SUTA-LM: Bridging Test-Time Adaptation and Language Model Rescoring for Robust ASR

Guan-Ting Lin; Hung-yi Lee; Wei-Ping Huang

REVIEW 4 major objections 5 minor 19 references

SUTA-LM: Bridging Test-Time Adaptation and Language Model Rescoring for Robust ASR

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Test-time adaptation and LM rescoring can be combined by letting each utterance pick its own adaptation step, cutting average WER to 19.9 across 18 ASR datasets.

desk verdict A solid, useful empirical paper on combining TTA with LM rescoring for ASR; the proxy assumption in the step-selection rule needs validation, but the core result is credible. read the letter →

arxiv 2506.11121 v1 pith:LDEE6M3G submitted 2025-06-10 cs.CL cs.AIcs.SDeess.AS

Wei-Ping Huang , Guan-Ting Lin , Hung-yi Lee This is my paper

classification cs.CLcs.AIcs.SDeess.AS

keywords test-timeadaptationautomaticspeechrecognitionlanguagemodelrescoringentropyminimizationauto-stepselectiondomainshiftrobustnessbeamsearchdecoding

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

SUTA-LM claims that test-time adaptation (TTA) and language model rescoring, two techniques usually applied independently, can be combined into one pipeline if the number of adaptation steps is chosen per utterance rather than fixed in advance. The paper shows that naive sequential combination is unreliable: the optimal adaptation depth varies sharply across domains, and too many steps can actually hurt the rescored output even when greedy decoding looks stable. Its solution is an auto-step selection mechanism that scores every intermediate adapted model acoustically and linguistically, then rescores the chosen step's output with an external language model. On 18 diverse datasets, this yields an average word error rate of 19.9, better than SUTA+Rescoring (21.1), SUTA (21.3), SGEM (21.4), and rescoring alone (22.4), while running about 7x faster than SGEM. The paper's broader point is that future TTA methods for ASR should treat adaptation depth as an input-dependent decision, not a hyperparameter.

What carries the argument

The load-bearing mechanism is the two-stage auto-step selection rule. Acoustic Score Thresholding computes, for each adaptation step t, the average log confidence $S_t = (1/L) Σ_{l=1}^L log c_t^l$ over frames and keeps only steps with $S_t ≥ τ$; Linguistic Score Selection then picks the retained step $t^* = arg max_{t∈T} p_lm(y_t)$, where $y_t$ is the greedy decoding of $θ_t(x)$ and $p_lm$ is the external 4-gram language model probability. Ties go to the smallest index, and if no step passes the acoustic threshold the final step is used. An early-stopping variant halts adaptation when the best linguistic score has not improved for P valid steps. The two stages filter different failure modes: the acoustic threshold blocks underconfident early steps, while the linguistic score blocks overconfident but wrong late steps.

What would settle it

Compute, over the 18 datasets, the rank correlation between $p_lm(y_t)$ for greedy decodings along the SUTA trajectory and the WER of the beam-search-rescored transcription from the same intermediate model; a low or negative correlation would show the selection rule is not picking steps for the reason claimed. A more direct falsifier is to compare SUTA-LM to a per-utterance oracle that picks the step with the lowest true rescored WER: a large average gap would indicate the proxy is too noisy to support the method's central claim.

Watch

Extended reading notes

Core claim

The central discovery is that TTA can interfere with LM rescoring: after adapting a wav2vec 2.0 model with SUTA's entropy-minimization objective, the same number of adaptation steps that helps one domain (e.g., 8 steps on Gaussian noise) hurts another (1 step on Spanish-accented speech, 2 steps on TED talks). Even when greedy decoding performance stays flat across adaptation steps, the WER after LM rescoring degrades noticeably, showing that acoustic confidence and final linguistic quality diverge. SUTA-LM exploits this by scanning the adaptation trajectory, keeping only steps whose average log confidence exceeds a threshold, and then selecting the step whose greedy transcription has the highest probability under the external language model. That intermediate model's logits are then passed to beam search with LM rescoring. The result is a per-utterance choice of adaptation depth that the paper argues approaches an oracle that knows the true WER at each step, while remaining computationally cheap via early stopping.

Load-bearing premise

The selection rule assumes that the probability the external language model assigns to the greedy transcription at an intermediate adaptation step is a trustworthy stand-in for how good the final beam-search-rescored transcription from that step will be; if that proxy is poor, SUTA-LM can pick a step that scores worse than a fixed-step baseline.

