Scriptum – “bezpečí”

The two plots below depict the usage of the term “bezpečí” in sentences from the Czech samizdat and exile literature from The position of individual points is based on similarities of the target term (i.e. “bezpečí”) token vectors within the corresponding sentences (the 3D coordinates are obtained via t-SNE). The token vectors are based on the FERNET-C5 BERT model. For more details, check the presentation and the code.

On the upper plot, the point colors represents the source: red – samizdat literature; green – exile literature. On the lower plot, the colors are based on k-means clustering (K=5).