---
title: "Timesteps of Mamba Align with Human Reading Times (2026)"
description: "The paper therefore presents Mamba as a new tool for studying incremental, memory-limited, noisy human language processing. 30 references."
authors: ["Yuji Yamamoto", "Shinnosuke Isono", "Yoshinobu Kawahara", "Sho Yokoi"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.1592/"
doi: "10.18653/v1/2026.findings-acl.1592"
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/yamamoto-2026-timesteps-mamba-align-human-reading/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yamamoto-2026-timesteps-mamba-align-human-reading/)

[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/53/) / Timesteps of Mamba Align with Human Reading Times

# Timesteps of Mamba Align with Human Reading Times

**Authors:** Yuji Yamamoto, [Shinnosuke Isono](https://md.research.elamur.ai/free/en/authors/shinnosuke-isono/), Yoshinobu Kawahara and [Sho Yokoi](https://md.research.elamur.ai/free/en/authors/sho-yokoi/) · **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.1592/) · [PDF](https://aclanthology.org/2026.findings-acl.1592.pdf) · [DOI 10.18653/v1/2026.findings-acl.1592](https://doi.org/10.18653/v1/2026.findings-acl.1592) · [OpenAlex](https://openalex.org/W7166874311)

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

## Summary
This study asks whether Mamba, a state-space language model, handles words on a time scale that resembles human reading. In each Mamba layer, the input decides the size of a discretization timestep ∆t. This value acts as a gate. A large step lets the current word overwrite the stored memory, and a small step keeps the memory intact. The authors treat ∆t as the model's implicit processing time for a word. They averaged it across dimensions for each layer of Mamba-130M and Mamba-2.8B, then entered it as a predictor in regression models of reading times from the Natural Stories and OneStop corpora. Per-word timesteps significantly predicted how long people spent on words. The best layer performed about as well as GPT-2 surprisal, and many layers stayed significant after controlling for surprisal and other standard predictors. A formal analysis adds two points. Layers differ in how long they retain information, and timestep size corresponds to uncertainty in state transitions. The paper therefore presents Mamba as a new tool for studying incremental, memory-limited, noisy human language processing.

## Key points
- Data: Natural Stories, collected with self-paced reading from 181 native English speakers over 10 stories (10,245 words). OneStop "ordinary reading" subset, collected with eye-tracking from 180 readers over 30 articles, in an advanced version (19,428 words) and an elementary version (15,737 words). The OneStop measures were first-pass time and regression-path duration.
- In Mamba-130M (24 layers), layer 17 gave the best predictions for Natural Stories. Unlike many other layers, it shows no ∆t peaks at sentence boundaries, which hints that information carried across sentences matters for explaining reading behavior.
- Eigenspectrum analysis of the transition matrices shows that each layer retains memory to a different degree. Together, the layers cover both short- and long-term information.
- Limitations: the authors do not explain how the model learns to raise ∆t at sentence boundaries or across long syntactic distances. The results are also limited to Mamba, which is the only architecture with an input-dependent SSM timestep. Code is available on OSF.

*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

- [The University of Osaka](https://md.research.elamur.ai/free/en/universities/the-university-of-osaka/) — University, Osaka, Japan · 17 works · 517 citations
- [Tohoku University](https://md.research.elamur.ai/free/en/universities/tohoku-university/) — University, Sendai, Japan · 23 works · 595 citations
- [RIKEN](https://md.research.elamur.ai/free/en/labs/riken/) — Research institute, Wako, Japan · 30 works · 607 citations

📚 Shelves (8): [🖥️ 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 · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,649 works · [💭 Developmental and Educational Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/developmental-and-educational-psychology/) · 523 works · [🧠 Neuroscience](https://md.research.elamur.ai/free/en/shelves/neuroscience/) · 186 works · [🧠 Cognitive Neuroscience](https://md.research.elamur.ai/free/en/shelves/neuroscience/cognitive-neuroscience/) · 185 works

🧭 Topics (1): [🧒 Psycholinguistics & language acquisition](https://md.research.elamur.ai/free/en/topics/psycholinguistics/) · 108 works

## Cite

Cite as: Yuji Yamamoto, Shinnosuke Isono, Yoshinobu Kawahara and Sho Yokoi. 2026. Timesteps of Mamba Align with Human Reading Times. 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{yamamoto2026timesteps,
  title = {{Timesteps of Mamba Align with Human Reading Times}},
  author = {Yuji Yamamoto, Shinnosuke Isono, Yoshinobu Kawahara and Sho Yokoi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1592},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1592}
}
```

- [← previous TimeSAF: Towards LLM-Guided Semantic Asynchronous Fusion for Time Series Forecasting](https://md.research.elamur.ai/free/en/papers/zhang-2026-timesaf-llm-guided-semantic-asynchronous/)
- [next → Timing In stand-up Comedy: Text, Audio, Laughter, Kinesics (TIC-TALK): Pipeline and Database for the Multimodal Study of Comedic Timing](https://md.research.elamur.ai/free/en/papers/zribi-2026-timing-stand-up-comedy-text/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/53/)

## References

30 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Yevgeni Berzak, Jonathan Malmaud, Omer Shubi, Yoav Meiri, Ella Lion, and Roger Levy. 2025. Onestop: A 360-participant English eye tracking dataset with different reading regimes. Scientific Data, 12(1995). [find](https://scholar.google.com/scholar?q=Yevgeni+Berzak%2C+Jonathan+Malmaud%2C+Omer+Shubi%2C+Yoav+Meiri%2C+Ella+Lion%2C+and+Roger+Levy%2E+2025%2E+Onestop%3A+A+360%2Dparticipant+English+eye+tracking+dataset+with+different+reading+regimes%2E+Scientific+Data%2C+12%281)
2. Aaron Blakeman, Aaron Grattafiori, Aarti Basant, Abhibha Gupta, Abhinav Khattar, Adi Renduchintala, Aditya Vavre, Akanksha Shukla, Akhiad Bercovich, Aleksander Ficek, and 1 others. 2025. Nvidia nemotron 3: Efficient and open intelligence. arXiv preprint arXiv:2512.20856. — [arXiv:2512.20856](https://arxiv.org/abs/2512.20856) [find](https://scholar.google.com/scholar?q=Aaron+Blakeman%2C+Aaron+Grattafiori%2C+Aarti+Basant%2C+Abhibha+Gupta%2C+Abhinav+Khattar%2C+Adi+Renduchintala%2C+Aditya+Vavre%2C+Akanksha+Shukla%2C+Akhiad+Bercovich%2C+Aleksander+Ficek%2C+and+1+others%2E+2025%2E+Nvidia+nemotr)
3. Guy E Blelloch. 1990. [Prefix sums and their applications](https://md.research.elamur.ai/free/en/papers/blelloch-1990-prefix-sums-their-applications/).
4. Vera Demberg and Frank Keller. 2008. Data from eye-tracking corpora as evidence for theories of syntactic processing complexity. Cognition, 109:192–210. [find](https://scholar.google.com/scholar?q=Vera+Demberg+and+Frank+Keller%2E+2008%2E+Data+from+eye%2Dtracking+corpora+as+evidence+for+theories+of+syntactic+processing+complexity%2E+Cognition%2C+109%3A192%E2%80%93210%2E)
5. Nir Endy, Idan Daniel Grosbard, Yuval Ran-Milo, Yonatan Slutzky, Itay Tshuva, and Raja Giryes. 2025. Mamba knockout for unraveling factual information flow. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 23457–23477, Vienna,… [find](https://scholar.google.com/scholar?q=Nir+Endy%2C+Idan+Daniel+Grosbard%2C+Yuval+Ran%2DMilo%2C+Yonatan+Slutzky%2C+Itay+Tshuva%2C+and+Raja+Giryes%2E+2025%2E+Mamba+knockout+for+unraveling+factual+information+flow%2E+In+Proceedings+of+the+63rd+Annual+Meeting+o)
