Towards Analogy-Based Story Generation.

Jichen Zhu, Santiago OntañónICCC 2010 · OpenReview

Abstract

Current AI storytelling often feels mechanical due to an over-reliance on goal-driven planning algorithms. This paper proposes a shift toward analogy-based story generation to capture the emotional depth and subjective nuance of human literature. By introducing a three-dimensional framework for classifying narrative systems and presenting the Riu system as a case study, the authors demonstrate how computational analogy can simulate complex literary techniques like "stream of consciousness."

Introduction

Most computational narrative systems, such as the classic Tale-Spin, treat storytelling as a planning problem where characters execute sequences of actions to reach specific goals. While logically consistent, this approach often results in a narrow "problem-solving" aesthetic. The authors argue that analogy—a fundamental human cognitive faculty—is significantly under-utilized in AI storytelling. They aim to move from "high focus" analytical planning toward "low focus" creative activities, such as associative thinking and emotional resonance, to bridge the gap between logical sequences and literary art.

A Framework for Analogy in Narrative

To organize the field, the authors propose a three-dimensional framework for classifying existing systems based on Scope, Technique, and Representation.

Scope and Narrative Hierarchy

The "Scope" dimension defines where the analogy is applied within the narrative structure. The authors adopt a standard narratological hierarchy to define these levels.

Hierarchy of Narrative components

As shown in Figure 1, a Narrative is divided into Story (the "what") and Discourse (the "how"). The Story further decomposes into Events (Actions and Happenings) and Existents (Characters and Setting). The authors' survey found that most current systems focus on the Story or Existent levels, leaving the Discourse level—the stylistic way a story is told—largely unexplored.

Taxonomy of Systems

The authors analyzed several landmark AI storytelling systems to populate their framework, categorizing them by their technique (identification vs. generation) and internal representation.

Classification of analogy-based narrative systems

Table 1 (Figure 2) illustrates that while systems like Minstrel and ProtoPropp use Case-Based Reasoning (CBR) for story generation, others like MEXICA focus on the "existents" level using engagement/reflection cycles. This taxonomy reveals a cluster of research around plot and character, highlighting a significant opportunity for research into analogical discourse.

Case Study: The Riu System

To demonstrate the power of analogy, the authors introduced Riu, an interactive narrative about a robot named Ales who has lost his memories. Riu utilizes the Structure Mapping Engine (SME) to perform two key tasks:

  1. Memory Retrieval: The system uses surface similarities (e.g., a squeaky door) to trigger memories with structural similarities (e.g., a past event with a similar logical structure).
  2. Analogical Inference: When a player chooses an action, the system "imagines" the outcome by mapping a past memory onto the current situation.

This allows for "subjective" behavior. For instance, a character might hesitate to perform a task not because it is logically impossible, but because it triggers a tragic analogical association. This creates a "stream of consciousness" effect reminiscent of Virginia Woolf’s Mrs. Dalloway.

Key Insights and Future Directions

The paper highlights that analogy is better suited for "aesthetic affordances" than planning. While planning excels at causality, analogy enables the simulation of human subjectivity. However, representing the complex data structures required for analogy at the "Narrative" or "Discourse" level remains a significant challenge.

Future work should focus on hybrid systems that combine the structural strengths of planning with the aesthetic flexibility of analogy. Exploring the full range of aesthetic possibilities beyond "stream of consciousness" could lead to entirely new genres of AI-driven interactive fiction.

Summary updated

BibTeX
@inproceedings{zhu2010towards,
  author = {Jichen Zhu and Santiago Ontañón},
  title = {{Towards Analogy-Based Story Generation.}},
  booktitle = {ICCC},
  year = {2010},
  url = {https://openreview.net/forum?id=abmzXBo24TRQ},
}

Loading related content...

Figure