# Transcribe, Then Reason: Two-Pass Decomposition for Multimodal Review

arXiv: https://arxiv.org/abs/2609.18958 · HTML: https://mlsanity.com/paper/2609.18958/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 26 of its 36 references that are in the map.
- First posted on arXiv 15 Jul 2026.

## Builds on (most cited first, 26 of 26)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Least-to-Most Prompting Enables Complex Reasoning in Large Language Models](https://mlsanity.com/paper/2205.10625/) (2022)
- [FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation](https://mlsanity.com/paper/2305.14251/) (2023)
- [On Faithfulness and Factuality in Abstractive Summarization](https://mlsanity.com/paper/2005.00661/) (2020)
- [Decomposed Prompting: A Modular Approach for Solving Complex Tasks](https://mlsanity.com/paper/2210.02406/) (2022)
- [Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents](https://mlsanity.com/paper/2410.05243/) (2024)
- [RARR: Researching and Revising What Language Models Say, Using Language Models](https://mlsanity.com/paper/2210.08726/) (2022)
- [Measuring Attribution in Natural Language Generation Models](https://mlsanity.com/paper/2112.12870/) (2021)
- [Nougat: Neural Optical Understanding for Academic Documents](https://mlsanity.com/paper/2308.13418/) (2023)
- [UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning](https://mlsanity.com/paper/2509.02544/) (2025)
- [ColPali: Efficient Document Retrieval with Vision Language Models](https://mlsanity.com/paper/2407.01449/) (2024)
- [The Consciousness Prior](https://mlsanity.com/paper/1709.08568/) (2017)
- [Step-Audio-R1 Technical Report](https://mlsanity.com/paper/2511.15848/) (2025)
- [DocVLM: Make Your VLM an Efficient Reader](https://mlsanity.com/paper/2412.08746/) (2024)
- [OCR-Reasoning Benchmark: Unveiling the True Capabilities of MLLMs in Complex Text-Rich Image Reasoning](https://mlsanity.com/paper/2505.17163/) (2025)
- [Closing the Gap Between Text and Speech Understanding in LLMs](https://mlsanity.com/paper/2510.13632/) (2025)
- [X-Talk: On the Underestimated Potential of Modular Speech-to-Speech Dialogue System](https://mlsanity.com/paper/2512.18706/) (2025)
- [VISTA-Bench: Do Vision-Language Models Really Understand Visualized Text as Well as Pure Text?](https://mlsanity.com/paper/2602.04802/) (2026)
- [Do Vision-Language Models Truly Perform Vision Reasoning? A Rigorous Study of the Modality Gap](https://mlsanity.com/paper/2604.16256/) (2026)
- [The Cascade Equivalence Hypothesis: When Do Speech LLMs Behave Like ASRrightarrowLLM Pipelines?](https://mlsanity.com/paper/2602.17598/) (2026)
- [Discrete JEPA: Learning Discrete Token Representations without Reconstruction](https://mlsanity.com/paper/2506.14373/) (2025)
- [Fairness of Explanations in Artificial Intelligence (AI): A Unifying Framework, Axioms, and Future Direction toward Responsible AI](https://mlsanity.com/paper/2605.09852/) (2026)

## Abstract (arXiv)

The natural way to review a long recording or document with a multimodal model is to hand it the raw source and ask for a review in one call. We show that this quietly fails: the model satisfices, dropping roughly a third of the content and embellishing the rest. The failure is not perception--almost all of the dropped content reappears when the same model is simply asked to transcribe the source. The bottleneck is generation under load: a single pass cannot perceive, reason over, and write a long faithful review at the same time, because doing all three competes for one output. We rule out the obvious alternatives. It is not the modality: models read text and an image of the same text equally well. And it is not merely a matter of thinking harder: giving the single pass a far larger reasoning budget does not recover the lost content, because the model spends that budget planning a review rather than writing the source down. What works is to split the labor across two same-weights passes--first transcribe, then review the transcript--so each step gets a full output budget of its own. This transcribe-then-review decomposition improves both faithfulness and coverage across a 21-source suite. The benefit is not uniform: we observe that it helps most where the one-pass baseline is weakest and little where that baseline is already strong, a pattern that also tracks the source's length and modality. Decomposition comes with two failure modes--the review pass running out of room on very long sources, and confabulating from memory once the grounding source is removed.