Historical and contemporary patterns of emotional expression in written texts

Abstract

AbstractWritten texts reflect the emotional state of humans that create them. It is however not always obvious how to interpret the observed trends and patterns. Here we use statistical analysis to extract information about emotions, and focus on the valence-energy space to ask the questions: (1) Can we detect a temporal change in emotional characteristics of texts? (2) Are there measurable differences of these emotional characteristics among different groups of people? To determine trends in emotion through writing, a searchable online Corpus of Historical American English was used (400 million words from 1810–2000), as well as a collection of 180 contemporary posts grouped by gender, age, and occupation. Sentiment analysis tools were applied to measure levels of positivity/negativity and energy of writing. It was found that through written text, energetic words decreased in frequency and less energetic words increased, indicating a decrease of strong feelings and a rise of apathy in the texts over time. In the present day’s blog texts, three pairwise comparisons were performed: males vs females, older vs younger age-groups, and individuals with background in arts vs those with backgrounds in science. While no statistically significant difference in energy levels were detected, there was a clear separation in the valence of these groups, with females, younger people, and those with a background in the arts displaying the most negativity. Mathematical modeling was used to interpret the findings of the historical analysis. It was shown that a flattening of emotions, or a rise in apathy, may not necessarily be caused by a corresponding population trend, but could be a simple consequence of a bell-shape curve in emotion distribution and imitation dynamics governing the production of written texts.