Machine Learning as a Model for Cultural Learning: Teaching an Algorithm What it Means to be Fat.

Area of Science:

• Computational linguistics
• Cultural psychology
• Media studies

Background:

• Public culture, particularly media, influences perceptions of body weight.
• The cognitive mechanisms by which individuals internalize public cultural meanings remain unclear.

Purpose of the Study:

• To propose neural word embeddings as a computational model for schema learning.
• To empirically investigate how individuals learn and activate cultural schemas related to body weight from natural language data.

Main Methods:

• Utilized neural word embeddings, a computational text analysis technique.
• Extracted four lower-order schemas (gender, moral, health, class) from news articles concerning body weight.
• Quantified the association between body weight language and these schemas.

Main Results:

• Demonstrated that neural word embeddings can model schema learning and activation.
• Identified specific gender, moral, health, and class meanings associated with body weight in public discourse.
• Quantified how body weight language "fills in the blanks" for these schemas.

Conclusions:

• Schema learning is a key mechanism for integrating public culture into personal understanding.
• Neural word embeddings offer a viable method for studying cultural learning and schema activation.
• Findings highlight the potential for machine learning models to perpetuate societal biases present in language data.