Automated Story Generation: Balancing Plot and Character

1
Automated Story GenerationBalancing Plot and
Character

• Mark Riedl
• Research ScientistInstitute for Creative
  TechnologiesUniversity of Southern California

2
Storytelling

• Storytelling is a pervasive part of our world and
  our culture
• Narrative intelligence we use same cognitive
  structures to understand stories as to understand
  world around us
• Relevant applications of storytelling systems
• Entertainment
• Education
• Training

3
State of the Art in Storytelling Systems

• Standard practice script a story at design time
• Drawbacks
• Every user has same experience
• Each user experiences same story every session
• Alternative approach generate stories
  automatically on a per session basis
• Benefits
• Customize story to user preferences/abilities
• Vary story from one session to the next

4
Computer Games and Storytelling

• Current game approach
• Use story as glue between interactive sessions
• Back story cut scenes

5
Interactive Storytelling

• Goal Tell a story in which user is an active
  participant
• User decisions and actions can affect the
  direction and outcome of plot
• Control vs. Coherence
• Approaches
• Branching story graphs
• Story scripted at design time like a flow chart
• Combinatorial complexity of authoring
• Generative techniques

6
Automated Story Generation

• Approaches
• Reactive, autonomous character agents
• Character agents drama manager
• Generate a linear story
• Re-generate when user does something unexpected

7
Introduction to Narrative

• What is narrative?
• Recounting of a sequence of events
• What is story?
• Narrative with plot that is structured
• Narrative has 3 levels of interpretation
• Text/Media
• Sjuet
• Fabula
• My work focuses on fabula generation

8
Challenges for Story Generation

• Challenges of this work involve choosing and
  structuring story content
• Focus balancing plot and character
• Plot coherence
• Perception that main events in the story have
  meaning and relevance to the outcome of the story
• Character believability
• Perception that characters act according to
  internal motivations, desires, and traits

9
History of Story Generation (1)

• Story understanding
• Tale-Spin
• Meehan, c. 1975
• Inference engine
• Forward-chaining goal-based character activities
• Characters are motivated by goals
• Quality dependent on world representation
• No dramatic goals

Once upon a time, there was a dishonest fox named
Henry who lived in a cave, and a vain and
trusting crow named Joe who lived in an elm tree.
Joe had gotten a piece of cheese and was holding
it in his mouth. One day, Henry walked from his
cave, across the meadow to the elm tree. He saw
Joe Crow and the cheese and became hungry. He
decided that he might get the cheese if Joe Crow
spoke, so he told Joe that he liked his singing
very much and wanted to hear him sing. Joe was
very pleased with Henry and began to sing. The
cheese fell out of his mouth, down to the ground.
Henry picked up the cheese and told Joe Crow
that he was stupid. Joe was angry, and didnt
trust Henry anymore. Henry returned to his cave.
Henry Ant was thirsty. He walked over to the
river bank where his good friend Bill Bird was
sitting. Henry slipped and fell in the river.
He was unable to call for help. He drowned.
10
History of Story Generation (2)
(tell (((churn neil liz)))) working on goal
(CHURN NEIL LIZ) - using plan FORCED-MARRIAGE work
ing on goal (DO-THREATEN STEPHANO LIZ forget
it) - using plan THREATEN STEPHANO threatens
LIZ forget it working on goal (DUMP-LOVER
LIZ NEIL) - using plan BREAK-UP LIZ tells
NEIL she doesnt love him working on goal
(TOGETHER NEIL) - using plan SEDUCTION
DAPHNE seduces NEIL working on goal (ELIMINATE
STEPHANO) - using plan ATTEMPTED-MURDER RENEE
tries to kill STEPHANO working on goal
(DO-DIVORCE TONY LIZ) - using plan DIVORCE
LIZ and TONY got divorced working on goal
(TOGETHER LIZ NEIL)

• Universe
• Lebowitz, c. 1985
• Hierarchical and backward-chaining planning
• Driven by authors goals
• Focus on coherence but not believability

Liz was married to Tony. Neither loved the
other, and, indeed, Liz was in love with Neil.
However, unknown to either Tony or Neil,
Stephano, Tonys father, who wanted Liz to
produce a grandson for him, threatened Liz that
if she left Tony, he would kill Neil. Convinced
that he was serious by a bomb that exploded near
Neil, Liz told Neil that she did not love him,
that she was still in love with Tony, and that he
should forget about her. Eventually, Neil was
convinced and he married Marie. Later, when Liz
was finally free from Tony (because Stephano had
died), Neil was not free to marry her and their
troubles went on.
11
History of Story Generation (3)

• Era of interactivity
• Oz Project
• Bates et al., c. 1995
• Reactive planning agents
• Drama manager
• Façade
• Mateas and Stern, c. 2000
• Reactively assembled beats
• Authoring!
• I-Storytelling (Friends)
• Cavazza et al., c. 2001
• Conflicting HTN character plans
• No drama management

