Using word embeddings to investigate cultural biases.

Area of Science:

• Computational Social Science
• Psychology
• Natural Language Processing

Background:

• Word embeddings represent word semantics quantitatively, useful for analyzing large text archives.
• Bias analysis is a key concern in social psychology, but traditional methods can miss subtle cues.

Approach:

• This article introduces social psychologists to word embedding construction and bias measurement techniques.
• Embeddings are used to measure bias along bipolar dimensions, similar to semantic differential scales.
• Recent studies demonstrating bias detection and temporal changes in embeddings are reviewed.

Key Points:

• Word embeddings provide valid and reliable estimates of social bias.
• They can detect subtle biases not explicitly communicated in text.
• Bias detection in embeddings correlates with real-world discriminatory practices and changes over time.

Conclusions:

• Word embedding research offers novel quantitative measures for studying prejudice and stereotyping.
• This methodology can extend the understanding of the bias environment influencing human thought and action.
• Social psychologists can leverage word embeddings to analyze bias in digital text data.