---
title: "FreeChunker: A Cross-Granularity Chunking Framework (2026)"
description: "Retrieval-Augmented Generation (RAG) depends heavily on how documents are split into chunks. 22 references."
authors: ["Zhang Wenxuan", "Yuan-Hao Jiang", "Yang Cao", "Yonghe Wu"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.730/"
doi: "10.18653/v1/2026.findings-acl.730"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/wenxuan-2026-freechunker-cross-granularity-chunking-framework/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wenxuan-2026-freechunker-cross-granularity-chunking-framework/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/33/) / FreeChunker: A Cross-Granularity Chunking Framework

# FreeChunker: A Cross-Granularity Chunking Framework

**Authors:** Zhang Wenxuan, Yuan-Hao Jiang, [Yang Cao](https://md.research.elamur.ai/free/en/authors/yang-cao/) and Yonghe Wu · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.730/) · [PDF](https://aclanthology.org/2026.findings-acl.730.pdf) · [DOI 10.18653/v1/2026.findings-acl.730](https://doi.org/10.18653/v1/2026.findings-acl.730) · [OpenAlex](https://openalex.org/W7166886201)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Retrieval-Augmented Generation (RAG) depends heavily on how documents are split into chunks. Existing chunkers assign each document a single granularity, and semantic chunkers often spend heavy compute, sometimes LLM inference, on finding boundaries. FreeChunker drops boundary detection entirely. It treats sentences as indivisible units, embeds each one once, and then uses a modified transformer encoder to build embeddings for many overlapping contiguous sentence spans. A banded Chunk Pattern Mask, applied as an attention mask between a learnable chunk query and the sentence embeddings, defines which sentences each candidate covers. This produces spans of several sizes in a single forward pass instead of re-encoding concatenated text. At query time, retrieval selects freely among these multi-size candidates.

On LongBench V2, the authors tested three embedding models with only the granularity set (1, 2, 4). FreeChunker achieved the best average retrieval performance against fixed-size, embedding-similarity, perplexity-based and LLM-guided chunkers, while running almost as fast as plain sentence accumulation. The result shows that RAG systems can gain flexible granularity at little cost.

## Key points
- The framework has three stages:
  - a Sentenizer splits the text and embeds each sentence;
  - a Cross-Granularity Encoder uses a Chunk Pattern Mask P ∈ {0, −∞} to aggregate sentence embeddings into span embeddings;
  - dense retrieval matches the query against the deduplicated multi-granularity candidates.
- Baselines were:
  - Traditional Chunking (sentence accumulation up to 256 tokens);
  - LangChain SemanticChunker (percentile threshold, amount 50.0);
  - Meta-Chunking with PPL Chunking (threshold 0.5) and Margin Sampling, both using Qwen/Qwen2.5-1.5B-Instruct;
  - LumberChunker.
- FreeChunker consistently outperforms Traditional Chunking, which uses the same sentence splitting. The authors treat this comparison as an ablation that confirms the benefit of the cross-granularity encoding.
- FreeChunker's processing time is close to Traditional Chunking and up to 30× faster than complex semantic chunkers. LLM-based methods such as LumberChunker take hundreds of seconds per instance.