6. Richard Futrell, Edward Gibson, and Roger Levy. 2020. Lossy-context surprisal: An information-theoretic model of memory effects in sentence processing. Cognitive Science, 44(3):e12814. [find](https://scholar.google.com/scholar?q=Richard+Futrell%2C+Edward+Gibson%2C+and+Roger+Levy%2E+2020%2E+Lossy%2Dcontext+surprisal%3A+An+information%2Dtheoretic+model+of+memory+effects+in+sentence+processing%2E+Cognitive+Science%2C+44%283%29%3Ae12814%2E)
7. Richard Futrell, Edward Gibson, Harry J. Tily, Idan Blank, Anastasia Vishnevetsky, Steven T. Piantadosi, and Evelina Fedorenko. 2021. The natural stories corpus: A reading-time corpus of english texts containing rare syntactic constructions. Language Resources and Evaluation, 55(1):63–77. [find](https://scholar.google.com/scholar?q=Richard+Futrell%2C+Edward+Gibson%2C+Harry+J%2E+Tily%2C+Idan+Blank%2C+Anastasia+Vishnevetsky%2C+Steven+T%2E+Piantadosi%2C+and+Evelina+Fedorenko%2E+2021%2E+The+natural+stories+corpus%3A+A+reading%2Dtime+corpus+of+english+texts)
8. Albert Gu and Tri Dao. 2024. Mamba: Linear-time sequence modeling with selective state spaces. In First Conference on Language Modeling. [find](https://scholar.google.com/scholar?q=Albert+Gu+and+Tri+Dao%2E+2024%2E+Mamba%3A+Linear%2Dtime+sequence+modeling+with+selective+state+spaces%2E+In+First+Conference+on+Language+Modeling%2E)
9. Laura Gwilliams, Jean-Remi King, Alec Marantz, and David Poeppel. 2022. Neural dynamics of phoneme sequences reveal position-invariant code for content and order. Nature communications, 13(1):6606. [https://sites.google.com/view/317cafe/](https://sites.google.com/view/317cafe/) [↗](https://sites.google.com/view/317cafe/)
10. Laura Gwilliams, Alec Marantz, David Poeppel, and Jean-Remi King. 2024. Hierarchical dynamic coding coordinates speech comprehension in the human brain. bioRxiv. [find](https://scholar.google.com/scholar?q=Laura+Gwilliams%2C+Alec+Marantz%2C+David+Poeppel%2C+and+Jean%2DRemi+King%2E+2024%2E+Hierarchical+dynamic+coding+coordinates+speech+comprehension+in+the+human+brain%2E+bioRxiv%2E)
11. Michael Hahn, Richard Futrell, Roger Levy, and Edward Gibson. 2022. A resource-rational model of human processing of recursive linguistic structure. PNAS, 119(43). [find](https://scholar.google.com/scholar?q=Michael+Hahn%2C+Richard+Futrell%2C+Roger+Levy%2C+and+Edward+Gibson%2E+2022%2E+A+resource%2Drational+model+of+human+processing+of+recursive+linguistic+structure%2E+PNAS%2C+119%2843%29%2E)
12. John Hale. 2001. [A probabilistic Earley parser as a psycholinguistic model](https://md.research.elamur.ai/free/en/papers/hale-2001-probabilistic-earley-parser-as-psycholinguistic/). In Second Meeting of the North American Chapter of the Association for Computational Linguistics.
13. John Hale. 2016. Information-theoretical complexity metrics. Language and Linguistics Compass, 10:397– 412. [find](https://scholar.google.com/scholar?q=John+Hale%2E+2016%2E+Information%2Dtheoretical+complexity+metrics%2E+Language+and+Linguistics+Compass%2C+10%3A397%E2%80%93+412%2E)
14. Marcel Adam Just, Patricia A. Carpenter, and Jacqueline D. Woolley. 1982. Paradigms and processes in reading comprehension. Journal of Experimental Psychology: General, 111(2):228–238. [find](https://scholar.google.com/scholar?q=Marcel+Adam+Just%2C+Patricia+A%2E+Carpenter%2C+and+Jacqueline+D%2E+Woolley%2E+1982%2E+Paradigms+and+processes+in+reading+comprehension%2E+Journal+of+Experimental+Psychology%3A+General%2C+111%282%29%3A228%E2%80%93238%2E)
15. Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, Jiaxi Hu, Fanqing Meng, Chengyin Liu, Xin Men, Songlin Yang, Zhiyuan Li, and 1 others. 2025. [Kimi linear: An expressive, efficient attention architecture](https://md.research.elamur.ai/free/en/papers/kimi-team-2025-kimi-linear-expressive-efficient/). arXiv preprint arXiv:2510.26692.
16. Tatsuki Kuribayashi, Yohei Oseki, Ana Brassard, and Kentaro Inui. 2022. Context limitations make neural language models more human-like. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10421–10436, Abu Dhabi, United Arab Emirates. Association for Comp… [find](https://scholar.google.com/scholar?q=Tatsuki+Kuribayashi%2C+Yohei+Oseki%2C+Ana+Brassard%2C+and+Kentaro+Inui%2E+2022%2E+Context+limitations+make+neural+language+models+more+human%2Dlike%2E+In+Proceedings+of+the+2022+Conference+on+Empirical+Methods+in+N)
17. Tatsuki Kuribayashi, Yohei Oseki, Takumi Ito, Ryo Yoshida, Masayuki Asahara, and Kentaro Inui. 2021. Lower perplexity is not always human-like. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language P… [find](https://scholar.google.com/scholar?q=Tatsuki+Kuribayashi%2C+Yohei+Oseki%2C+Takumi+Ito%2C+Ryo+Yoshida%2C+Masayuki+Asahara%2C+and+Kentaro+Inui%2E+2021%2E+Lower+perplexity+is+not+always+human%2Dlike%2E+In+Proceedings+of+the+59th+Annual+Meeting+of+the+Associa)
18. Tatsuki Kuribayashi, Yohei Oseki, Souhaib Ben Taieb, Kentaro Inui, and Timothy Baldwin. 2025. [Large language models are human-like internally](https://md.research.elamur.ai/free/en/papers/kuribayashi-2025-large-language-models-are-human/). Transactions of the Association for Computational Linguistics, 13:1743–1766.
19. Roger Levy. 2008. Expectation-based syntactic comprehension. Cognition, 106(3):1126–1177. [find](https://scholar.google.com/scholar?q=Roger+Levy%2E+2008%2E+Expectation%2Dbased+syntactic+comprehension%2E+Cognition%2C+106%283%29%3A1126%E2%80%931177%2E)
20. Wei Liu and Nai Ding. 2025. Information integration in large language models is gated by linguistic structural markers. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 6903–6915, Suzhou, China. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Wei+Liu+and+Nai+Ding%2E+2025%2E+Information+integration+in+large+language+models+is+gated+by+linguistic+structural+markers%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Methods+in+Natural+Language+P)
21. David Marr. 1982. Vision. WH Freeman, San Fransisco, CA. [find](https://scholar.google.com/scholar?q=David+Marr%2E+1982%2E+Vision%2E+WH+Freeman%2C+San+Fransisco%2C+CA%2E)
22. Amirkeivan Mohtashami and Martin Jaggi. 2023. Random-access infinite context length for transformers. In Advances in Neural Information Processing Systems, volume 36, pages 54567–54585. Curran Associates, Inc. [find](https://scholar.google.com/scholar?q=Amirkeivan+Mohtashami+and+Martin+Jaggi%2E+2023%2E+Random%2Daccess+infinite+context+length+for+transformers%2E+In+Advances+in+Neural+Information+Processing+Systems%2C+volume+36%2C+pages+54567%E2%80%9354585%2E+Curran+Associa)
23. Byung-Doh Oh and William Schuler. 2023. [Why does surprisal from larger transformer-based language models provide a poorer fit to human reading times?](https://md.research.elamur.ai/free/en/papers/oh-2023-why-does-surprisal-larger-transformer/) Transactions of the Association for Computational Linguistics, 11:336–350.
