Paper Presentation: A Survey on Story Generation Techniques

This talk presents the findings of Kybartas and Bidarra in their survey paper on story generation techniques:
https://graphics.tudelft.nl/Publications-new/2016/KB16/Survey.pdf

Transcript

1. A Survey on Story Generation Techniques for Authoring Computational Narratives Paper by Ben Kybartas and Rafael Bidarra Presentation by Chris Martens CSC 791: Generative Methods Monday, October 24, 2016 1
2. Purpose of this paper 2 Q: What is the purpose of any survey paper?
3. Purpose of this paper 3 Define common goals of a research community Review progress on those goals Classify progress along different axes Identify gaps for future work
4. Human-Computer Collaboration 4 Which parts does the computer do? Which parts does the human do?
5. Outline 5 Terminology Examples Classification Scheme Examples to Test
6. Terminology 6
10. Story vs. Space 10 Playwright vs. Director script: characters act structure scenes lines performance: actors props set delivery
12. Plot Automation 12 Manual Structure Template Constrained Automated
13. Plot Automation 13 Manual Structure Template Constrained Automated Computer is present at a mostly invisible level
14. Plot Automation 14 Manual Structure Template Constrained Automated Computer is present at a mostly invisible level Attempts to minimize author involvement as much as possible
15. 15 Examples
16. 16 Manual Authoring Tools Interactive Fiction Creation Programming Education Therapy
17. 17 Storyspace Twine Inform 7
18. Scrivener 18
19. 19 Plot Generation Systems Grammar-based Planning-based Other
20. Grammars as Templates 20 Grammar describes an outline of the narrative but does not populate it with existents.
21. Proppian Grammars 21
22. Cohn Comic Grammars 22
23. Grammars as Complete Plot Generators 23 GESTER: rules defining relations between existents TEATRIX: grammar as director, player as performer
24. Grammars as Complete Plot Generators 24 GESTER
25. ReGen: Story Graph Rewriting 25
26. Planning-Based Systems 26 Story Canvas
27. Porteous et al. “Applying planning to interactive storytelling” 27
28. Case-Based Reasoning 28 Engage/Reflect Cycle Reflection uses a corpus of previous stories to evaluate and revise the current story on the basis of novelty. MEXICA
29. Social Physics Engines 29 Prom Week
30. Social Physics Engines 30 Versu
31. 31 Space Generation Systems
32. Text-to-Image Systems 32
33. 33 Skyrim Radiant Quests
34. 34 GameForge: As seen in the PCG Book Story as input Valid space as output
35. 35 Complete Story Generation Systems
36. 36 Fabulist Initial State Revision Example: location of a hidden weapon needed for stealthy assassination
37. 37 Generating space on the fly Virtual Storyteller: Late commitment Inspired by improvisational theater Example: Character want to fight each other? Spawn some weapons Similarly, Li and Riedl: Gadget generation
38. 38 Scheherezade Crowd-sourced story choices
39. 39 Universe Generating backstories for characters with family tree simulation Temporal coherence: if plot requires two characters to fall in love, their birthdates must be within a reasonable range of one another
40. 40 Dwarf Fortress “Emergent Storytelling”
41. 41 Classification Scheme
42. Plot Automation 42 Manual Structure Template Constrained Automated
43. Plot Automation 43 Structure Computer provides plot structure, but no specific events or event orderings. e.g. Scrivener
44. Plot Automation 44 Template Computer provides plot template with events in order, but instantiation with existents is left to the human author. e.g. Propp grammars
45. Plot Automation 45 Constrained Author provides initial state or other narrative constraints; computer gives complete plot populated with existents. e.g. GME, Mexica, Prom Week
46. Space Automation 46 Manual Modification Simulation Constrained Automated
47. Space Automation 47 Modification Author provides a starting point, but the computer can modify it, e.g. to suit the plot structure. e.g.: initial state revision; late commitment
48. Space Automation 48 Simulation New spatial content is generated by simulating interactions between hand-authored existents, resulting in an initial state for story. e.g.: Universe’s family tree generation
49. Space Automation 49 Constrained New spatial content is generated to satisfy authored constraints. e.g.: Dwarf Fortress
52. 52 Examples to test classification
54. Ice-Bound 54
55. Epitaph 55 https://mkremins.itch.io/epitaph