Historical representations of social groups across 200 years of word embeddings from Google Books

Abstract

Using word embeddings from 850 billion words in English-language Google Books, we provide an extensive analysis of historical change and stability in social group representations (stereotypes) across a long timeframe (from 1800 to 1999), for a large number of social group targets (Black, White, Asian, Irish, Hispanic, Native American, Man, Woman, Old, Young, Fat, Thin, Rich, Poor), and their emergent, bottom-up associations with 14,000 words and a subset of 600 traits. The results provide a nuanced picture of change and persistence in stereotypes across 200 y. Change was observed in the top-associated words and traits: Whether analyzing the top 10 or 50 associates, at least 50% of top associates changed across successive decades. Despite this changing content of top-associated words, the average valence (positivity/negativity) of these top stereotypes was generally persistent. Ultimately, through advances in the availability of historical word embeddings, this study offers a comprehensive characterization of both change and persistence in social group representations as revealed through books of the English-speaking world from 1800 to 1999.

Keywords: attitude change; natural language processing; stereotype change; word embeddings.

Fig. 1.

Valence timeseries of top 10 traits. Valence computed from the average valence rating (from ref. 19) for the top 10 traits in each decade (on a scale from −4 [very negative] to +4 [very positive]). For plots of valence timeseries for words, see SI Appendix.

Fig. 2.

Within-group correlations of trait representations across 20 decades of historical text. Red colors indicate lower correlations, and blue colors indicate higher correlations. Correlations are computed between the vectors of all trait-group cosine similarities for each pair of decades. The first column of the triangle indicates the correlations between representations in 1800 with all other decades (1810 to 1990); the last row of the triangle indicates the correlations in 1990 with all other decades (1800 to 1980); and the bottom corner indicates the furthest pair of decades (1800 to 1990) while increasing toward the diagonal indicates closer pairs of decades. For correlation plots of word representations, see SI Appendix.

Fig. 3.

Cross-group correlations of trait representations for two groups over 20 decades (1800 to 1990) of historical English-language text. Higher correlations indicate greater overlap across the two comparison groups listed above each correlation.