Letter Lemmatization: One-to-one and Banded RNNs for Reversing Character-Set Simplification and Abbreviation in Medieval Text

2026-07-10 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

This research introduces "Letter Lemmatization," a method employing one-to-one and Banded Recurrent Neural Networks (RNNs) to address character-set fluidity, simplification, and abbreviation in medieval texts. The approach trains character-level one-to-one RNNs using self-supervision to reverse character mappings, recovering half the Character Error Rate (CER) even with just 20 text lines. These networks significantly improve Handwritten Text Recognition (HTR) post-correction, even while ignoring insertions/deletions. The same networks, when configured as Banded RNNs with character-level alignment from parallel corpora, successfully expand abbreviations in medieval charter transcriptions. The work also defines a heuristic for semantic character similarity and provides a rich Python library implementing all presented methods.

Key takeaway

For NLP engineers or digital humanists working with historical documents, this research offers a robust solution for normalizing diverse medieval text corpora. You can apply the proposed one-to-one and Banded RNNs to automatically correct character-set variations and expand abbreviations, significantly improving data quality for downstream analysis. Consider integrating the provided Python library to streamline your preprocessing workflows for challenging historical datasets.

Key insights

Flexible character-set transformation and abbreviation expansion in medieval texts are achievable using self-supervised and aligned RNNs.

Principles

• Self-supervision can reverse character mappings.
• Banded RNNs expand abbreviations effectively.
• Semantic character similarity aids mapping.

Method

Train character-level one-to-one RNNs with self-supervision for character mapping reversal, then use the same networks as Banded RNNs with character-level alignment for abbreviation expansion.

In practice

• Post-correct Handwritten Text Recognition (HTR).
• Expand medieval text abbreviations.
• Develop custom character-set mappings.

Topics

• Letter Lemmatization
• Recurrent Neural Networks
• Medieval Texts
• Character-Set Simplification
• Abbreviation Expansion
• Handwritten Text Recognition
• Natural Language Processing

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Letter Lemmatization: One-to-one and Banded RNNs for Reversing Character-Set Simplification and Abbreviation in Medieval Text — Character-level RNNs and similarity heuristics effectively manage character set variations and abbreviations in historical texts, even with limited data.
• Neural edit-tree lemmatization for spaCy — A new ML-based lemmatizer for spaCy achieves over 95% accuracy by learning rules, eliminating manual linguistic engineering.
• Twenty’s Plenty: Semantic Scaffolding and Span Architecture for 19-Label NER in Medieval Latin Charters — High-quality 19-label NER for medieval Latin charters is achievable with limited data via semantic scaffolding and custom architecture.
• RALS: Resources and Baselines for Romanian Automatic Lexical Simplification — RALS provides the first Romanian lexical simplification resources, with a hybrid rule-based system outperforming LLMs for this low-resource language.
• CHiPS: Character Histograms and Positional Signals for Lightweight Authorship Attribution in Romanian Texts — CHiPS offers a lightweight, tokenization-free authorship attribution method using character histograms and positional signals for Romanian texts.
• Analysing LLMs for spelling normalization of 18th century Portuguese — Specialized LLMs significantly outperform multilingual models for historical language normalization.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.