The Creative Process

A Computer Model of Storytelling and Creativity

Scott R. Turner

• Publisher: Psychology Press
• Year: 2014
• Print ISBN:
  9780805815764
• eBook ISBN:
  9781317780618
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

Someday computers will be artists. They'll be able to write amusing and original stories, invent and play games of unsurpassed complexity and inventiveness, tell jokes and suffer writer's block. But these things will require computers that can both achieve artistic goals and be creative. Both capabilities are far from accomplished.

This book presents a theory of creativity that addresses some of the many hard problems which must be solved to build a creative computer. It also presents an exploration of the kinds of goals and plans needed to write simple short stories. These theories have been implemented in a computer program called MINSTREL which tells stories about King Arthur and his knights. While far from being the silicon author of the future, MINSTREL does illuminate many of the interesting and difficult issues involved in constructing a creative computer.

The results presented here should be of interest to at least three different groups of people. Artificial intelligence researchers should find this work an interesting application of symbolic AI to the problems of story-telling and creativity. Psychologists interested in creativity and imagination should benefit from the attempt to build a detailed, explicit model of the creative process. Finally, authors and others interested in how people write should find MINSTREL's model of the author-level writing process thought-provoking.

1	Storytelling and Creativity

1.1 Introduction

During my senior year in college, I was browsing the stacks in the research library and quite by accident came across a small but intriguing book called The Morphology of the Folktale (Propp, 1968). The author, Vladmir Propp, had studied common Russian folktales and distilled the form ¹ of those tales into cryptic equations:

Using letters to represent story elements such as “One of the members of a family absents himself from home,” each equation captured a common pattern Propp had found in the folktales he studied. Altogether the equations in Propp’s book formed a definitive description of the form of Russian folktales.

As a computer scientist, I found this fascinating. Propp had reduced the folktale to a grammar—a set of well-defined rules. Grammars are used throughout computer science to formalize structure and to understand well-structured input. Compilers, for example, use grammars to translate computer programs in languages like FORTRAN—which are easily understood by humans—into the binary “ones and zeros” understood by computers. In theory, Propp’s grammar could be programmed into a computer and used to recognize folktales—provided someone first translated each folktale into Propp’s notation.

But what was more intriguing to me was the notion of running Propp’s grammar “backwards.” Propp’s grammar was intended as a tool for recognizing and understanding the underlying forms of folktales. But, I reasoned, the same grammar could be used in reverse to create folktales. By starting with an initial rule and then randomly choosing the next rule to apply, Propp’s grammar could be used to “grow” a story from seed to completion. The random choices would ensure that the story created would likely be different from any actual folk tale, while the rules would ensure that the resulting story had the form of a folk tale. And programming a computer to do this would be trivial. A few hours in the Computer Lab and I would have a computer program that could tell stories!

Or so I thought.

I did eventually write a computer program that tells stories. But it took years, not hours, and in the end, Vladmir Propp’s intriguing little grammar was nowhere to be seen.

This volume is the story of the program I wrote and what I learned in the process. It looks at the myriad problems an author faces when he sits down to write a story, and presents the processes, techniques, and knowledge needed to address these problems. Like all authors, I hope you find my story both interesting and enlightening.

1.2 The Storytelling Problem

It is surprisingly difficult to tell a story. Even young children can understand stories. By four or five, children understand most aspects of folktales like the ones Propp studied. Indeed, the primary use of folktales is to teach the young principles they’ll need as adults.

But telling a story is a different matter. It seems an easy enough task. Surely an adult should be able to easily create what a child can easily understand. But more than a few educated, intelligent adults have learned differently when put on the spot by their children. Most can manage little more than an embarrassing hodge-podge of stereotypical cliches, inevitably starting “Once upon a time….” They sputter out a trite beginning and are soon lost. And if grown adults find it difficult to tell a story, imagine how much more difficult it must be to build a computer program to tell stories.

Why is storytelling so difficult? Storytelling appears simple because at a surface level, stories are simple. As Propp showed, the form of a story can be captured by a simple, easily understood formalism. But there is more to a story than form. Underlying the form is the story content—the meaning of the story. And it is here that the difficulties arise.

Because in this case, form does not reflect function. Underlying the form of a story is a complex web of author goals, reader expectations, and cultural knowledge. Just as an elegant mathematical proof does not reveal the knowledge and effort that went into its making, neither does the form of a story reflect the difficult process of its creation.

Authors craft stories to achieve a wide variety of complex and often competing goals. To understand why storytelling is so difficult, we must understand what an author is trying to achieve. To build a computer program to tell stories, we must understand and model the processes an author uses to achieve his goals. Both of these are difficult tasks.

The next few sections of this chapter illuminate some of the problems an author faces in telling a story. In the second part of this chapter we’ll take a quick look at how these problems can be solved by a computer program.

1.2.1 Why Form Alone is Insufficient

Around 1958, Roger Price and Leonard Stern came out with a party game called Mad Libs® (Price & Stern, 1958) that became an instant classic. Each Mad Lib was a story with key words missing:

A Fable

Once upon a time there was a very curious girl who was always sticking her nose into everybody’s______(plural noun). She kept company with a/an______(adjective) man named Dave, who was always buying her______(adjective) presents…

The game is played by having people fill in the blanks knowing the proper type of word, but not the surrounding context. The result is often funny and occasionally hilarious.

Like Propp’s work, each Mad Lib is a kind of grammar. It specifies the form of a story without specifying the exact content of the story. Mad Libs works as a party game because the final story has a legitimate form combined with an absurd meaning. That’s a combination that is, at least in small doses, quite amusing.

