Story Generation After TALE-SPIN
Proceedings of the Seventh International Joint Conference on Artificial Intelligence. Natural Language. Pages 16-17.
© 1981 International Joint Conferences on Artificial Intelligence
Abstract
TALE-SPIN, the last major AI attempt at story generation, approached the problem of making up stories primarily from the perspective of an impartial world simulator. Alll1IQl is a program (under development) which generates stories as a creative reasoner in pursuit of her own narrative goals. It is thus intended to simulate an author·s mind as she makes up a story, rather than the world as things happen in it. The four major forces driving the story generation process, according to the AUTHOR model, are (1) author intentionality, (2) conceptual reformulation, (3) reminding, and (4) the opportunity enhancement metagoal.
- B.EFEB.ENCES Klein, Sheldon, et al. (1976). Automatic Novel Writing: A Status Report. Technical Report #186. Computer Sciences Department. University of Wisconsin-Madison.
- Kolodner, Janet L. 0980). Retrieval and Organizational Strategies in Conceptual Memory; A Computer Hodel. Ph. D. Thesis. Research Report #187. Department of Computer Science. Yale University, New Haven, CT.
- Meehan, James R. (1976). The Hetanovel: Writing Stories By Computer. Ph. D. Thesis. Research Report #74. Department of Computer Science. Yale University, New Haven, CT.
- Schank, Roger C. (in press). Dynamic Memory: A Theory of Learning in Computers and People.
References were extracted automatically from the PDF and may contain errors.
Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.
1. Wor\d Siaulation Baaed Story Generation
TALE-SPIN [Meehan, 1976] is 1 ·program which made up Aesop-fable-like conceptual stories, which it then expreued in natural language. Each story TALE-SPIN made up was based on a model of the storyworld, including the characters in it and their goals, personalities, and interpersonal relations. Stories were spun by working out consequences of the model. Thus, central to TALE-SPIN was the idea of world aiaulation.
The three levels TALE-SPIN was most concerned with simulating were:
- character intentionality - e.g., if Joe Bear was hungry, he would try to get something to eat
- sociai constraints - e.g., if Joe Bear believed Irving Bird to be his friend, he could ask an outright favor; if Joe Bear believed their relation to be a bit cooler, he would feel obliged to offer something in return (such as a worm)
- physical causality e.g., if Irving Bird ate a worm, the worm would cease to exist; after "fooling around",
* As opposed to Sheldon Klein's system which only produced story texts and which is incapable, in principle, of understanding the stories it '~akes up" [Klein et al., 1976].
STORY GENERATION AFTER TALE-SPIN
Natalie Dehn
Yale University Department of Computer Science Artificial Intelligence Project P.O. Box 2158 10 Hillhouse Avenue New Haven, CT. 06520
characters felt "wiped out"
World simulation, however, fails to reflect the process that an author goes through in making up a story - storyworlds are developed by authors as needed, frequently as post hoc justification for events that the author has already decided she wants as part of the story. As illustration, let us consider the question of why, in the Star Wars sequel, "The Empire Strikes Back", Princess Lei a and Solo go to visit Landeau.
If one asks a reader (or viewer) this question, the answer will be along the lines of, "They needed a place to hide from imperial stormtroopers until they got the hyperdrive fixed, and Landeau was an old buddy of Solo's, and ··· ".
Were this story being made up by a world simulation approach, Landeau would have to have been invented as part of the initial specification of the story world, back before any of the story action had been decided. The decisions that Solo liked Landeau a lot and trusted him some but that Landeau was dishonest, would also have to have been made before starting the action. Landeau would then have remained dormant until it developed that Leia and Solo ran into hyperdrive problems while fleeing imperial stormtroopers. At this point, storyworld causal neceessity would have to require Solo to do some further planning and thence think of asking Landeau for help.
Setting up an initial story world such that things work out in an interesting way is, at the very least, tricky business.
2. Author Intent Baaed Story Generation
A more plausible mechanism for making up this story, however, is one driven by author intent, in which Landeau would not be invented until needed. His invention would be motivated by the need to get Solo out of action for a while so that Luke Skywalker could get the girl, get the credit for some heroic behavior, and eventually rescue Solo and repay his debt.