Editorial extensions

If this is right

Fixed-step TTA baselines are suboptimal: the paper's step-variation curves imply that any method using a constant number of adaptation steps is leaving accuracy on the table on at least some domains.
Sequential combination of TTA and LM rescoring is not automatically beneficial; it needs a controller that stops adaptation before the model's output becomes harmful to rescoring.
SUTA-LM's efficiency claims follow directly from the mechanism: early stopping cuts the average number of adaptation steps, making the full pipeline about 7x faster than SGEM and over 2x faster than SUTA+Rescoring while achieving the lowest WER.
The ablation results indicate that both selection components are needed: removing the acoustic threshold hurts on noisy domains, while replacing linguistic selection with random selection raises WER substantially on accented and clean domains.
The method generalizes beyond the wav2vec 2.0 backbone, with consistent gains on data2vec-base and HuBERT-large, suggesting the selection rule is not tied to one acoustic model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not explicitly test whether $p_lm(y_t)$ correlates with the WER of the final rescored output; a natural extension is to check that correlation across the 18 datasets, since the entire selection rule hinges on it.
The same two-stage selection idea could be lifted onto other TTA objectives beyond SUTA, because the selection only reads logits and an external LM score rather than the TTA loss itself.
A stronger linguistic scorer than a 4-gram LM (e.g., a neural LM or LLM rescorer) might make the linguistic selection component even more discriminative, especially on domains where several steps are acoustically plausible.
One could construct a testable oracle comparison: measure the gap between SUTA-LM and the per-utterance oracle that selects the step with the lowest true rescored WER; if that gap is large in high-noise domains, the acoustic threshold is filtering out useful steps.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The short version: this paper has a genuinely useful empirical finding — fixed-step TTA can interfere with LM rescoring, and the optimal number of adaptation steps varies a lot across domains. SUTA-LM, a simple extension of SUTA with an auto-step selection mechanism, beats fixed-step baselines by about 1.2 WER on average across 18 datasets, while being faster than SGEM. The code is released, the benchmark is broad, and the ablation shows each component (acoustic threshold, linguistic selection, early stopping) contributes. That is a decent contribution to the ASR robustness subfield.

The main thing I'd want from the authors before believing the mechanism as described: they select the step t* by maximizing plm(yt), where yt is the greedy decoding from θt(x), but the reported output is the beam-search-rescored transcript from θt*(x). They never show that plm(yt) is a reliable proxy for the rescored WER. The stress-test note is right that Fig. 2 shows greedy WER stays fairly flat while rescored WER degrades, so the ranking could easily be off. That said, the ablation redeems the method partially: replacing linguistic selection with random selection hurts a lot on Korean and Spanish, so the linguistic signal is doing something. What's missing is a direct check against the oracle step — e.g., a regret analysis or a scatter plot of plm(yt) vs. final rescored WER. That's an addressable weakness, not a fatal one.

Other soft spots in proportion: no variance or significance tests are reported despite 'averaged over 3 runs,' and many of the per-dataset gains are small (0.1–0.2 WER). The acoustic threshold τ is tuned on a representative subset, and Table IV shows about 1 WER swing across reasonable thresholds, so robustness to that choice is not fully established. Also, LI-TTA is cited but not compared experimentally, which is a gap given the claim of covering LM-integrated TTA methods.

Who is this for? People working on TTA for ASR, or on practical domain robustness. It's not a theoretical paper; it's an engineering-empirical contribution with a clean setup and honest discussion of the interference issue. I'd bring it to a reading group for the interaction between TTA and LM rescoring, which is often ignored.

Recommendation: send it to peer review, but with a request for variance reporting and an analysis that directly validates or qualifies the plm proxy (oracle regret or correlation). It's a conditional accept, not a reject.

Referee Report

4 major / 5 minor

Summary. The paper proposes SUTA-LM, an extension of the SUTA test-time adaptation (TTA) method that combines entropy-minimization adaptation with external 4-gram language model rescoring. The key novelty is an auto-step selection mechanism that, for each test utterance, chooses an intermediate adapted model along the SUTA trajectory using two criteria: an acoustic-score threshold (Eq. 3) that filters out low-confidence steps, and a linguistic score (Eq. 4) that selects the step whose greedy transcript has the highest probability under the external LM. An early-stopping rule reduces computation. The authors report experiments on 18 ASR datasets spanning synthetic noise, accented speech, and real-world environments, claiming an average WER of 19.9, outperforming SUTA+Rescoring (21.1), SUTA (21.3), SGEM (21.4), and Rescoring (22.4), while running about 7x faster than SGEM. They also report ablations on the individual components and on different source ASR models.

Significance. If the claims hold, SUTA-LM is a practically useful, low-cost recipe for combining TTA with LM rescoring, and the observation that fixed-step adaptation can hurt rescoring is a useful caution for the ASR community. The paper's strengths include a broad 18-dataset evaluation, a comparison against a strong TTA baseline (SGEM), source-code release, and ablations that isolate the contribution of each mechanism. However, the central selection rule relies on an unverified proxy assumption: the LM probability of a greedy decoding is used to rank adaptation steps, while the reported output is the beam-search-rescored transcript from the selected step. The current evidence does not establish that this proxy ranks steps correctly in terms of final WER, which is the load-bearing claim of the method. The significance is therefore conditional on additional analysis demonstrating the ranking quality and on more rigorous statistical reporting.

major comments (4)

[Section IV-C, Eq. (4); Section III-B; Fig. 2]
[Tables I and II]
[Section VI-D, Table IV]
[Section VI-B, Fig. 2]

minor comments (5)

[Table III]
[Sections VI-D and VI-E]
[Fig. 4]
[Section III-A]
[Section IV-B, Eq. (2)]

Circularity Check

0 steps flagged · score 2.0 of 10

No definitional circularity; the central empirical claims are self-contained, though the linguistic step selection and the final rescoring share the same external LM, creating a mild alignment rather than a circular reduction.

full rationale

SUTA-LM is an empirical systems paper. The proposed auto-step selection mechanism (Eq. 4) chooses the adaptation step whose greedy transcript maximizes the external 4-gram LM probability, and the same LM is then used in beam-search rescoring of the selected model's output. This is a design choice that aligns the selection criterion with the linguistic component of the final decoding, but the reported metric is WER against ground truth, which is not defined in terms of plm; no prediction is equivalent to its input by construction. The threshold tau is fixed at -0.05 across all datasets, and the main result (Table II) is evaluated on 18 datasets against baselines that use the same rescoring; the ablation (Table III) shows random step selection degrades performance, indicating the linguistic score carries signal beyond the LM's own preference. The self-citations (SUTA [6], DSUTA [9]) are baselines re-run in the paper and are not load-bearing justifications of the central claim. The weakest link is the unverified proxy assumption that plm(yt) tracks the WER of the rescored output; Fig. 2 even shows greedy WER is stable while rescored WER degrades, so the proxy may fail. This is a correctness risk, not circularity, and does not reduce the central derivation to its inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method rests on two assumptions not proven in the paper: that LM score of greedy output predicts rescored WER, and that SUTA's adaptation trajectory contains an intermediate model well suited to rescoring. The free parameters are hyperparameters, with only tau explicitly swept.

free parameters (4)

Acoustic score threshold tau = -0.05, chosen from a sweep over five values on a representative subset.
Used in Eq. (3) to filter unreliable adaptation steps; Section VI-D explores tau values and the default is selected from this sweep.
Early stopping patience P = 3, fixed without sensitivity analysis.
Controls when the online adaptation stops if the linguistic score does not improve; stated in Section V-C with no robustness study.
Shallow fusion weights alpha, beta = 0.5 and 0, fixed in all experiments.
Define the LM rescoring objective in Section III-A; chosen for simplicity rather than tuned.
Maximum adaptation steps N = 20, used as the cap in all experiments.
The trajectory length over which auto-step selection operates; reported in Section V-B.

assumptions (4)

domain assumption The external 4-gram LM score plm(yt) of a greedy-decoded hypothesis is predictive of the WER of the final beam-search-rescored transcription.
Eq. (4) selects t by maximizing plm(yt); if this proxy fails, the selected step can be worse than a fixed-step baseline.
domain assumption SUTA's entropy-minimization update over a fixed number of steps produces a sequence of models in which intermediate steps are valid ASR models suitable for rescoring.
Section III defines theta0,...,thetaN and the method selects among them; the paper assumes no step collapses or becomes degenerate beyond the acoustic score filter.
standard math The beam search shallow fusion formula with alpha=0.5 and beta=0 is a correct and sufficient rescoring procedure.
Section III-A cites Deep Speech and uses pyctcdecode; the formula is standard and the specific constants are assumed.
domain assumption Excluding test utterances longer than 20 seconds does not bias the comparison.
Section V-C excludes less than 1% of data following prior work; the impact is small but not quantified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of SUTA-LM: Bridging Test-Time Adaptation and Language Model Rescoring for Robust ASR." pith.science (2026). https://pith.science/paper/LDEE6M3G

@misc{pith2026250611121,
  author       = {Pith},
  title        = {Pith review of: SUTA-LM: Bridging Test-Time Adaptation and Language Model Rescoring for Robust ASR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LDEE6M3G}},
  note         = {Machine review of arXiv:2506.11121}
}

read the original abstract

Despite progress in end-to-end ASR, real-world domain mismatches still cause performance drops, which Test-Time Adaptation (TTA) aims to mitigate by adjusting models during inference. Recent work explores combining TTA with external language models, using techniques like beam search rescoring or generative error correction. In this work, we identify a previously overlooked challenge: TTA can interfere with language model rescoring, revealing the nontrivial nature of effectively combining the two methods. Based on this insight, we propose SUTA-LM, a simple yet effective extension of SUTA, an entropy-minimization-based TTA approach, with language model rescoring. SUTA-LM first applies a controlled adaptation process guided by an auto-step selection mechanism leveraging both acoustic and linguistic information, followed by language model rescoring to refine the outputs. Experiments on 18 diverse ASR datasets show that SUTA-LM achieves robust results across a wide range of domains.

Figures

Figures reproduced from arXiv: 2506.11121 by the authors.

**Figure 2.** WER(%) of SUTA and SUTA+Rescoring over different numbers of [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Comparison of the average number of adaptation steps selected by [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

19 extracted references · 15 canonical work pages

[1]

Tent: Fully test-time adaptation by entropy minimization,

D. Wang, E. Shelhamer, S. Liu, B. Olshausen, and T. Darrell, “Tent: Fully test-time adaptation by entropy minimization,” in International Conference on Learning Representations , 2020

work page 2020
[2]

Ef- ficient test-time model adaptation without forgetting,

S. Niu, J. Wu, Y . Zhang, Y . Chen, S. Zheng, P. Zhao, and M. Tan, “Ef- ficient test-time model adaptation without forgetting,” in International conference on machine learning . PMLR, 2022, pp. 16 888–16 905

2022
[3]

Towards stable test-time adaptation in dynamic wild world,

S. Niu, J. Wu, Y . Zhang, Z. Wen, Y . Chen, P. Zhao, and M. Tan, “Towards stable test-time adaptation in dynamic wild world,” in The Eleventh International Conference on Learning Representations , 2022

work page 2022
[4]

Continual test-time domain adaptation,

Q. Wang, O. Fink, L. Van Gool, and D. Dai, “Continual test-time domain adaptation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 7201–7211

2022
[5]

Test time adaptation via conjugate pseudo-labels,

S. Goyal, M. Sun, A. Raghunathan, and J. Z. Kolter, “Test time adaptation via conjugate pseudo-labels,” Advances in Neural Information Processing Systems, vol. 35, pp. 6204–6218, 2022

work page 2022
[6]

Listen, Adapt, Better WER: Source-free Single-utterance Test-time Adaptation for Automatic Speech Recognition,

G.-T. Lin, S.-W. Li, and H. yi Lee, “Listen, Adapt, Better WER: Source-free Single-utterance Test-time Adaptation for Automatic Speech Recognition,” in Proc. Interspeech 2022 , 2022, pp. 2198–2202

work page 2022
[7]

SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization,

C. Kim, J. Park, H. Shim, and E. Yang, “SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization,” in Proc. INTERSPEECH 2023, 2023, pp. 3367– 3371

work page 2023
[8]

Awmc: Online test-time adap- tation without mode collapse for continual adaptation,

J.-H. Lee, D.-H. Kim, and J.-H. Chang, “Awmc: Online test-time adap- tation without mode collapse for continual adaptation,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) . IEEE, 2023, pp. 1–8

work page 2023

Show all 19 references

[9]

Continual test-time adaptation for end-to-end speech recognition on noisy speech,

G.-T. Lin, W. P. Huang, and H.-y. Lee, “Continual test-time adaptation for end-to-end speech recognition on noisy speech,” in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, Eds. Miami, Florid...