- Pre-trained models and code are released at github.com/mazehart/FreeChunker. Non-contiguous span compositions are left for future work, and the extra encoder VRAM is reported as manageable.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [East China Normal University](https://md.research.elamur.ai/free/en/universities/east-china-normal-university/) — University, Shanghai, China · 69 works · 2,490 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 3,486 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,730 works

## Cite

Cite as: Zhang Wenxuan, Yuan-Hao Jiang, Yang Cao and Yonghe Wu. 2026. FreeChunker: A Cross-Granularity Chunking Framework. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{wenxuan2026freechunker,
  title = {{FreeChunker: A Cross-Granularity Chunking Framework}},
  author = {Zhang Wenxuan, Yuan-Hao Jiang, Yang Cao and Yonghe Wu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.730},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.730}
}
```

- [← previous Free-Gaza at NakbaArchiveClassifier Shared Task: Towards Distinguishing the Destructive Effect of Nakba: NakbaImage Classification Using Artificial Intelligence Techniques](https://md.research.elamur.ai/free/en/papers/yassin-2026-free-gaza-nakbaarchiveclassifier-shared-task/)
- [next → Frequency modulates structural choice in Turkish suspended affixation: a latent-process account](https://md.research.elamur.ai/free/en/papers/turk-2026-frequency-modulates-structural-choice-turkish/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/33/)

## References

22 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Yushi Bai, Shangqing Tu, Jiajie Zhang, Hao Peng, Xiaozhi Wang, Xin Lv, Shulin Cao, Jiazheng Xu, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. 2024. [Longbench v2: Towards deeper understanding and reasoning on realistic long-context multitasks](https://md.research.elamur.ai/free/en/papers/bai-2024-longbench-v2-deeper-understanding/). arXiv preprint arXiv:2412.15204.
2. Henrik Brådland, Morten Goodwin, Per-Arne Andersen, Alexander S. Nossum, and Aditya Gupta. 2025. A new hope: Domain-agnostic automatic evaluation of text chunking. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’25), pages 1–… [find](https://scholar.google.com/scholar?q=Henrik+Br%C3%A5dland%2C+Morten+Goodwin%2C+Per%2DArne+Andersen%2C+Alexander+S%2E+Nossum%2C+and+Aditya+Gupta%2E+2025%2E+A+new+hope%3A+Domain%2Dagnostic+automatic+evaluation+of+text+chunking%2E+In+Proceedings+of+the+48th+Internati)
3. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind André V. Duarte, João Marques, Miguel Graça, Miguel Freire, Lei Li, and Arlindo L. Oliveira. 2024. Lumberchunker: Long-form narrative document segmentation. In Findings of the Association for Computatio… [find](https://scholar.google.com/scholar?q=Tom+B%2E+Brown%2C+Benjamin+Mann%2C+Nick+Ryder%2C+Melanie+Subbiah%2C+Jared+Kaplan%2C+Prafulla+Dhariwal%2C+Arvind+Andr%C3%A9+V%2E+Duarte%2C+Jo%C3%A3o+Marques%2C+Miguel+Gra%C3%A7a%2C+Miguel+Freire%2C+Lei+Li%2C+and+Arlindo+L%2E+Oliveira%2E+2024%2E+Lum)
4. Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason Phang, Horace He, Anish Thite, Noa Nabeshima, et al. 2020. The pile: An 800gb dataset of diverse text for language modeling. arXiv preprint arXiv:2101.00027. [find](https://scholar.google.com/scholar?q=Leo+Gao%2C+Stella+Biderman%2C+Sid+Black%2C+Laurence+Golding%2C+Travis+Hoppe%2C+Charles+Foster%2C+Jason+Phang%2C+Horace+He%2C+Anish+Thite%2C+Noa+Nabeshima%2C+et+al%2E+2020%2E+The+pile%3A+An+800gb+dataset+of+diverse+text+for+lan)
5. Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2025. [A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions](https://md.research.elamur.ai/free/en/papers/huang-2023-survey-hallucination-large-language-models/). ACM Transactions on Information Systems, 43(2):…
6. Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts. Neural computation, 3(1):79–87. [find](https://scholar.google.com/scholar?q=Robert+A+Jacobs%2C+Michael+I+Jordan%2C+Steven+J+Nowlan%2C+and+Geoffrey+E+Hinton%2E+1991%2E+Adaptive+mixtures+of+local+experts%2E+Neural+computation%2C+3%281%29%3A79%E2%80%9387%2E)
7. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. [Efficient memory management for large language model serving with pagedattention](https://md.research.elamur.ai/free/en/papers/kwon-2023-efficient-memory-management-large-language/). In Proceedings of the 29th Symposium on Operating Systems Principles (SOSP ’23).