24. Preferred Networks, Kaizaburo Chubachi, Yasuhiro Fujita, Shinichi Hemmi, Yuta Hirokawa, Kentaro Imajo, Toshiki Kataoka, Goro Kobayashi, Kenichi Maehashi, Calvin Metzger, and 1 others. 2025. Plamo 2 technical report. arXiv preprint arXiv:2509.04897. — [arXiv:2509.04897](https://arxiv.org/abs/2509.04897) [find](https://scholar.google.com/scholar?q=Preferred+Networks%2C+Kaizaburo+Chubachi%2C+Yasuhiro+Fujita%2C+Shinichi+Hemmi%2C+Yuta+Hirokawa%2C+Kentaro+Imajo%2C+Toshiki+Kataoka%2C+Goro+Kobayashi%2C+Kenichi+Maehashi%2C+Calvin+Metzger%2C+and+1+others%2E+2025%2E+Plamo+2+te)
25. Soo Hyun Ryu and Richard Lewis. 2021. Accounting for agreement phenomena in sentence comprehension with transformer language models: Effects of similarity-based interference on surprisal and attention. In Proceedings of the Workshop on Cognitive Modeling and Computational Linguistics, pages 61– 71,… [find](https://scholar.google.com/scholar?q=Soo+Hyun+Ryu+and+Richard+Lewis%2E+2021%2E+Accounting+for+agreement+phenomena+in+sentence+comprehension+with+transformer+language+models%3A+Effects+of+similarity%2Dbased+interference+on+surprisal+and+attention)
26. Arnab Sen Sharma, David Atkinson, and David Bau. 2024. [Locating and editing factual associations in mamba](https://md.research.elamur.ai/free/en/papers/sharma-2024-locating-editing-factual-associations-mamba/). In First Conference on Language Modeling.
27. Ethan G. Wilcox, Tiago Pimentel, Clara Meister, Ryan Cotterell, and Roger P. Levy. 2023. [Testing the predictions of surprisal theory in 11 languages](https://md.research.elamur.ai/free/en/papers/wilcox-2023-testing-predictions-surprisal-theory-11/). Transactions of the Association for Computational Linguistics, 11:1451–1470.
28. Songlin Yang and Yu Zhang. 2024. Fla: A triton-based library for hardware-efficient implementations of linear attention mechanism. [find](https://scholar.google.com/scholar?q=Songlin+Yang+and+Yu+Zhang%2E+2024%2E+Fla%3A+A+triton%2Dbased+library+for+hardware%2Defficient+implementations+of+linear+attention+mechanism%2E)
29. Ryo Yoshida, Shinnosuke Isono, Kohei Kajikawa, Taiga Someya, Yushi Sugimoto, and Yohei Oseki. 2025. If attention serves as a cognitive model of human memory retrieval, what is the plausible memory representation? In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguist… [find](https://scholar.google.com/scholar?q=Ryo+Yoshida%2C+Shinnosuke+Isono%2C+Kohei+Kajikawa%2C+Taiga+Someya%2C+Yushi+Sugimoto%2C+and+Yohei+Oseki%2E+2025%2E+If+attention+serves+as+a+cognitive+model+of+human+memory+retrieval%2C+what+is+the+plausible+memory+rep)
30. Jingwei Zuo, Maksim Velikanov, Dhia Eddine Rhaiem, Ilyas Chahed, Younes Belkada, Guillaume Kunsch, and Hakim Hacid. 2024. Falcon mamba: The first competitive attention-free 7b language model. arXiv preprint arXiv:2410.05355. — [arXiv:2410.05355](https://arxiv.org/abs/2410.05355) [find](https://scholar.google.com/scholar?q=Jingwei+Zuo%2C+Maksim+Velikanov%2C+Dhia+Eddine+Rhaiem%2C+Ilyas+Chahed%2C+Younes+Belkada%2C+Guillaume+Kunsch%2C+and+Hakim+Hacid%2E+2024%2E+Falcon+mamba%3A+The+first+competitive+attention%2Dfree+7b+language+model%2E+arXiv+pr)

*Updated 2 November 2026*