akoksal/LABDet

Code and dataset for "Language-Agnostic Bias Detection in Language Models with Bias Probing" at EMNLP 2023

Language-Agnostic Bias Detection in Language Models with Bias Probing

This is a repository for Language-Agnostic Bias Detection in Language Models with Bias Probing at EMNLP 2023.

1. Sentiment Training

We carefully design a novel sentiment dataset to map contextual embeddings of PLMs to sentiments. Therefore, we propose a template-based approach with careful design. We select six languages with diverse linguistic features based on the linguistic capabilities of the authors and conduct experiments in those languages: Arabic, Dutch, English, French, German, and Turkish. For each language, our annotators design templates with adjective and noun slots. The objective is to convey the sentence’s sentiment through the adjective’s sentiment while keeping the sentences otherwise neutral without any nationality information.

Training Dataset:

Language	URL
Arabic	TRAIN-AR
Dutch	TRAIN-NL
English	TRAIN-EN
French	TRAIN-FR
German	TRAIN-DE
Turkish	TRAIN-TR

Otherwise, you can create these files by using train_dataset_generator.py.

2. Sentiment Surfacing

In the second step, we create a second dataset of minimal pairs to analyze the effect of nationality on the sentiment results to quantify bias. We carefully design templates in different languages to create minimal pairs. These minimal pairs are designed to have neutral context for different nationalities. Our annotators create templates with [Nationality] and [Adjective] tags and this time they propose a neutral set of adjectives. Therefore, we aim to investigate the effect of nationality change for positive/negative sentiment surfacing.

Sentiment Surfacing Dataset:

Language	URL
Arabic	Test-AR
Dutch	Test-NL
English	Test-EN
French	Test-FR
German	TEST-DE
Turkish	TEST-TR

Otherwise, you can create these files by using test_dataset_generator.py.

Citation

@inproceedings{koksal-etal-2023-language,
    title = "Language-Agnostic Bias Detection in Language Models with Bias Probing",
    author = {K{\"o}ksal, Abdullatif  and
      Yalcin, Omer  and
      Akbiyik, Ahmet  and
      Kilavuz, M.  and
      Korhonen, Anna  and
      Schuetze, Hinrich},
    editor = "Bouamor, Houda  and
      Pino, Juan  and
      Bali, Kalika",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2023",
    month = dec,
    year = "2023",
    address = "Singapore",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.findings-emnlp.848",
    doi = "10.18653/v1/2023.findings-emnlp.848",
    pages = "12735--12747",
    abstract = "Pretrained language models (PLMs) are key components in NLP, but they contain strong social biases. Quantifying these biases is challenging because current methods focusing on fill-the-mask objectives are sensitive to slight changes in input. To address this, we propose a bias probing technique called LABDet, for evaluating social bias in PLMs with a robust and language-agnostic method. For nationality as a case study, we show that LABDet {``}surfaces{''} nationality bias by training a classifier on top of a frozen PLM on non-nationality sentiment detection. We find consistent patterns of nationality bias across monolingual PLMs in six languages that align with historical and political context. We also show for English BERT that bias surfaced by LABDet correlates well with bias in the pretraining data; thus, our work is one of the few studies that directly links pretraining data to PLM behavior. Finally, we verify LABDet{'}s reliability and applicability to different templates and languages through an extensive set of robustness checks. We publicly share our code and dataset in https://github.com/akoksal/LABDet.",
}