Censorship of Online Encyclopedias: Implications for NLP Models

By Eddie Yang, Margaret E. Roberts. 2021

22 citations recorded by the source corpus.

While artificial intelligence provides the backbone for many tools people use around the world, recent work has brought to attention that the algorithms powering AI are not free of politics, stereotypes, and bias. While most work in this area has focused on the ways in which AI can exacerbate existing inequalities and discrimination, very little work has studied how governments actively shape training data. We describe how censorship has affected the development of Wikipedia corpuses, text data which are regularly used for pre-trained inputs into NLP algorithms. We show that word embeddings trained on Baidu Baike, an online Chinese encyclopedia, have very different associations between adjectives and a range of concepts about democracy, freedom, collective action, equality, and people and historical events in China than its regularly blocked but uncensored counterpart - Chinese language Wikipedia. We examine the implications of these discrepancies by studying their use in downstream AI applications. Our paper shows how government repression, censorship, and self-censorship may impact training data and the applications that draw from them.

Subfields: Artificial Intelligence, Communication, Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Iason Gabriel
2. Po-Sen Huang
3. Jonathan Uesato
4. William S. Isaac
5. John F. J. Mellor
6. Borja Balle
7. Geoffrey Irving
8. Myra Cheng

In Communication

1. William S. Isaac

In Natural Language Processing

1. Iason Gabriel
2. John F. J. Mellor
3. Borja Balle
4. Geoffrey Irving
5. Myra Cheng
6. Lisa Anne Hendricks
7. Po-Sen Huang
8. Laura Rimell

In Safety, Robustness & Interpretability

1. Iason Gabriel
2. Po-Sen Huang
3. William S. Isaac
4. John F. J. Mellor
5. Borja Balle
6. Geoffrey Irving
7. Myra Cheng
8. Lisa Anne Hendricks

Citations in the corpus

• Ethical and social risks of harm from Language Models (2021, background): “In a distinct feedback loop, censorship of web text, for example of online encyclopedias, can then affect the quality of a LM trained on such data (Yang and Roberts, 2021).”
• Mitigating harm in language models with conditional-likelihood filtration (2021, methodology): “For example, they could use the method to remove oppositional speech from training data and encourage language models to proliferate their ideas [Yang and Roberts, 2021].”