8. LangChain. 2024. Semantic chunker. [https://python.langchain.com/docs/how_to/](https://python.langchain.com/docs/how_to/) semantic-chunker/. Accessed: 2025-06-06. [↗](https://python.langchain.com/docs/how_to/)
9. Michel Ledoux. 2001. The concentration of measure phenomenon. 89. American Mathematical Soc. [find](https://scholar.google.com/scholar?q=Michel+Ledoux%2E+2001%2E+The+concentration+of+measure+phenomenon%2E+89%2E+American+Mathematical+Soc%2E)
10. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. [Retrieval-augmented generation for knowledge-intensive nlp tasks](https://md.research.elamur.ai/free/en/papers/lewis-2020-retrieval-augmented-generation-knowledge-intensive/). Advances in neural information processing systems, 33:9459–9474.
11. Junyi Li, Xiaoxue Cheng, Xin Zhao, Jian-Yun Nie, and Ji-Rong Wen. 2023. [Halueval: A large-scale hallucination evaluation benchmark for large language models](https://md.research.elamur.ai/free/en/papers/li-2023-halueval-large-scale-hallucination-evaluation/). In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages 6…
12. Yuanjie Lyu, Zhiyu Li, Simin Niu, Feiyu Xiong, Bo Tang, Wenjin Wang, Hao Wu, Huanyong Liu, Tong Xu, and Enhong Chen. 2025. Crud-rag: A comprehensive chinese benchmark for retrieval-augmented generation of large language models. ACM Transactions on Information Systems, 43(2):1– 32. [find](https://scholar.google.com/scholar?q=Yuanjie+Lyu%2C+Zhiyu+Li%2C+Simin+Niu%2C+Feiyu+Xiong%2C+Bo+Tang%2C+Wenjin+Wang%2C+Hao+Wu%2C+Huanyong+Liu%2C+Tong+Xu%2C+and+Enhong+Chen%2E+2025%2E+Crud%2Drag%3A+A+comprehensive+chinese+benchmark+for+retrieval%2Daugmented+generatio)
13. Vitali D Milman and Gideon Schechtman. 1986. Asymptotic theory of finite dimensional normed spaces. Springer. [find](https://scholar.google.com/scholar?q=Vitali+D+Milman+and+Gideon+Schechtman%2E+1986%2E+Asymptotic+theory+of+finite+dimensional+normed+spaces%2E+Springer%2E)
14. Li. 2025. Moc: Mixtures of text chunking learners for retrieval-augmented generation system. CoRR, abs/2503.09600. [find](https://scholar.google.com/scholar?q=Li%2E+2025%2E+Moc%3A+Mixtures+of+text+chunking+learners+for+retrieval%2Daugmented+generation+system%2E+CoRR%2C+abs%2F2503%2E09600%2E)
15. Jihao Zhao, Zhiyuan Ji, Pengnian Qi, Simin Niu, Bo Tang, Feiyu Xiong, and Zhiyu Li. 2024. Meta-chunking: Learning efficient text segmentation via logical perception. CoRR, abs/2410.12788. — [arXiv:2410.12788](https://arxiv.org/abs/2410.12788) [find](https://scholar.google.com/scholar?q=Jihao+Zhao%2C+Zhiyuan+Ji%2C+Pengnian+Qi%2C+Simin+Niu%2C+Bo+Tang%2C+Feiyu+Xiong%2C+and+Zhiyu+Li%2E+2024%2E+Meta%2Dchunking%3A+Learning+efficient+text+segmentation+via+logical+perception%2E+CoRR%2C+abs%2F2410%2E12788%2E+%E2%80%94+%5BarXiv%3A241)
