, how blameworthy or praiseworthy an action is, how certain the situation is, how real it seems, how surprising it is, how deserving the agent is of the situation, and so on. For example, if an adversary is particularly deserving of her bad fortune, an agent observing that bad fortune may gloat in a particularly strong way.
The second category, the stable disposition variables are those variables that are internal to each agent. Changes in one of these will not affect any other agent’s interpretation of situations. Unlike the simulation-event variables these values can generally be changed at will, and thus are easy to use with story-morphing. That is, how an agent is disposed to see situations that arise is, in essence, up to them. Among these variables are how important it is to achieve a goal or keep it from getting blocked; or uphold a principle, or not have it violated. For example, after losing a game three times to a rival, the importance of the goal of winning might become increasingly important. Included in this category would be the emotional interrelatedness of two agents. The more (unilaterally) interrelated they are seen to be by one of the agents, the stronger the emotions generated in the context of this relationship.
There is room for some finesse here, as well. For example, we have independent variables for how an agent sees the importance of upholding of a principle, and for the importance of not violating it. In this way one version of an agent’s disposition might have strong emotions over admirability of hearing romantic passion in music, but not care much at all if someone does not. An alternate version of the agent’s disposition might find the agent greatly disdainful of those that cannot hear romantic passion in music, but not feel much admiration those that do—taking it to be expected of them. Or, the agent might feel strongly in both instances, or only mildly experience emotions in both cases.
Next, we have mood variables, which are intended to vary over time and affect both agents’ dispositions, and also their temperaments. Non-relationship mood variables which include changes in values like arousal and physical well-being, can make emotions stronger or weaker. Also included (among others) are a bias toward negative or positive emotions across a strength spectrum (a good mood, a bad mood), and anxiety-invincibility, which affects the strength of prospect-based emotions. Relationship mood variables affect how an agent is diposed toward judging another agent, or, differently, other agents as a whole. This affects how an agent is biased toward praising the actions of another, or toward condemning them.
Concerns of others: Lastly, because emotions are sometimes generated on behalf of agents according to how they believe others to interpret situations—stored in what we call “concerns of others” structures (COOs) -- we can alter these beliefs and thus change the appearance and strength of the fortunes-of-others emotions.
Constraints on plots: Sometimes emotions are generated by the emotions of others and in this way drive the plot. For example, some people lash out when others don’t like them. When this happens in a plot we require some level of consistency and must either (a) leave the appraisals of the emotions of others alone at that point, or (b) put some constraints on changes in the emotions being generated—such as requiring that the valence remain the same.
To generate new stories using the story-morphing system, in any context, we must give some thought to the nature of each part of the plot. In particular, we have to decide which parts of the plot—“what happened”—fall into one of three general categories: (1) those parts that are independent of how agents feel about them (e.g., the invaders overrun the castle walls), (2) those parts that allow for emotional freedom, but within constraints (e.g., an agent has to feel positively (for some reason) about the prospect of meeting a stranger because they are next going to pursue a meeting in the plot, and (3) those parts of the plot that require specific emotions such that those emotions cannot be changed in any significant way. To meet these constraints we have to give some thought to possible appraisals that agents might make for each of the situations that arise during the unfolding of the plot.
Within this context we typically still find a wide range of concerns that we can build for our agents, and a wide range of ways in which they express their resulting emotions, yielding dozens, and sometimes hundreds, of valid variations of the narrative.
One serendipitous reason that story-morphing works is that we do not require a perfect product. Humans consider the reading of the emotions of others (and even themselves) an inexact science: we are often willing to jump to conclusions and provide our own (posssibly incorrect) assumptions about what someone is feeling, and why they are feeling it. This is a normal part of the human condition.
How development proceeds
The story-morphing system is a structure-based procedure that can apply in many contexts, and development for each will be different. The handling of text is an important consideration. The Affective Reasoner has no built-in text-generation capabilities, or language understanding capabilities. However, it is a system designed to drive such processes by providing computer-controlled manipulation and understanding of the underlying emotion fabric of stories, which we claim is one of the most critical elements of stories. Themes arise from the emotion fabric, and can thus be transferred from one context to another, from one story to another. Memorable scenes in stories are often dependent on the underlying emotion structure. Identification with characters often comes from identifying with the emotions they are experiencing, independent of the context of the character’s lives.
To illustrate how development might proceed, let us suppose that we wish to manipulate characters within a computer-controlled presentation of our stories. [Ref story-morphing paper] For the purposes of this position paper we need not be more specific. Such underlying work would apply to a number of applications. We might, for example, wish to generate teaching stories as part of an automated tutoring system. We might wish to play out our stories through virtual actors, using emotion-appropriate background music. We might wish to generate real-time characters as part of a computer game that act in novel (surprising, internally consistent...) ways according to the current configuration of their personalities.
