How Much Data Do You Need? About the Creation of a Ground Truth for Black Letter and the Effectiveness of Neural OCR (2020.lrec-1)
Copied to clipboard
| Challenge: | Recent advances in Optical Character Recognition and Handwritten Text Recognition have led to more accurate text recognition of historical documents. |
| Approach: | They propose to build a ground truth for a German-language newspaper published in black letter . they also evaluate the performance of different OCR engines and estimate how much data is needed to achieve high-quality OCR results. |
| Outcome: | The proposed model can recognise black letter text and performs well on data they have not seen during training. |
Similar Papers
OCR Improves Machine Translation for Low-Resource Languages (2022.findings-acl)
Copied to clipboard
| Challenge: | Despite many recent successes, Machine Translation still lacks support or fails to achieve good performance for most low-resource languages. |
| Approach: | They propose a benchmark to evaluate OCR systems on low-resource languages and low- resource scripts. |
| Outcome: | The proposed benchmark evaluates state-of-the-art OCR systems on low-resource languages and low-rural scripts. |
Low-resource Post Processing of Noisy OCR Output for Historical Corpus Digitisation (L18-1)
Copied to clipboard
| Challenge: | 7.6% of the words in the original OCR text contain an error; fully manual correction would take thousands of hours due to the size of the corpus. |
| Approach: | They propose a post-processing system to efficiently correct OCR errors in a 2.7 million word Faroese corpus. |
| Outcome: | The proposed method reduces the word error rate to 1.3% with around 65 hours of human annotator work. |
Efficient OCR for Building a Diverse Digital History (2024.acl-long)
Copied to clipboard
| Challenge: | Current optical character recognition (OCR) systems are poorly extensible to low-resource document collections, as learning a language-vision model requires extensive labeled sequences and compute. |
| Approach: | They propose to model optical character recognition as a character level image retrieval problem using a contrastively trained vision encoder. |
| Outcome: | The proposed model is more sample efficient and extensible than existing architectures, enabling accurate OCR in settings where existing solutions fail. |
Gold Standard Bangla OCR Dataset: An In-Depth Look at Data Preprocessing and Annotation Processes (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Existing datasets designed specifically for the Bengali language have been limited. |
| Approach: | They propose to use a large collection of labeled Bangla text image datasets to improve the performance of Bangla OCR. |
| Outcome: | The proposed system is the most extensive gold standard corpus for Bangla characters and words, comprising over 4 million human-annotated images. |
Digitizing Nepal’s Written Heritage: A Comprehensive HTR Pipeline for Old Nepali Manuscripts (2026.acl-long)
Copied to clipboard
| Challenge: | Using a line-level transcription approach, we explore encoder-decoder architectures and data-centric techniques to improve recognition accuracy for Old Nepali manuscripts. |
| Approach: | They propose a line-level transcription approach and explore encoder-decoder architectures and data-centric techniques to improve recognition accuracy. |
| Outcome: | The proposed model achieves a 4.9% error rate and is highly reliable. |
PEaCE: A Chemistry-Oriented Dataset for Optical Character Recognition on Scientific Documents (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing open-source OCR models focus on scientific texts or generic printed English . Nougat is unable to parse tables in PubMed articles . |
| Approach: | They propose to train OCR models for scientific or generic printed English . Nougat is a popular tool for parsing academic documents, but unable to parse PubMed tables . |
| Outcome: | The proposed models perform better when trained on real-world records than those trained on synthetic records. |
OCR Post Correction for Endangered Language Texts (2020.emnlp-main)
Copied to clipboard
| Challenge: | Currently, there is little to no data available to build natural language processing models for endangered languages. |
| Approach: | They propose a benchmark dataset of transcriptions for scanned books in three critically endangered languages and a method to improve OCR in these data-scarce settings. |
| Outcome: | The proposed method reduces the recognition error rate by 34% across the three endangered languages. |
When Good OCR Is Not Enough: Benchmarking OCR Robustness for Retrieval-Augmented Generation (2026.acl-industry)
Copied to clipboard
Lin Sun, null Wangdexian, Jingang Huang, Linglin Zhang, Change Jia, Zhengwei Cheng, Xiangzheng Zhang
| Challenge: | Existing OCR benchmarks rely on character-level metrics to measure downstream performance . high OCR accuracy does not translate into strong downstream performance, authors say . |
| Approach: | They propose an OCR benchmark for industrial RAG systems that measures character-level metrics . they find that high OCR accuracy does not translate into strong downstream RAG performance . |
| Outcome: | The proposed benchmark shows that high OCR accuracy does not translate into strong downstream performance . structural and semantic errors can cause substantial retrieval failures even when WER/CER remains low. |
Evaluating Transformers for OCR Post-Correction in Early Modern Dutch Theatre (2025.coling-main)
Copied to clipboard
| Challenge: | a new study examines the effectiveness of two types of transformer models for OCR post-correction in early modern Dutch plays. |
| Approach: | They propose to use large generative models and sequence-to-sequence models for OCR post-correction in early modern Dutch plays. |
| Outcome: | The proposed model outperforms generative models on the OCR post-correction task . the model outpersforms the model with the lowest error rate on the historical English dataset . |
Cleaning Dirty Books: Post-OCR Processing for Previously Scanned Texts (2021.findings-emnlp)
Copied to clipboard
| Challenge: | a large amount of work is required to clean digitized books for NLP analysis because of errors in the scanned text and duplicate volumes in the corpora. |
| Approach: | They propose methods to handle optical character recognition errors in scanned texts . they identify the canonical version for each of 17,136 repeatedly-scanned books . |
| Outcome: | The proposed method corrects over six times as many errors as it introduces, the authors show . the authors evaluate a collection of 19,347 texts from the Gutenberg dataset and 96,635 from the HathiTrust Library . |