File:T-SNE visualisation of word embeddings generated using 19th century literature.png

English: Word embedding algorithms derive a set of real-valued vectors representing the vocabulary of a text corpus in a new embedded space. This provides a useful means of measuring the underlying similarity between words.

This image consists of word embeddings generated from 19th century literature. Gender-encoded unigrams, such as ‘she’ and ‘he’, by female authors are depicted as large, pink circles while the corresponding male authored unigrams are depicted as large, grey circles. Gender-encoded embeddings occupy four different spaces within this embeddings projection annotated A-D.

A: Female- and male-authored plural nouns {fellows, women, men,..} surrounded by past-participles verbs. No family related nouns such as {daughters, sisters, brothers} by female authors despite presence of male-authored counterparts.

B: Singular gender-encoded nouns by both female and male authors nested within nouns referring to (typically male) occupations {priest, clerk, magistrate, farmer,..}. All male-authored pronouns but only one female authored pronoun, "himself".

C: Family related nouns (singular and plural) by only female authors, nested within a cluster of characters predominately from Jane Austen’s novels.

D: Female authored pronouns next to past-participles and past verbs. Provides interesting counterpoint to Argamon et al. [1] who found differences in how women and men use words particularly personal pronouns.

[1] Argamon, S., Koppel, M., Fine, J., Shimoni, A.R.: Gender, genre, and writing style in formal written texts. TEXT 23, 321–346 (2003)