To build a platform that supports such systems we would encode the plot as a series of events unfolding in a simulation. These sim-events (simulation events) contain ground instances of “what happens” in our modeled world, played out as a discrete series of states within the system. Included in this series of sim-events is what the characters themselves—our agents—do. Actions of agents generated by the emotion system are inserted into the plot as additional sim-events. Our agents contain specialized internal, matching, versions of all those sim-events that are important to them for one reason or another. These internal agent sim-events, which we call agent-frames, form the basis of each agent’s disposition. The agent-frames support complex matching algorithms containing variables and functions, and themselves form the left hand side of rules such that when a match succeeds between an agent-frame and the current sim-event the rule fires and an emotion is generated. And, notably, the emotion is generated along with all the variable bindings that were created during the match process. In this way, for example, an agent doesn’t just get angry, she gets angry at the other agent that was bound to the “agent-that-violated-my-principle” variable during the match of the offending sim-event.
Once an emotion is generated, it is expressed through the particular temperament channels that are activated as part of the agent’s current personality. The resulting expression of the emotion is formed as a new sim-event, and placed in the simulator’s event queue.
In this way all of the events in the plot are simulated, along with all of the emotion events that the plot generates. To play the original story we configure our agent personalities with (our interpretation of) the concerns and temperaments of the original characters. To create a story-morph we variously alter the agent-frames (comprising an agent’s disposition) to embody different concerns of the characters, different moods, different interpretations of the concerns of others, and different relationships, and also alter the activation of the emotion-expression channels (comprising an agent’s temperament) to embody different ways of the agents expressing themselves, and then re-run the simulation.
An important feature of this system is that novel, new, internally-consistent stories can be generated automatically by the system without intervention by human authors.
Some additional comments.
Humor: Certain types of stories are humorous because of their particular emotion structure. [Ref. Motorcyle paper] When this structure can be fully captured by our emotion theory, or minor extensions of it, we can also generate humorous stories.
Case-based reasoning: Agents have case-based reasoning intelligence that allows them to dynamically choose different internal representations (COOs) for how they believe others see the world. These are based on the features of eliciting situations that arise, and the responses of agents to those situations, used as remindings for how others they have known, or they themselves, see the world.
Applications: A computational emotion-based theory of stories is widely applicable. One obvious application is in addressing what is known as the content bottleneck in computer games. In the gaming industry there is a delicate balance between (a) making a computer game too easy to “figure out,” so that interest is not maintained for long, and (b) making it too complex so that it is too hard to understand, and interest is never piqued. Appropriately complex content is expensive to develop, and varied plots are both difficult to generate and burdensome to make cohesive and interesting. Using story-morphing techniques we claim that highly complex and novel game-stories can be automatically generated based on how the characters feel about what is unfolding—possibly controlling their dispositional behavior—rather than on the external complexities of the plot.
Other applications include role-playing therapy, story-telling and story-understanding systems, applications for children, and as a component part of emotionally-intelligent agents.
Morhping the Monster
Let us now look at a more extended example which will help to illustrate the richness of the story-morphing pallet. We borrow a short passage from Chapter Five of Mary Shelley’s Frankenstein—or the Modern Prometheus.
A paraphrase of the original narrative:
Background: Victor Frankenstein has created an artificial human being which is just now coming to life.
Victor is in a high state of arousal because the monster is finally coming to life.
Physical beauty is important to Victor. Victor likes correct design proportions in the human body. Victor likes clear beautiful skin, and dislikes yellow pasty skin and yellow eyes.
Victor has worked very hard to make a beautiful artificial human.
Victor was hoping to celebrate creating new life, but is now disappointed that it is ugly.
Victor leaves to get some sleep, and dreams of his wife.
The Monster gets up, goes to see Victor, and loves him. He wants approval. He smiles at Victor and reaches out his hand.
Victor wakes, is overcome with disgust, and rushes downstairs.
Victor is bitterly disappointed that the new life he has created is not beautiful.
Now, using only those story-morphing techniques that manipulate computable aspects of characters’ personalities within the context of the narrative we can create different meaning, using the same action steps that take place in the original plot. For the purposes of illustration we present only snippets of what would be significantly more extensive passage in a full story-morph treatment.
Story morph snippet one:
Victor feels mixed emotions. He loves his creation (admires the struggle for life; feels the newly created life itself is beautiful). Victor feels that parents should love their children and see them as attractive no matter what (a principle), and he is the parent of the monster. However, Victor is repulsed by the ugliness of the monster.
Victor is ashamed that he does not see his creation as beautiful.
The monster is afraid that Victor will not like him.
The monster’s fears are confirmed when Victor runs away from him.
The monster is angry at Victor for hurting his feelings, and for not taking care of him.
Story morph snippet two:
Victor makes an assessment of the monster. He feels nothing but scientific curiosity. He is tired and leaves to get some sleep.
The monster comes to life and is desperate for the affection of his creator.
When victor does not respond the monster is sad and thrown into depression.
Victor feels reproach for the monster for being so emotional and leaves.
The monster feels abandoned.
Story morph three:
Victor loves his monster. He fears that others will harm his monster because they will see the monster as ugly. Victor feels guilty he did not make a beautiful creation and it is his fault that others will harm his creation.
Victor admires the monster’s strength.
Victor leaves to sleep.
The monster is curious and goes to see Victor.
Victor is hoping to see signs of love in the monster’s eyes, but sees none.
The monster feels nothing for Victor.
