Automatic Named Entity Linking for Ancient Greek with a Domain-Specific Knowledge Base (model weights)

This record documents the BLINK model weights developed for the paper “Automatic Named Entity Linking for Ancient Greek with a Domain-Specific Knowledge Base”, which will be presented at CHR 2025 in Luxembourg.

Code and additional information can be found on github: https://github.com/MarijkeBeersmans/CHR_code_Greek_NEL.

There are four versions of the model, based on two criteria.:

1. Training Data Type

• Gold: Trained exclusively on manually annotated (gold standard) data.
• Gold_Silver: Trained on a combination of manually annotated (gold) and synthetic (silver) data.

2. Text Source

• Kurz: Trained on the Kurztexts of the REALencyclopädie.
• Voll: Trained on the Volltexts of the REALencyclopädie.

Each model directory includes:

• A biencoder
• A crossencoder