Published 2026 | Version 1
Poster Open

A Dataset for Metaphor Detection in Early Medieval Hebrew Poetry

Authors/Creators

  • 1. Technion − Israel Institute of Technology
  • 2. The Open University of Israel

Description

This poster presents a new dataset of early medieval Hebrew poetry with expert annotations of metaphors. The dataset consists of 180 labeled texts, 7,000 sentences, and 45,000 words, with 15% of the words being metaphors. The inter-annotator agreement is 0.618, considered substantial. The results show that AlephBERT, a pre-trained language model, outperforms AlephBERT Adapted, and BEREL, pre-trained on medieval texts, shows the best results with an f1 score of 48.3 on the test set.

Files

8228476.pdf

Files (2.7 MB)

Name Size Download all
md5:428bfff030c96134828b870f5bbe9378
2.7 MB Preview Download
md5:077ea3b4821b7ce04f88f2b7beb6f3bf
8.2 kB Preview Download

Additional details

Related works

Is described by
Other: https://sandbox.posters.science/discover/34154 (URL)

Dates

Submitted
2026-07-29
Submitted to Zenodo through Posters.science