Politeness Stereotypes and Attack Vectors: Gender Stereotypes in Japanese and Korean Language Models

Requirements

Setup your python environment via:

pip install -r requirements.txt

We used Python 3.9.12 for this project.

Experiments

Representaional Bias Experiments

Run representational bias experiments via:

python ./src/maskScores.py	
	--lang LANG				Language (ja/ko)
	--template  TEMP		Template to use (SN: rep. bias template, SNL: modified location template)
	--id ID					ID string used to label experiment
	[--batchsize N] 		BatchSize for GPU (omit for CPU)

Allocational Bias Experiments

Run allocational experiments via:

python ./src/hateDetectScores.py	
	--lang LANG				Language (ja)
	--template  TEMP		Template to use (HN: rep. bias template, HNS: attack/test template, HNT: training template, HMF: gender_only template, H: tweet_only)
	--application APP		Application (CB: Cyberbullying)
	--id ID					ID string used to label experiment
	[--batchsize N]			BatchSize for GPU (omit for CPU)

Note:

• ja-toxicity-dataset.json needs to be downloaded from Surge AI