Fig 3. Projecting a large sentiment lexicon (HGI, N = 3623) with positive/negative labeled terms on a set of cultural axes along the lines of gender, ethnicity, age, physical appearance, socioeconomic status and political/sexual/religious orientation often shows asymmetrical associations of positive and negative terms with the poles of cultural axes in seven popular pre-trained word embedding models.

A positive correlation value denotes association of positive terms with Pole 2 of the cultural axes and conversely association of negative terms with Pole 1. A negative correlation denotes association of positive terms with Pole 1 of the cultural axes and conversely, association of negative terms with Pole 2. Asterisks adjacent to a model bar indicate a statistically significant result (p<0.01) after applying the Bonferroni adjustment for multiple comparisons.