This model distinguishes the names of authors who primarily wrote in Ancient Greek from the names of authors who wrote primarily in Latin.

The specific purpose for this model is to assist in processing bibliographic metadata about editions of Latin texts.

Most critical editions of ancient Greek texts bear a Latin version of the title of the original work and the author's name. For example, Hesiod's Theogony is Hesiodi Theogonia instead of Ἡσιόδου Θεογονία. Consequently, these works are often cataloged under the subject "Latin" by catalogers doing their best with languages they do not understand.

Consequently, metadata records tagged with the subject "Latin" from collections such as the HathiTrust Digital Library inevitably include Greek works.

Since the Digital Latin Library is interested only in records of Latin works, we need a good way of winnowing out the Greek editions. This model does a good job of that.

Here is the codecarbon output from training on Google Colab with an A100 runtime: