Archivists Turn to LLMs to Decipher Handwriting at Scale

2026-05-13 · Source: IEEE Spectrum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Novice, medium

Summary

General-purpose AI models, specifically large language models (LLMs) like GPT-4 and Gemini, are significantly improving the ability to transcribe historical handwritten documents, a task that has historically challenged AI researchers. A study by Mark Humphries and colleagues at Wilfrid Laurier University, published in May 2025 in "Historical Methods," demonstrated that LLMs outperformed specialized handwriting recognition software like Transkribus in accuracy, speed, and cost on 18th and 19th-century English-language documents. LLM-based approaches achieved character error rates below 2% compared to Transkribus's 8%, while being 50 times faster and 1/50th the cost. This advancement is making previously inaccessible archival collections searchable, enabling new research questions for scholars and family historians, and is being adopted by institutions like the University of North Carolina at Chapel Hill and the Federal Reserve Bank of Philadelphia.

Key takeaway

For archivists and historians managing large collections of handwritten documents, the emergence of highly capable LLMs like GPT-4 and Gemini fundamentally changes the economics and feasibility of transcription. You should explore integrating these general-purpose AI models to rapidly digitize and make searchable previously inaccessible materials, significantly reducing costs and processing times compared to traditional specialized software or manual methods. This shift enables new research avenues and democratizes access for a broader audience.

Key insights

General-purpose LLMs now reliably transcribe historical handwriting, outperforming specialized software in speed, cost, and accuracy.

Principles

• General methods leveraging computation often outperform specialized ones.
• Vast training data enables LLMs to infer complex patterns like handwriting.

Method

Feed handwritten document images to LLMs (e.g., GPT-4, Gemini) for transcription. This method leverages the models' broad training to interpret diverse handwriting styles and even tabular structures.

In practice

• Use LLMs for bulk transcription of diverse historical documents.
• Apply LLMs to extract data from complex historical ledgers.
• Explore tools like Archive Pearl for democratized access to AI transcription.

Topics

• Large Language Models
• Handwriting Recognition
• Archival Digitization
• Historical Research
• Generative AI

Best for: NLP Engineer, AI Scientist, Research Scientist, Domain Expert, General Interest

Related on AIssential

• Why LLMs Rewrite History (And How Multi-Agent Systems Can Help Restore It) — Multi-agent AI systems can effectively counter LLM hallucinations, particularly in historical document processing.
• Technical advances in document understanding — Document processing has evolved from basic OCR to sophisticated multimodal models that preserve structure and fine detail.
• A Text Recognition Dataset from Sahidic Coptic Ancient Manuscripts — A new dataset, SCAM, highlights significant challenges for Handwritten Text Recognition in low-resource historical scripts like Sahidic Coptic.
• How Grab Built a Vision LLM to Scan Images — Custom-built, lightweight Vision LLMs can surpass larger general-purpose models for specialized document processing tasks.
• OCR vs LLM Extraction — The Evolution of Document AI — OCR digitizes text, while LLMs understand document meaning and extract contextual information.
• Analysing LLMs for spelling normalization of 18th century Portuguese — Specialized LLMs significantly outperform multilingual models for historical language normalization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.