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

Abstract

Public culture is a powerful source of cognitive socialization; for example, media language is full of meanings about body weight. Yet it remains unclear how individuals process meanings in public culture. We suggest that schema learning is a core mechanism by which public culture becomes personal culture. We propose that a burgeoning approach in computational text analysis – neural word embeddings – can be interpreted as a formal model for cultural learning. Embeddings allow us to empirically model schema learning and activation from natural language data. We illustrate our approach by extracting four lower-order schemas from news articles: the gender, moral, health, and class meanings of body weight. Using these lower-order schemas we quantify how words about body weight “fill in the blanks” about gender, morality, health, and class. Our findings reinforce ongoing concerns that machine-learning models (e.g., of natural language) can encode and reproduce harmful human biases.

Using word embeddings to investigate cultural biases

“…On the other hand, jazz and rap projected more strongly onto the ‘black pole’ of the race dimension while opera and bluegrass projected more strongly onto the ‘white pole’. A similar approach showed that obesity‐related words carry strong ‘unhealthy’ and ‘female’ bias in New York Times articles (Arseniev‐Koehler & Foster, 2020).…”

“…Arseniev‐Koehler & Foster, 2020; Garg et al, 2018; Kozlowski et al, 2019; Rodman, 2020). For example, Arseniev‐Koehler and Foster (2020) found that boyish was masculine on each of 25 replications, and on every model, it was less masculine than tough_guy . Even small differences in bias can be reliably detected.…”

“…Generally, bootstrapping has shown bias estimates to be remarkably stable and precise (cf. Arseniev‐Koehler & Foster, 2020; Garg et al, 2018; Kozlowski et al, 2019; Rodman, 2020). For example, Arseniev‐Koehler and Foster (2020) found that boyish was masculine on each of 25 replications, and on every model, it was less masculine than tough_guy .…”

“…Word embeddings have an uncanny power to ‘absorb’ biases from culture even when the associations are alluded to or implied but not explicitly stated (Caliskan et al, 2017, p. 1). They can thus be used to detect even subtle biases such as the association of body weight terms ( obese and slender ) with women and muscularity terms ( lean ) with men (Arseniev‐Koehler & Foster, 2020).…”

A Systematic Evaluation of Text Mining Methods for Short Texts: Mapping Individuals’ Internal States from Online Posts

“…Yet, text mining methods are only as good as we are as researchersthus, rather than replacing our efforts, they should only augment our capabilities .,Substantive research questions might even call for a combination of multiple methods that tackle different tasks in text analysis (Bonikowski and Nelson 2022). It is important to note that using data generated by individuals in their everyday life creates a particularly large gap between the raw data and meaningful measures of scientifically relevant concepts (Lazer et al 2021) and that relying on complex machine learning models carries a risk of unwillingly reproducing existing biases and cultural perceptions (Arseniev-Koehler and Foster 2022;Waseem 2016;Whittaker et al 2018).…”