12
Categorizing Story Generation Systems

• Character-centric
• Model goals and plans of autonomous characters
• Responsibility of story generation placed on
  character agents

• In general, character-centric systems
• More reliably produce stories with character
  believability
• Less reliably produce stories with plot coherence

• Author-centric
• Model thought processes of human author
• Responsibility of story generation placed on
  author agent

• In general, author-centric systems
• More reliably produce stories with plot coherence
• Less reliably produce stories with character
  believability

13
The Problem

• Question How can stories be generated that
  combine both
• Strong plot coherence
• Strong character believability
• An answer Merge author-centric and
  character-centric techniques
• Partial-order planning
• Reasoning about character intentions

? Author-centric approach
? Character-centric approach
14
Causal Link Planning

• Plan is tuple,
• S steps
• B variable bindings
• O ordering constraints
• L causal links
• Steps have preconditions and effects
• Causal links capture temporal and causal
  relationship between steps

p3
S1
Goal
Init
p1
p6
p4
p4
S2
p1
p7
p7
p5
p2
p8
S3
p2
p5
p9
15
Causal Link Planning

• Plan creation is process of flaw revision
• Consider all ways of repairing flaw
• Open condition flaws
• Reuse an existing step
• Insert a new step
• Search tree of all possible plan revisions
• Heuristic functions guide search

p3
S1
Goal
Init
p1
p6
p4
p4
S2
p1
p7
p7
p5
p2
p8
S3
p2
p5
p9
16
Planning and Narrative

• Why use planning?
• Planners reason about action, causality, and
  temporality

• Conventional planning
• Single agent
• Goal is agents desired world state
• Goal state is intended by agent

• Narrative planning
• Multiple characters
• Goal describes outcome of the story
• Outcome is not necessarily intended by any
  characters

• Mismatch due to assumption of intentionality
• Augment planner with models of character
  believability and intentionality

17
Modeling Story with Plans

• Causal-dependency planners provide plot coherence
  for free!
• Character believability
• Consistency
• Defined relative to personality traits
• Observable based on selection among alternative
  actions
• Intentionality
• Defined relative to internal character goals
• Observable based on actions performed in pursuit
  of internal goals
• Other stuff
• Appearance, animation, emotion, etc

18
Intent-Driven Planning

• Intent-driven Partial Order Causal Link (IPOCL)
  planner
• Conventional causal dependency planning
• Provides soundness and coherence
• Reasoning about character intention
• Determine internal character goals for characters
  and further drive content creation process by
  motivating internal character goals

Riedl Young (2004). An Intent-Driven Planner
for Multi-Agent Story Generation. Proceedings of
AAMAS 2004.
19
Character Intention in Narrative Planning

• Look at actions in the plan and determine
  plausible internal character goals
• When a new action is added
• Why would the character do that?
• To achieve some new internal goal
• To facilitate another action that achieves some
  internal goal
• Why does the character have that internal goal?
• Find a motivating action
• When an existing action is reused
• Can that action be part of more than one
  intention?
• Yes or no

20
Character Intention in Narrative Planning

• Plan is tuple,
• C frames of commitment
• Frame of commitment
• Records a characters intention to achieve an
  internal goal
• Actions associated with frame are intended to
  achieve the goal
• Motivating actions cause characters to have goals

Char1 intends p
Intention level
Domain level
S4
S1
Goal
Init
S2
S3
21
IPOCL Algorithm

• Causal planning
• Bookkeeping as normal
• If new step, s
• Choose an effect, e, of s or nil.
• If e ? nil, let c be a new frame of commitment
  with internal character goal, e. Create a new
  open condition flaw on c.
• For each existing frame of commitment, c, that
  can explain s, create a new intent flaw, f
  c
• Intent planning
• Let s be a step that can possibly be explained by
  frame of commitment, c.
• Choose one of the following
• Make s part of the interval of c and recursively
  propagate flaw to steps immediately preceding s.
• Do not make s part of the interval of c.