Victor feels rejected by the monster and this leads to bitter disappointment because Victor has been hoping for two years to build someone that will love him. He has invested a great deal of effort in this project.
Victor feels shame that he has not provided his monster with a family where members love one another. He can’t bear his shame and leaves.
Story morph four:
Victor is in a strong adversarial relationship with (toward) the monster he has created. His goal is to create life that he can mistreat with impunity. The monster has a strong freindship relationship with (toward) Victor.
Victor is gloating because his monster has ugly skin and he will be able to use this against the monster. He is sad that the monster has good proportions. He is afraid that because the monster is so strong he will not be able to mistreat him very extensively.
Victor looks forward to the moment when the monster realizes that Victor despises him.
Victor leaves to get some rest.
The monster comes to life and has a strong desire for human contact. He is lonely, but is hopeful of being loved by others.
He finds Victor and is satisfied to find human company. He feels love for Victor and reaches out to him.
Victor expresses his disgust at the monster. He is very excited about the impending feelings of rejection the monster will feel.
The monster is now terribly sad to be rejected by Victor.
Victor gloats over the monster’s distress.
The monster gets angry at Victor for behaving so badly.
Victor fears that the monster will hurt him and runs away.
Story-morph snippet five:
Victor loves his monster very much. He believes that parents should love their children and also that they should always find their children beautiful despite their faults. He is proud of loving his creation despite his ugliness, but he is remorseful that he finds the monster repulsive. He does not express his love strongly because such emotions are repressed in his temperament, but his temperament is also such that disgust is shown in a communicative-verbal way. He calls out at the monster saying “Disgusting!”
The monster, who desperately wants love, does not see that Victor loves him, but only that Victor is repulsed by him.
Story-morph snippet six:
As above in story-morph five, but...
The monster is very happy to find himself alive. This puts him in a very good mood, and he is predisposed to appraising the world in a positive light.
The monster ignores Victor’s comment that he, the monster, is disgusting, but notices Victor’s obsessive attentional focus on him as an object of love. He is satisfied to feel Victor’s love.
Story-morph snippet seven:
[Using the case-based reasoning capabilities of agents]
The monster knows that he has pasty skin and yellow eyes. He sees that Victor is disgusted. Taking these features together he changes his his internal representation of how he believes Victor sees the world. He feels sorry for Victor because he, the monster, is so ugly, and now believes this makes Victor unhappy.
Story-morph snippet eight:
[Using extensions for humor]
The monster sees Victor as an authority figure. He believes Victor holds everyone to high standards of behavior. He believes Victor will hold him, the monster, to high standards. Victor has created a monster that is ugly, thus violating one his own standards. The monster observes that Victor knows the monster knows that Victor has violated his own standards. The monster finds Victor’s chagrin funny, and laughs.
Victor is embarrassed. He resents the monster laughing at him.
Story-morph snippet nine:
[Using extensions for altering Concerns-of-Others structures]
Victor looks at his sleeping monster and feels pity for him because he assumes the monster will feel very badly about being so ugly.
Victor leaves to sleep.
The monster wakes and is happy to be alive. This happiness trumps all other feelings.
Later when the monster comes, Victor realizes that the monster is happy, and changes how he believes the monster perceives himself. He stops feeling pity and now feels happy for the monster that he enjoys being alive.
Some finer-grained variations:
The monster is now terribly sad to be rejected by Victor:
The intensity of this rejection is increased because this is unexpected by the monster, and the surprisingness contributes to intensity.
The importance of not being rejected is very high for the monster, and this increases the intensity.
Feeling rejected causes the monster to express this by reappraising [himself] as being ugly and unlovable.
The disgust on Victor’s face is taken by the monster to be a very strong indication of intense dislike, and this [sim-event variable] contributes to the intensity.
Victor’s temperament is manipulative and he expresses his disgust through Other-directed emotion modulation channels to make the monster feel as badly as possible.
Each of these snippets from different story-morphs is based exclusively on components of emotion that the Affective Reasoner system can manipulate. In addition because they are based entirely on the logical structure of how emotions arise, and are expressed, they come with sophisticated explanations, embodied as explicit values in what might amount to be hundreds of details for each emotion generated.
Conclusions: Complex, but precise, emotion structure can be teased out of all stories, based on how the characters appraise the events that arise in the narrative. This emotion structure is essential to what makes a story a story. The emotion structure is both portable (it can be repeated in an entirely different context) and subject to manipulation. The Affective Reasoner, which does manipulate such emotion structures, can be used to automatically generate new, and novel, stories which are nonetheless internally consistent, because of the consistent nature of the artificial personalities that are dynamically constructed by the computer.
Summary: We have discussed the nature of what makes a story a story, and not simply a collection of plot events, claiming that how an character cares about unfolding events is critical, and also, necessarily, gives rise to emotions. We were introduced to the underlying emotion theory that drives the Affective Reasoner. We worked through several snippets of stories to see how story-morphing works, and concluded with a more extended example from chapter five of Mary Shelley’s Frankenstein.
· Ortony, A.; Clore, G. L.; and Collins, A.
1988. The Cognitive Structure of Emotions.
Cambridge University Press.