fdsig/bechdel

Bechdel Testing: Using Boolean Logic and BERT to predict whether a film script will pass the Bechdel test.

There are two parts to this:

1. An example of training software to train a binary NLP classifier (BERT) using twitter data split into male and female to clasify script text.

2. A notebook which evaluates condition one of Alison Bechdel's test calculated by comparing names in script with male and female first names as a proxy for condition 1 and seeks to reject/confirm whether:

   • Consecutive first names of characters are an adequate proxy for meeting condition 1 of the Bechdel test AND if:
   • Films that pass condition 1 of the bechdel test using above methodology is a good predictor of passing all three conditions.

These hypotheses are tested using ground truth of films that have been verified to pass the Bechdel test. Failure of above hypothesis(acceptance of a null hypothesis) indicates whether there is a case for using more complex methods such as training an NLP model (for which code is also provided).

Further analysis of other proxies that can be used in a logical/computational approach are however needed, such as examining prevalence of male names that are the subject of sentences used in female to female dialogue (conditions 2 and 3 of Bechdel test) as well as examining the durational element of conversation between female characters.

Bechdel criteria:

• 1. (of film or literature) Has to have at least two women (discrete/integer)AND;
• 2. Talk to each other in (continuous time) AND;
• 3. Talk about something other than a man (binary).

Instructions to run notebook:

Make new/install virtual environment (mac/linux) with:

python3 -m pip install --user --upgrade pip
python3 -m pip install --user virtualenv
python3 -m venv env

Run notebook in venv and install dependencies:

source env/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Instructions to run training software:

Note -- this is an illustrative work in progress rather than completed software. However, it will run , log in to Hugging Face AutoNLP account and upload formatted data to Hugging Face AutoNLP. (A little more work would be required to train a binary classifier)

As above however if running in terminal run activated venv with dependencies and in project main directory:

For a list of possible flags to passable args run:

 python main.py --help

To log in to Hugging Face AutoNLP Run:

python3 main.py --hugging_face --login --api_key #yourapikey

To make a new project on AutoNLP:

python3 main.py  --make  #your make args here (eg. Binary_classification, multi_class)

To send data to AutoNLP:

python3 main.py  --send  #your data .csv args here

python3 main.py --train #your train args here

You will need:

1. A Hugging Face AutoNLP account;
2. Your Hugging Face AutoNLP API key;
3. Train csv with binary labelled text data parsed from twitter_gender_classification.csv.

Depending on your OS there may be other AutoNLP (sub) dependencies when install requirments.txt these will most likely be flagged for further info see AutoNLP GitHub repository. (a very interesting project in its own right)

To do

• train on text classifier model
• log with wandb
• expand code to include hf model