A Dataset for Metaphor Detection in Early Medieval Hebrew Poetry

TL;DR

This study addresses metaphor detection in early medieval Hebrew poetry (Piyyut) by constructing two expert-annotated corpora totaling 309 poems and 73,179 words. It evaluates transformer-based detectors (BEREL and AlephBERT) with BIO sequence labeling, showing that BEREL with domain-adaptive pretraining and weighted cross-entropy achieves near 48.7–49.4 F1 across corpora, with inter-annotator agreement around 0.62 and 71% of validation words perfectly predicted. The contributions include the first Hebrew metaphor annotation resource for Piyyut, detailed intrinsic statistics, baseline and advanced model results, and error analysis, all released under CC-BY to support humanities research and Hebrew NLP development. The findings demonstrate the value of adapting models to historical genres for figurative language detection and pave the way for semi-automatic annotation workflows in philological studies.

Abstract

There is a large volume of late antique and medieval Hebrew texts. They represent a crucial linguistic and cultural bridge between Biblical and modern Hebrew. Poetry is prominent in these texts and one of its main haracteristics is the frequent use of metaphor. Distinguishing figurative and literal language use is a major task for scholars of the Humanities, especially in the fields of literature, linguistics, and hermeneutics. This paper presents a new, challenging dataset of late antique and medieval Hebrew poetry with expert annotations of metaphor, as well as some baseline results, which we hope will facilitate further research in this area.

Figures (2)

• Figure 1: Distribution of the metaphor ratio in the Pre-Classical Piyyut corpus.
• Figure 2: Distribution of the metaphor ratio in the Pinchas corpus.