22
Intent-Driven Planning Example
Villain intends (controls Vil Prez)
(intends Vil (controls Vil Prez))
Villains Intention level
Domain level
Init
Goal
(person hero)
(person villain)
Bribe (Vil, Prez, )
(has Vil )
Give (Hero, Vil, )
(intends Vil )
(has Hero )
(corrupt Prez)
(has Hero )
Coerce (Vil, Hero, (has Vil ))
Domain level
Heros Intention level
Hero intends (has Vil )
(intends Hero (has Vil ))
23
Aladdin Example
There is a woman named Jasmine. There is a
king named Mamoud. This is a story about how
King Mamoud becomes married to Jasmine. There is
a magic genie. This is also a story about how
the genie dies. There is a magic lamp.
There is a dragon. The dragon has the magic
lamp. The genie is confined within the magic
lamp. There is a brave knight named
Aladdin. Aladdin travels from the castle to the
mountains. Aladdin slays the dragon. The dragon
is dead. Aladdin takes the magic lamp from the
dead body of the dragon. Aladdin travels from
the mountains to the castle. Aladdin hands the
magic lamp to King Mamoud. The genie is in the
magic lamp. King Mamoud rubs the magic lamp and
summons the genie out of it. The genie is not
confined within the magic lamp. The genie casts
a spell on Jasmine making her fall in love with
King Mamoud. Jasmine is madly in love with King
Mamoud. Aladdin slays the genie. King Mamoud is
not married. Jasmine is very beautiful. King
Mamoud sees Jasmine and instantly falls in love
with her. King Mamoud and Jasmine wed in an
extravagant ceremony. The genie is dead.
King Mamoud and Jasmine are married. The end.
There is a woman named Jasmine. There is a
king named Mamoud. This is a story about how
King Mamoud becomes married to Jasmine. There is
a magic genie. This is also a story about how
the genie dies. There is a magic lamp.
There is a dragon. The dragon has the magic
lamp. The genie is confined within the magic
lamp. King Mamoud is not married.
Jasmine is very beautiful. King Mamoud sees
Jasmine and instantly falls in love with her.
King Mamoud wants to marry Jasmine. There is a
brave knight named Aladdin. Aladdin is loyal to
the death to King Mamoud. King Mamoud orders
Aladdin to get the magic lamp for him. Aladdin
wants King Mamoud to have the magic lamp.
Aladdin travels from the castle to the mountains.
Aladdin slays the dragon. The dragon is dead.
Aladdin takes the magic lamp from the dead body
of the dragon. Aladdin travels from the
mountains to the castle. Aladdin hands the magic
lamp to King Mamoud. The genie is in the magic
lamp. King Mamoud rubs the magic lamp and
summons the genie out of it. The genie is not
confined within the magic lamp. King Mamoud
controls the genie with the magic lamp. King
Mamoud uses the magic lamp to command the genie
to make Jasmine love him. The genie wants
Jasmine to be in love with King Mamoud. The
genie casts a spell on Jasmine making her fall in
love with King Mamoud. Jasmine is madly in love
with King Mamoud. Jasmine wants to marry King
Mamoud. The genie has a frightening appearance.
The genie appears threatening to Aladdin.
Aladdin wants the genie to die. Aladdin slays
the genie. King Mamoud and Jasmine wed in an
extravagant ceremony. The genie is dead.
King Mamoud and Jasmine are married. The end.
24
Limitations and Future Work

• Limitations
• Only addressed a small portion of the problem
• Failed intentions / failed actions
• Pacing and timing issues
• Computational complexity

25
Evaluation

• Do enhancements increase perception of character
  believability?
• Objective metrics
• Character intention understanding
• Elicit evidence without subjectivity
• Compare mental model of reader with some standard
  mental model

26
QUEST Model of Question-Answering

• (Graesser et al., 1991)
• Events and character goals are stored in memory
• Represented as a graph with nodes represent facts
  known by reader
• Relationships between nodes capture understanding
• Consequence
• Reason
• Initiate
• Outcome
• Implies

27
Example QUEST Knowledge Structure
Im
EVENT 22 Heroes returned daughters to palace
GOAL 21 Heroes return daughters to palace
EVENT 16 Heroes rescued daughters
GOAL 15 Heroes rescue daughters
EVENT 14 Heroes heard cries
EVENT 11 Daughters got help
GOAL 10 Daughters get help
O
O
O
I
C
C
R
R
C
C
GOAL 12 Daughters cry
EVENT 13 Daughters cried
GOAL 17 Heroes fight dragon
EVENT 18 Heroes fought dragon
EVENT 23 Czar heard of the rescue
C
O
O
C
R
I
GOAL 19 Heroes go to daughters and dragon
EVENT 20 Heroes came to daughters and dragon
GOAL 24 Czar reward heroes
O
O

• Question/answer pair
• Q Why did the heroes want to fight the dragon?
• A Because the heroes wanted to rescue the
  daughters

EVENT 25 Czar rewarded heroes
28
(No Transcript)
29
Character Intentionality Evaluation

• QUEST predicts the goodness-of-answer rating
• Procedure
• Generate 2 stories Control, Test
• Generate QUEST model of generated stories
• For each condition, use QUEST to separate
  question/answer pairs into good and poor
• Test subjects
• Hypothesis
• If system generates character believable
  narratives, then ...
• Subjects in the test condition will have higher
  GOA ratings for good question/answer pairs than
  subjects in the control condition
• Subjects in the test condition will have lower
  GOA ratings for poor question/answer pairs than
  subjects in the control condition
• Why this works
• Comparing reader models to QUEST predictions

30
Conclusions

• Plot and Character are intertwined
• Cant reason about one without the other
• Specialized algorithms needed for story
  generation
• Character believability hard Character
  intentionality easier
• Empirical evaluation
• Motivation beliefs desires intentions
• Explanation Expectation?
• Making stories interactive
• Reconcile character autonomy and story direction

31
The End