# When VLMs 'Fix' Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR

arXiv: https://arxiv.org/abs/2604.22774 · HTML: https://mlsanity.com/paper/2604.22774/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 15 of its 38 references that are in the map.
- First posted on arXiv 1 Apr 2026.

## Builds on (most cited first, 15 of 15)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [COMET: A Neural Framework for MT Evaluation](https://mlsanity.com/paper/2009.09025/) (2020)
- [Image-to-Markup Generation with Coarse-to-Fine Attention](https://mlsanity.com/paper/1609.04938/) (2016)
- [The Hidden Life of Tokens: Reducing Hallucination of Large Vision-Language Models via Visual Information Steering](https://mlsanity.com/paper/2502.03628/) (2025)
- [Image Over Text: Transforming Formula Recognition Evaluation with Character Detection Matching](https://mlsanity.com/paper/2409.03643/) (2024)
- [Large Language Model as an Assignment Evaluator: Insights, Feedback, and Challenges in a 1000+ Student Course](https://mlsanity.com/paper/2407.05216/) (2024)
- [Can Vision-Language Models Evaluate Handwritten Math?](https://mlsanity.com/paper/2501.07244/) (2025)
- [Robust ASR Error Correction with Conservative Data Filtering](https://mlsanity.com/paper/2407.13300/) (2024)
- [Leveraging What's Overfixed: Post-Correction via LLM Grammatical Error Overcorrection](https://mlsanity.com/paper/2509.20811/) (2025)

## Built on by (most cited first, 4 of 4)

- [BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models](https://mlsanity.com/paper/2608.03884/) (2026)
- [Large Scale AI Grading of Handwritten Physics Assessments: Score Agreement and Olympiad Team Selection Outcomes](https://mlsanity.com/paper/2608.20521/) (2026)
- [When VLMs Trust Context: Evaluating Scene Text Recognition under Misleading Context](https://mlsanity.com/paper/2609.34781/) (2026)
- [Equal CER, Unequal Damage: Severity-Weighted Character Error Rate for OCR](https://mlsanity.com/paper/2610.20461/) (2026)

## Abstract (arXiv)

Accurate transcription of handwritten mathematics is crucial for educational AI systems, yet current benchmarks fail to evaluate this capability properly. Most prior studies focus on single-line expressions and rely on lexical metrics such as BLEU, which fail to assess the semantic reasoning across multi-line student solutions. In this paper, we present the first systematic study of multi-line handwritten math Optical Character Recognition (OCR), revealing a critical failure mode of Vision-Language Models (VLMs): over-correction. Instead of faithfully transcribing a student's work, these models often "fix" errors, thereby hiding the very mistakes an educational assessment aims to detect. To address this, we propose PINK (Penalized INK-based score), a semantic evaluation metric that leverages a Large Language Model (LLM) for rubric-based grading and explicitly penalizes over-correction. Our comprehensive evaluation of 15 state-of-the-art VLMs on the FERMAT dataset reveals substantial ranking reversals compared to BLEU: models like GPT-4o are heavily penalized for aggressive over-correction, whereas Gemini 2.5 Flash emerges as the most faithful transcriber. Furthermore, human expert studies show that PINK aligns significantly better with human judgment (55.0% preference over BLEU's 39.5%), providing a more reliable evaluation framework for handwritten math OCR in educational settings.