Modeling the "Dalet" Clitic in Historical Hebrew Texts: A New Prefix-Segmented BERT Model and Stylistic Analysis

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computational Linguistics · Depth: Expert, quick

Summary

The new BERT model for historical Hebrew addresses the challenge of distinguishing the Aramaic proclitic "dalet"'s two orthographically identical grammatical functions: subordinating conjunction and possessive preposition. This model segments all prefixes, encoding them as independent tokens to allow direct evaluation of proclitics. Utilizing a probe-based unsupervised method with masked language modeling predictions, the model determines the "dalet" clitic's grammatical role. It achieved an average F1 score of over 0.89 on a manually annotated dataset derived from historical Hebrew literature. Applied to a corpus exceeding 300 million words, the method facilitated large-scale stylistic analyses, uncovering geographic and diachronic trends and identifying distinct stylistic clusters related to the choice between the Aramaic "dalet" and Hebrew alternatives. The prefix-segmented model and annotated dataset are released for unrestricted use.

Key takeaway

For NLP engineers or computational linguists working with historical texts or agglutinative languages, this prefix-segmented BERT model offers a robust approach to disambiguating orthographically identical grammatical functions. You should consider adapting this segmentation strategy for similar challenges in other languages, particularly when analyzing subtle stylistic variations or historical linguistic shifts. The released model and dataset provide a valuable starting point for such research.

Key insights

A prefix-segmented BERT model accurately disambiguates the "dalet" clitic's grammatical role in historical Hebrew, enabling large-scale stylistic analysis.

Principles

• Prefix segmentation improves BERT's clitic evaluation.
• Unsupervised probing can determine grammatical roles.
• Orthographic identity complicates NLP tasks.

Method

The method involves segmenting prefixes in a BERT model, then using a probe-based unsupervised approach with masked language modeling predictions to determine the grammatical role of orthographically ambiguous proclitics.

In practice

• Apply prefix segmentation to other agglutinative languages.
• Use probe-based methods for unsupervised grammatical analysis.
• Analyze diachronic linguistic trends with NLP models.

Topics

• Historical Hebrew
• BERT Model
• Prefix Segmentation
• Clitic Analysis
• Stylistic Analysis
• Natural Language Processing
• Digital Humanities

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Modelling the Morphology of Verbal Paradigms: A Case Study in the Tokenization of Turkish and Hebrew — Tokenization strategy critically impacts Transformer models' ability to represent complex morphology, especially in non-concatenative languages.
• MiqraBERT: Regression-Based Sentence-BERT Finetuning for Biblical Hebrew Parallel Detection — Finetuning a BERT model with regression for semantic similarity effectively detects textual reuse in ancient languages beyond lexical overlap.
• AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages — AfriSUD provides the first large-scale, native-speaker verified dependency treebanks for nine diverse African languages, revealing a significant syntax gap in current NLP models.
• From Vajrayana Tara to Bengali Baul: A Computational Study of Lexical Transmission Across Buddhist, Shakta, and Vaishnava Traditions in Bengal — Computational linguistics can quantitatively corroborate historical claims of lexical transmission and syncretism across religious traditions.
• Lexical and Orthographic Variation in Portuguese Financial Tweets: Annotation, Analysis, and Implications for Embedding Models — Lexical and orthographic variations in financial tweets are systematic, influencing embedding model performance and requiring curated lexicons.
• How Far Can Classical NLP Go? From Bag-of-Words to Stacking on Spooky Author Identification — Classical NLP, with careful feature engineering and ensemble stacking, excels at stylistic text classification.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.