The difference between a world simulation and an author-intentional model of the story generation process hinges on the difference between character and author goals. While a character's purpose (like a real person's) is to have things go well for himself and for those he cares about, the author's purpose is to make up a good story. The author can think about the story world, but is not part of it; she cannot be personally threatened by dangers in the story world, nor can she take advantage of opportunities that arise there. Though the author will frequently "intervene" to ensure that things work out well for a character or to ensure that he loses, she ia never in the relation of goal cooperation or goal coapetition (as described in [Wilensky, 197 8]) with the character, because she ia functioning on a different level than the character whose fate she determines. To the author, the character is mere modelling clay like the rest of the story world which serves as his environment.
The claim, then, is that making up a story is a deeply goal-based activity, the final story standing as the achievement of a complex web of author goals. The author"s goals serve as a sort of scaffolding in constructing the story; they are no longer directly visible in the final story, but are reflected in the storyworld situations and resolutions they gave rise to.
3. Conceptual Reforaulation
Author goals do not remain static as story generation progresses. In fact, a large part of the work of making up a story is the successive reformulation of author goals. Story ideas, fragments, etc. also go through numerous conceptual drafts, each reformulation supplying new perspectives which aid further retrieval of related material and constraints from the author"s memory.
Conceptual reformulation is an activity that happens on many different levels in the story generation process - e.g., getting the author from an idea to a kernal episode, from a kernal episode to a sequence of episodes, from an characterization of someooe to an episode illustrating that characterization, from a change in relation between two characters to a dialogue triggering or reflecting that change.
Suppo·e, for instance, the author has developed the following kernal episode:
Shy person (protagonist) in need of employment gets job as salesman but is incapable of facing $DOOR-TO-DOOR. Failure seems i111111inent when protagonist comes up with (and tries) $FORM-LETTER as substitute script towards same goa 1. flying colors! Succeeds with
After several passes of succeuive reformulation, the story sketch might now be:
- EPl: anecdote demonstrating shyness of protagonist
- EP2: episode demonstrating closeness of protagonist to mother [further development of protagonist's shyness; drama enhancer for EP4]
- EP3: episode revealing continuing unemployment of protagonist [and his interview avoidance?]
- EP4 : protagonist's expensively) increases mother falls ill; urgency seriously (and of employment
- EP5: starts ·responding by mail to newspaper want ads [foreshadowing of final solution]
- EP6: series of job application failures, leading to decreasing discrimination in selection of ads to respond to
- EP7: success (?) opportunity as co11111issioned door-to-door salesman
- EP8: details of conditions (through letter): should show up at office for one .day of orientation, starting on road the next, commission system with minimum monthly quota
- EP9: first day at work: obnoxious friendliness of salesmen
- EPlO: $DOOR-TO-DOOR script presented, in story round among salesmen, in exceedingly unfavorable light
- EPll: at home after first day at work; in shock over encounter; in further shock at thought of coming month
- EP12: more bad news about mother"s health, increasing drama
- EP13: agonizing by protagonist about about mess be's in, and recapitulation of it, leads to idea of $FORM-LETTER
- EP14: writing and sending out letters, applying old mass mailing system [from EP5]
- EP15: first mixed results coming in; pessimistic interpretation
- EP16: orders pouring in; rave success
- EP17: epilogue: appreciation and money pouring in from company
There is insufficient space to list all the reformulations made, but the following handful can serve as illustration. The author goal to
- aake the 1tory plau1ible got reformulated into the author goal to
- justify the protagonist getting into such an unlikely situation for a shy person
- justify the protagonist coming up with the solution he did
- aake the 1tory dramatic got reformulated into the author goal to
- emphasize the aversiveness of the salesman situation
- reduce the obviousness of the letter campaign solution
- illu1trate key fact· got reformulated into the author goal to make up episodes illustrating and presenting
- the protagonist'& shyness
- the protagonist's mother"s illness
Each of these reformulations, of course, will undergo numerous further ones as the story develops further.
