Fig. 5.

Text plot of words that distinguished the language of the Converters from the language of 30,000 Reddit users. Word positions were determined after dimensionality reduction of the word embeddings and clustering the positions using k-means++. The encircled clusters concern concepts related to voices and sounds