16. Zijie Zhong, Hanwen Liu, Xiaoya Cui, Xiaofan Zhang, and Zengchang Qin. 2025. [Mix-of-granularity: Optimize the chunking granularity for retrieval-augmented generation](https://md.research.elamur.ai/free/en/papers/zhong-2025-mix-granularity-optimize-chunking-granularity/). In Proceedings of the 31st International Conference on Computational Linguistics, COLING 2025, Abu Dhabi, UAE, January 19-24, 2025, p…
17. Renyi Qu, Ruixuan Tu, and Forrest Sheng Bao. 2025. Is semantic chunking worth the computational cost? In Findings of the Association for Computational Linguistics: NAACL 2025, Albuquerque, New Mexico, USA, April 29 - May 4, 2025, pages 2155–2177. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Renyi+Qu%2C+Ruixuan+Tu%2C+and+Forrest+Sheng+Bao%2E+2025%2E+Is+semantic+chunking+worth+the+computational+cost%3F+In+Findings+of+the+Association+for+Computational+Linguistics%3A+NAACL+2025%2C+Albuquerque%2C+New+Mexico%2C)
18. Dongyu Ru, Lin Qiu, Xiangkun Hu, Tianhang Zhang, Peng Shi, Shuaichen Chang, Cheng Jiayang, Cunxiang Wang, Shichao Sun, Huanyu Li, Zizhao Zhang, Binjie Wang, Jiarong Jiang, Tong He, Zhiguo Wang, Pengfei Liu, Yue Zhang, and Zheng Zhang. 2024. Ragchecker: A fine-grained framework for diagnosing retriev… [find](https://scholar.google.com/scholar?q=Dongyu+Ru%2C+Lin+Qiu%2C+Xiangkun+Hu%2C+Tianhang+Zhang%2C+Peng+Shi%2C+Shuaichen+Chang%2C+Cheng+Jiayang%2C+Cunxiang+Wang%2C+Shichao+Sun%2C+Huanyu+Li%2C+Zizhao+Zhang%2C+Binjie+Wang%2C+Jiarong+Jiang%2C+Tong+He%2C+Zhiguo+Wang%2C+Pengfe)
19. Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc V. Le, Geoffrey E. Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2… [find](https://scholar.google.com/scholar?q=Noam+Shazeer%2C+Azalia+Mirhoseini%2C+Krzysztof+Maziarz%2C+Andy+Davis%2C+Quoc+V%2E+Le%2C+Geoffrey+E%2E+Hinton%2C+and+Jeff+Dean%2E+2017%2E+Outrageously+large+neural+networks%3A+The+sparsely%2Dgated+mixture%2Dof%2Dexperts+layer%2E+In)
20. Qwen Team. 2025. Qwen3 technical report. [find](https://scholar.google.com/scholar?q=Qwen+Team%2E+2025%2E+Qwen3+technical+report%2E)
21. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng He, Junyang Lin, Kai D… [find](https://scholar.google.com/scholar?q=An+Yang%2C+Baosong+Yang%2C+Binyuan+Hui%2C+Bo+Zheng%2C+Bowen+Yu%2C+Chang+Zhou%2C+Chengpeng+Li%2C+Chengyuan+Li%2C+Dayiheng+Liu%2C+Fei+Huang%2C+Guanting+Dong%2C+Haoran+Wei%2C+Huan+Lin%2C+Jialong+Tang%2C+Jialin+Wang%2C+Jian+Yang%2C+Jian)
22. Jihao Zhao, Zhiyuan Ji, Jason Zhaoxin Fan, Hanyu Wang, Simin Niu, Bo Tang, Feiyu Xiong, and Zhiyu [find](https://scholar.google.com/scholar?q=Jihao+Zhao%2C+Zhiyuan+Ji%2C+Jason+Zhaoxin+Fan%2C+Hanyu+Wang%2C+Simin+Niu%2C+Bo+Tang%2C+Feiyu+Xiong%2C+and+Zhiyu)

*Updated 2 November 2026*