# Finding the Chekhov's Gun

Our focus is on computational methods for analyzing narrative structure and coherence using graphs, with attention to AI and NLP tools.

---

## Step 1: Narrative Graph Construction

**Objective:** Extract and structure the narrative into a graph of events, characters, and objects.

- **NarCo (Xu et al., 2024)**
Uses retrospective question generation with GPT-4 to construct a coherence-based narrative graph.
_[ACL Anthology](https://aclanthology.org/2024.acl-long.317/)_

- **Hamilton et al. (2025)**
Constructs large-scale character interaction networks from novels using LLM-based structured prompts.
_[arXiv](https://arxiv.org/html/2502.19590v1)_

---

## Step 2: Detecting Narrative Promises

**Objective:** Identify early narrative elements that signal potential future relevance.

- **NarCo (Xu et al., 2024)**
Retrospective coherence questions can reveal narrative tension and highlight "promises" through inferred expectations.

- **GraphPlan (Chen et al., 2021)**
Learns plausible event progressions from corpora and could be used to flag unusual or unresolved setups.
_[ACL Anthology](https://aclanthology.org/2021.inlg-1.42/)_

---

## Step 3: Tracing Narrative Payoff

**Objective:** Determine whether narrative elements introduced early are used meaningfully later.

- **Yan & Tang (2023)**
Builds temporal and causal event-centric knowledge graphs from stories to model evolution and coherence.
_[PubMed](https://pubmed.ncbi.nlm.nih.gov/37128597/)_

- **NarCo (Xu et al., 2024)**
Graph traversal using coherence links reveals how and whether early elements are revisited.

---

## Step 4: Evaluating Satisfaction of the Principle

**Objective:** Assess whether the “promise” was fulfilled in a meaningful way in the story.

- **NarCo (Xu et al., 2024)**
Retrospective question-answerability directly assesses whether parts of the narrative are connected logically.

- **GENEVA (Rao et al., 2024)**
Branching narrative graphs allow multiple storyline tracking; helpful for visualizing diverging or resolved threads.
_[arXiv](https://arxiv.org/abs/2311.09213)_

---

## Step 5: Reporting and Visualization

**Objective:** Visualize narrative graphs and identify satisfied/unsatisfied Chekhov elements.

- **GENEVA (Rao et al., 2024)**
Provides a system for visualizing branching narrative DAGs, which can be adapted to show payoff of narrative setups.

---

## Summary Table

| Method Step | Relevant Papers |
|---------------------------|----------------------------------------------------------------------------------|
| Narrative Graph Construction | Xu et al. (2024), Hamilton et al. (2025) |
| Detecting Narrative Promises | Xu et al. (2024), Chen et al. (2021) |
| Tracing Narrative Payoff | Xu et al. (2024), Yan & Tang (2023) |
| Evaluating Satisfaction | Xu et al. (2024), Rao et al. (2024) |
| Reporting and Visualization | Rao et al. (2024) |