Vision‑Language Models (VLMs) are increasingly replacing traditional OCR for document understanding, but this study shows they often rewrite imperfect text into more plausible forms, a flaw that clean‑text OCR benchmarks miss. The authors created FaithC4, a multilingual perturbation benchmark of 1,455 single‑page documents with scramble, random substitution, and visually similar substitution attacks, and evaluated 15 systems across general‑purpose VLMs, OCR‑specialized VLMs, and traditional OCR pipelines. Results reveal that general‑purpose VLMs suffer up to 6.9 WER points under perturbation, OCR‑specialized VLMs 0.1–3.4 points, and traditional OCR less than 0.8 points on English; probing Qwen3‑VL‑4B shows rewriting occurs only when a perturbed word’s final‑layer representation remains close to the original, with short words (4–6 characters) rewritten up to 10% of the time.
whyItMatters":"The findings highlight a critical limitation of VLMs in document transcription, underscoring the need for robust evaluation benchmarks that capture rewriting behavior beyond clean‑text accuracy."
By Gwang Gook Lee, Kenan Emir Ak, Jay Mohta, Yan Xu, Dimitrios Dimitriadis
Vision-language models (VLMs) have achieved strong performance on OCR-based benchmarks and increasingly focused on text-rich understanding, but their robustness under controlled visual degradation remains insufficiently understood. This gap is critical for OCR reasoning, where visual corruption can induce OCR errors and structural distortions, thereby introducing uncertainty into the reasoning task.
arXiv:2607. 03836v1 Announce Type: cross Abstract: Despite remarkable progress in machine translation, Vision Language Models (VLMs) struggle on historical manuscripts, a domain that stresses core Natural Language Processing (NLP) capabilities: low-resource transliteration, archaic vocabulary, and noisy input signals.
By Nguyen Kim Hai Bui, Md. Easin Arafat, Tam\'as G\'abor Orosz, Mufti Mahmud
arXiv:2607. 24077v1 Announce Type: cross Abstract: Optical Character Recognition (OCR) is a key component in the digitization of historical archives.
By Marina Gardella (CB), Camilo Mari{\~n}o (UDELAR, CB), Diego Belzarena (UDELAR, CB), Ignacio Ram{\'i}rez (UDELAR), Gregory Randall (UDELAR), Jean-Michel Morel (LU - Hong Kong)
The paper introduces a controlled benchmark for evaluating large language models (LLMs) on key‑value pair extraction from documents with varying levels of OCR noise. It tests 136 configurations across five instruction‑tuned open‑weight LLMs, three datasets, and four text‑quality conditions, using deterministic decoding to generate 17,688 document‑level inferences. The study finds that clean‑text performance does not reliably predict real‑world robustness, model rankings can reverse under noisy conditions, and few‑shot demonstrations do not always improve accuracy, highlighting reliability risks in OCR‑to‑LLM pipelines.
By Zahra Anvari
The paper introduces an entity‑aware evaluation framework for OCR systems applied to historical documents, focusing on how well names, dates, and identifiers are preserved. Using a benchmark of 175 Berrutti Archive documents and 5,209 manually annotated entity mentions, the authors compare sequence‑recognition OCR, OCR‑specialized vision‑language models, and general‑purpose vision‑language models. Their results show that OCR rankings vary with the evaluation criterion: Qwen 3.6 achieves the highest entity coverage, while some systems recover many entities when small character‑level errors are tolerated, indicating that aggregate transcription metrics alone do not capture meaningful information preservation.
By Camilo Mari\~no (UDELAR, CB), Diego Belzarena (UDELAR, CB), Marina Gardella (CB), Ignacio Ram\'irez (UDELAR), Gregory Randall (UDELAR), Jean-Michel Morel (LU - Hong Kong)
arXiv:2605. 16409v3 Announce Type: replace-cross Abstract: Optical character recognition (OCR) and multilingual scene-text understanding remain challenging for multimodal large language models (MLLMs), particularly in real-world images containing small or degraded text, cluttered layouts, occlusion, handwriting, and complex typography.
By Qinwu Xu, Yifan Jiang, Haoyu Ren
arXiv:2606. 06890v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence.
By Runyu Zhou, Qi Zhang, Qixun Wang, Yisen Wang
The paper investigates how general vision‑language models (VLMs) develop specialized optical character recognition (OCR) capabilities. By applying a causal intervention protocol, the authors identify sparse, stable OCR‑head sets in several VLMs and show that these heads largely overlap with textual retrieval/copy heads found in general VLMs. The study concludes that full‑sequence OCR functions as a dense multimodal copy‑and‑paste mechanism, and that when a VLM is fine‑tuned for OCR, it largely preserves the same head identities while redistributing their functional and causal strengths.
By Yuanxiang Huangfu, Hanmeng Zhong, Linqing Chen, Jeffrey Tiong Jee Hui
The paper investigates how vision‑language models (VLMs) perform optical character recognition (OCR) by identifying attention heads that are causally necessary for OCR across four models. These heads are shown to be general‑purpose, producing interpretable semantic features for any image token, such as recognizing the word "bike" or the concept "feathers". By collapsing the heads’ attention weights into a verbalization lens transformation, the authors reveal that image representations align with language from early layers and can even be used to edit non‑word concepts in images, demonstrating the broader utility of this subspace.
By Sheridan Feucht, Benno Krojer, Sarah Wang, Henry Abrahamsen, Byron C. Wallace, David Bau
The paper introduces the concept of Squint Text—text embedded in an image’s coarse structure that multimodal LLMs often miss but humans can read by squinting. It presents SquintBench, a dataset of 1,076 moderation images, and SquintCtrl, a synthetic suite for diagnosing the issue. The authors show that spatial selection is a key bottleneck and propose Squint OPSD, a training method that uses a downscaled teacher view and privileged transcription to supervise a student model, achieving a significant performance boost on SquintBench.
By Chen Peng, Shan Huang, Jianbo Huang, Linrui Qin, Qibo Sun, Mengwei Xu
ReWEIGH the Evidence is a training‑free decoding technique that calibrates token‑level ordinal visual evidence to reduce hallucinations in large vision‑language models. It aggregates vocabulary ranks across visual positions, compares candidates to a token‑specific reference derived from unlabeled images, and applies a bounded penalty only when evidence falls below this reference. Experiments on four 7B backbones show up to a 21.3% reduction in hallucinated object mentions while largely preserving or improving descriptive and general performance, with minimal added latency.
By Jihae Jeong, Junha Choi, Hwanjo Yu