4. Author M~ry and Story Generation
As TALE-SPIN was basically simulating the world, rather than an author's mind, "memory issues" were restricted to matters of comput;r data structures and representation of knowledge. AUTHOR, however, relies very heavily on a memory that is human-like in ita organization and contents. This memory needs to contain not only facts about the storyvorld thus far constructed, but also prior knowledge of memorable episodes, characters, etc. in the author's life. How all this is organized and accessed is very critical to the story generation process. Of special importance are memory search strategies [Kolodner, 1980] and reainclina mechanisms [Schank, in press]. A "by-product" of conceptual reformulation (actually, this may be why it is so effective in the first place) is greater opportunity for relevant reminding&.
Reminding playa two important roles in the story generation process:
- as a source of relevant external material - incidents, characters, etc. from the author' a life - to be worked into the story, and
- as a way of keeping track of internal material, narrative goals, etc. allowing the author to become distracted (another important memory phenomenon) yet return to earlier ideas as they become relevant again.
5. Creative Reasonina
Story generation is predominantly a process of creative reasonina. As such, there are two essential characteristics of the process that any model of story generation should capture: the degree to which the process is deliberate and directed, and the degree to which the process is serendipitous.
Alt'hough story generation is driven by the author's narrative goals, story generation is not planful in the traditional, top-down, means-end analysis sense. Rather, there are two metagoals:
- thst of achieving the current narrative goal, and
- that of finding better narrative goals to puraue.
* Meehan was very concerned with restricting TALE-SPIN's ltnovlecl&e to the kinds of knowledge people have; thus, for instance, he used a symbolic, rather than a more efficient numerical, representation for storyvorld geography. Nonethele81, it vas sufficient, for his purposes, to indeK "facts" in a very simplistic manner.
The author (or any creative thinker) has to be able to pursue the former, while recognizing that it is the latter that is actually the more important.
There are several essential characteristics of creative reasoning:
- sensitivity to unforeseen opportunities, when one has the good fortune for them to arise
- villingneu to be distracted from what one was doing, if something better comes up
- a process of successive reformulation, dramatically increasing the probability that useful opportunities will arise
- having at all times a sense of direction which serves both
- to keep occupied her goals the author usefully in progressing towards
- to provide new environments in which fortuitous opportunities are likely to arise.
This last characteristic, the opportunity enbanceaent metagoal, accounts for the dual directed-yet-serendipidous nature of creative thinking: why conceptual reformulation is so powerful a mechanism, and how a phenomenon such as being reminded can be constructively channeled in the creative thought process comprising story generation.
B.EFEB.ENCES
- Klein, Sheldon, et al. (1976). Automatic Novel Writing: A Status Report. Technical Report #186. Computer Sciences Department. University of Wisconsin-Madison.
- Kolodner, Janet L. 0980). Retrieval and Organizational Strategies in Conceptual Memory; A Computer Hodel. Ph. D. Thesis. Research Report #187. Department of Computer Science. Yale University, New Haven, CT.
- Meehan, James R. (1976). The Hetanovel: Writing Stories By Computer. Ph. D. Thesis. Research Report #74. Department of Computer Science. Yale University, New Haven, CT.
- Schank, Roger C. (in press). Dynamic Memory: A Theory of Learning in Computers and People.
- Wilensky, Robert (1978). Understanding goal-based stories . Ph. D. Thesis. Research Report #140. Department of Computer Science. Yale University, New Haven, CT.
@inproceedings{ijcai1981p4,
title = {Story Generation After TALE-SPIN},
author = {Dehn, Natlie},
booktitle = {Proceedings of the Seventh International Joint Conference on
Artificial Intelligence, {IJCAI-81}},
publisher = {International Joint Conferences on Artificial Intelligence},
editor = {Patrick J. Hayes},
pages = {16--17},
year = {1981},
month = {8},
note = {Natural Language},
url = {https://ijcai.org/proceedings/1981/4},
}
TY - CPAPER TI - Story Generation After TALE-SPIN AU - Dehn, Natlie T2 - Proceedings of the Seventh International Joint Conference on Artificial Intelligence, IJCAI-81 PB - International Joint Conferences on Artificial Intelligence PY - 1981 SP - 16 EP - 17 UR - https://ijcai.org/proceedings/1981/4 ER -Copied
Machine-readable versions of this paper, for tools and AI agents: