OCR Error Post-Correction with LLMs in Historical Documents: No Free Lunches

2025-02-03

AI Summary

This study addresses post-OCR error correction for historical documents, systematically evaluating the generalization and practicality of open-source large language models (LLaMA, Phi series) in bilingual English–Finnish settings. We propose a joint optimization framework tailored to historical texts, incorporating dynamic input segmentation, 4-bit quantization, and context-aware continuation strategies. Crucially, we identify— for the first time—the performance bottlenecks of LLMs in low-resource language OCR correction, specifically for Finnish. Experimental results show significant CER reduction on English, validating method efficacy; however, Finnish performance remains below practical thresholds due to scarce linguistic resources, training data distribution shift, and historical text characteristics—including archaic orthography and layout-induced noise. The work provides a reproducible technical pathway for multilingual historical document proofreading and delivers critical, causally grounded analysis of cross-lingual LLM limitations in heritage text processing.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Optical Character Recognition (OCR) systems often introduce errors when transcribing historical documents, leaving room for post-correction to improve text quality. This study evaluates the use of open-weight LLMs for OCR error correction in historical English and Finnish datasets. We explore various strategies, including parameter optimization, quantization, segment length effects, and text continuation methods. Our results demonstrate that while modern LLMs show promise in reducing character error rates (CER) in English, a practically useful performance for Finnish was not reached. Our findings highlight the potential and limitations of LLMs in scaling OCR post-correction for large historical corpora.

Problem

• Language Models
• OCR Error Correction
• Historical Documents

Innovation

• Large-scale Language Models
• OCR Error Correction
• Historical Document Processing