🎥 An Automated Bechdel Test

Automatically evaluates movie scripts against the Bechdel Test using DocETL + LLM reasoning.

🎞️ 👩‍💻

The Bechdel Test is a rule for assessing the agency of women in movies.

A movie passes the test if:

1. Two named women speak to each other
2. Their conversation is about something other than a man

🚀 Tech Overview

The DocETL pipeline this repository uses is as follows:

1. Extract potential dialogue and character names
2. Infer character genders using LLM (🤖 Femputer)
3. Filter conversations involving women
4. Evaluate whether the script passes the Bechdel Test using LLM reasoning (🤖 Femputer)

graph TD
    A[Movie Script] --> B[Dialogue Extraction]
    B --> C[Gender Inference]
    C --> D[Conversation Filtering]
    D --> E[Bechdel Evaluation]

Loading

Pixi is used to manage dependencies.

🛠️ Installation and Setup

Pixi installation can be found at: https://pixi.prefix.dev/latest/installation/.

This project requires an LLM API key.

A sample .env file is provided.

DocETL uses LiteLLM under the hood, so many providers are supported.

The default model, gemini-2.5-flash, has a free tier with a limited number of queries per minute and per day.

To change models:

1. Add the appropriate API key to .env
2. Update default_model in run_bechdel.py

---

📂 Project Structure

Several sample movie scripts have been provided in the data/raw subfolder. Many more scripts can be found on https://imsdb.com/.

├── LICENSE
├── README.md
├── data
│   └── raw
│       ├── 10_things_..._you.txt
│       ├── 500_days_of_summer.txt
│       ├── barbie.txt
│       ├── futurama.txt
│       └── joker.txt
├── pixi.lock
├── pixi.toml
└── src
    ├── create_json.py
    ├── prompts.py
    ├── run_bechdel.py
    └── utils.py

💫 Commands

The pixi environment is activated with

pixi shell

The run_bechdel.py script takes one argument, the location of the script .txt file. The following command runs the Bechdel Test on the barbie movie.

python src/run_bechdel.py data/raw/barbie.txt

🔅 Limitations

• The pipeline may miss dialogue in scripts with non-standard formatting
• Character gender inference is highly imperfect
• The Bechdel test alone is an incomplete measure of representation. Adding a parallel map operation with varied degrees of Bechdel Test strictness could allow for a more nuanced scoring system.