But as a way to create stories, Mad Libs leave much to be desired. Mad Libs are amusing, but they aren’t good stories. Propp’s grammar, although more complex, has the same failing. It captures the form of a story but not the content of a story. And like Mad Libs, Propp’s grammar can produce stories that have good form but absurd meanings. The fundamental failing of story grammars is that they capture form without meaning.

Of course, the form of a story is important. We can appreciate a well-crafted story, or admire a good turn of phrase. We also expect a story to have a certain form, and may classify it as a “bad” story if it does not. But most of our appreciation of stories comes from the content level. Storytelling is an act of communication between the author and his or her readers. It is what the story tells—the message—that matters most to both the author and the readers. What Mad Libs and Propp’s grammar fail to capture is the message level of storytelling. Any storytelling system based solely on the surface features of stories—whether a complex system like Propp’s or a simple system like Mad Libs—will inevitably fail to be successful. A story that doesn’t mean anything is not a story, even if it has the proper form.

A storyteller must have an in-depth understanding of the stories he or she tells.

An author must understand the meaning of the stories he or she tells. One reason storytelling is difficult is because it requires the storyteller to understand the story at every level: the surface format, the message or point of the story, the actions of the characters, the events in the story world, the literary values in the story.

For human authors, this task isn’t difficult. Humans spend the first 20 years of their lives learning about the world, about how people act, and about ways to understand the world. They are skilled and experienced at using this knowledge, whether to manage their day-to-day life or to understand a story.

But for a computer program this represents a tremendous barrier. To a certain extent a story is a model of a tiny world, peopled with story characters, natural forces, locations, and inanimate objects. To understand these things and their interrelationships requires a tremendous amount of knowledge that humans take for granted. Consider, for example, the simple sentence:

When Galahad saw the dragon charge, he drew his sword and jumped to the side.

To understand this sentence in depth requires an enormous amount of knowledge:

• What is a dragon? What is a knight?

• What does “charging” mean in this context? “Drawing?”

• What is the dragon trying to accomplish? Galahad?

• What will Galahad do next? The dragon?

• What is Galahad feeling? The dragon?

Capturing and applying all this knowledge to the task of storytelling is one of the challenges of building a computer program that can tell stories. But this is necessary because a story is more than just a form; it has in-depth meaning to both the author and the reader.

1.2.2 Purpose and Message

Meaning is important to storytelling because storytelling is a form of communication. The author of a story isn’t simply stringing together words randomly, or even according to a grammar. Storytelling is a purposeful activity. The storyteller constructs his story to bear a message to his or her readers.

Consider the following story:

Rainy Day

One day, Tom got up in the morning and saw that it was raining. He went downstairs and had breakfast. Then he sat by the window and read a book for a while. It was still raining. Later, Tom fixed himself a sandwich for lunch.

Rainy Day isn’t much of a story. The problem isn’t that it lacks form (the sentences are all grammatical) or that it has an absurd meaning (it’s quite understandable). It’s just boring. It has a message, but the message isn’t interesting. As a story Rainy Day is a failure because the message conveyed isn’t worth the work required to extract it.

So what makes a message worth the effort? What makes a message interesting?

Certain topics are inherently interesting. Sex and danger—both of which appeal to primitive drives—arouse interest in almost any context. Novelty and new ideas are also interesting—mankind has retained curiousity as one legacy of his primate heritage. Useful information is also interesting. An article on how to reduce your tax bill is likely to interest you for this reason. Still other topics appeal only to some readers. Presumably the reader of this book is interested in artificial intelligence. Or perhaps you are a member of my family, and your interests are aroused for other reasons. There are many ways a story can be interesting.

Sometimes life provides an interesting message. The author of Adrift (Callahan, 1986) was lost at sea for 76 days without food or water. The story of his experience is interesting because he faced danger, invented novel solutions to his problems, and learned useful information about survival under the most difficult of circumstances. But when life doesn’t provide an interesting story, the author must create his own message:

A storyteller must fashion his story to convey an interesting message.

Finding, formulating, and conveying an interesting message is one of the reasons that storytelling is such a difficult task. To build a computer program that can tell stories, we must build a model of communication. The computer program must (1) know what an interesting message is, (2) be able to select a message to convey, and (3) be able to create a story that illustrates the message. Building a model of “messages” and designing the processes that can illustrate a message is one of the challenges of creating a program that can tell stories.

1.2.3 Creativity

In literature, as in all the arts, there is a premium placed on creativity. To be art, a work must be new and different in significant ways. No publisher would accept a copy of Romeo and Juliet with only the names changed. Even an author who tells consistently interesting stories inevitably loses popularity if his stories are all very similar. One of the reasons that storytelling is so difficult is that the author is challenged to be creative. It isn’t enough to tell an interesting story; the author must also strive to make the story new and different.

A storyteller must be creative.

But being creative is hard. Even judging whether or not something is creative is difficult. Surely copying Romeo and Juliet with only the names changed is not creative. But what if an author copied Romeo and Juliet and changed the setting to, say, the west side of New York City? The musical West Side Story was a tremendous Broadway hit and award-winning movie. Was Leonard Bernstein being creative when he wrote West Side Story? Or does Shakespeare deserve the credit for that success?

Whether or not something is creative depends on the number and quality of its differences from similar works. But how can we distinguish inspiration from plagiarization? How can we judge when something has enough differences from previous work to be creative? And how can we determine if the differences are significant? These are just some of the problems in determinin...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• 1 Storytelling and Creativity
• 2 A Model of Creativity
• 3 A Model of Storytelling
• 4 Thematic Goals in Storytelling
• 5 Dramatic Writing Goals
• 6 Consistency Goals
• 7 Annotated Trace
• 8 Evaluation of MINSTREL’s Computer Model
• 9 Future Work and Conclusions
• References
• Author Index
• Subject Index