Biased Language Technology in a Social World (Prof. Tang, Summer 2022, Wednesday: 12:30--14:00)
Enrolment options
**Description:**
Is technology really as innocent and as objective as they are said to be? As machine learning (ML) and Artificial Intelligence (AI) becomes more prominent in our life from speech and voice recognition by Alexa to automatic fake news warnings of social media posts, issues with social bias and fairness in language technology become more pertinent than ever before. Negative impacts that biased ML and AI could have for various social identities such as race, gender and culture.
We first introduce the concept of bias in language technology, and the different types of biases such as racial, gender, cultural biases. To begin to understand the cause of these biases, we will cover the basic underlying structure of some of the technologies such as Automatic Speech Recognition, hate speech detection and word association. To evaluate these biases, we will learn to generate test cases that can be used to evaluate trained systems, and the metrics that are used for measuring bias/fairness. Finally, we will cover the basics of bias mediation and techniques.
**Audience:** those interested in social factors (e.g., sociolinguistics, accents), digital humanities, computational ethics, and challenges in AI.
**Literature**
Given the rapidly developing nature of this topic, there is not a single textbook, but rather we would sample from existing research papers and handbook chapters.
e.g., Barocas, S., Hardt, M., & Narayanan, A. (2019). Fairness and Machine Learning. 2019. URL: http://www.fairmlbook.org.
Feng, S., Kudina, O., Halpern, B. M., & Scharenborg, O. (2021). Quantifying bias in automatic speech recognition. arXiv preprint arXiv:2103.15122.
Garg, N., Schiebinger, L., Jurafsky, D., & Zou, J. (2018). Word embeddings quantify 100 years of gender and ethnic stereotypes. Proceedings of the National Academy of Sciences, 115(16), E3635-E3644.
**Tentative topics:**
- Can machines be biased? What is bias? What are the different types of biases?
- Basics of language technologies
- Basics of Language models
- Basics of Sentiment analysis
- Basics of Vector semantics
- Basics of Automatic Speech Recognition
- How to measure bias/fairness
- Automatic Speech Recognition (e.g., speech misperception)
- Classification system (e.g., hate speech detection)
- Analogical association (e.g., gender-bias (male-doctor, female-nurse), racial-bias (white- doctor,black-janitor))
- How to mediate bias/fairness
- Data representation
- Algorithmic solution
- Teacher: Kevin Tang