2024
[10]

Li-tta: Language informed test-time adaptation for automatic speech recognition,

E. Yoon, H. S. Yoon, J. Harvill, M. Hasegawa-Johnson, and C. D. Yoo, “Li-tta: Language informed test-time adaptation for automatic speech recognition,” in Interspeech 2024 , 2024, pp. 3490–3494

2024
[11]

Deep speech: Scaling up end-to-end speech recognition,

A. Hannun, C. Case, J. Casper, B. Catanzaro, G. Diamos, E. Elsen, R. Prenger, S. Satheesh, S. Sengupta, A. Coates, and A. Y . Ng, “Deep speech: Scaling up end-to-end speech recognition,” 2014. [Online]. Available: https://arxiv.org/abs/1412.5567

2014 arXiv
[12]

Advancing test-time adaptation in wild acoustic test settings,

H. Liu, H. Huang, and Y . Wang, “Advancing test-time adaptation in wild acoustic test settings,” in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, Eds. Miami, Florida, USA: Association for Co...

2024
[13]

Generative speech recognition error correction with large language models and task-activating prompting,

C.-H. H. Yang, Y . Gu, Y .-C. Liu, S. Ghosh, I. Bulyko, and A. Stolcke, “Generative speech recognition error correction with large language models and task-activating prompting,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , 2023, pp. 1–8

2023
[14]

Listen again and choose the right answer: A new paradigm for automatic speech recognition with large language models,

Y . Hu, C. Chen, C. Qin, Q. Zhu, E. Chng, and R. Li, “Listen again and choose the right answer: A new paradigm for automatic speech recognition with large language models,” in Findings of the Association for Computational Linguistics: ACL 2024 , L.-W. Ku, A. Martins, and V . S...

2024
[15]

Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,

F. Hernandez, V . Nguyen, S. Ghannay, N. Tomashenko, and Y . Est `eve, “Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,” in Speech and Computer. Springer International Publishing, 2018, pp. 198–208

2018
[16]

L2-arctic: A non-native english speech corpus,

G. Zhao, S. Sonsaat, A. Silpachai, I. Lucic, E. Chukharev-Hudilainen, J. Levis, and R. Gutierrez-Osuna, “L2-arctic: A non-native english speech corpus,” in Interspeech 2018 , 2018, pp. 2783–2787

2018
[17]

A scalable noisy speech dataset and online subjective test framework,

C. K. Reddy, E. Beyrami, J. Pool, R. Cutler, S. Srinivasan, and J. Gehrke, “A scalable noisy speech dataset and online subjective test framework,” Proc. Interspeech 2019 , pp. 1816–1820, 2019

2019
[18]

Librispeech: An asr corpus based on public domain audio books,

V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Librispeech: An asr corpus based on public domain audio books,” in 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2015, pp. 5206–5210

2015
[19]

The third ‘chime’speech separation and recognition challenge: Analysis and out- comes,

J. Barker, R. Marxer, E. Vincent, and S. Watanabe, “The third ‘chime’speech separation and recognition challenge: Analysis and out- comes,” Computer Speech & Language , vol. 46, pp. 605–626, 2017

2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Tent: Fully test-time adaptation by entropy minimization,

D. Wang, E. Shelhamer, S. Liu, B. Olshausen, and T. Darrell, “Tent: Fully test-time adaptation by entropy minimization,” in International Conference on Learning Representations , 2020

work page 2020

[2] [2]

Ef- ficient test-time model adaptation without forgetting,

S. Niu, J. Wu, Y . Zhang, Y . Chen, S. Zheng, P. Zhao, and M. Tan, “Ef- ficient test-time model adaptation without forgetting,” in International conference on machine learning . PMLR, 2022, pp. 16 888–16 905

2022

[3] [3]

Towards stable test-time adaptation in dynamic wild world,

S. Niu, J. Wu, Y . Zhang, Z. Wen, Y . Chen, P. Zhao, and M. Tan, “Towards stable test-time adaptation in dynamic wild world,” in The Eleventh International Conference on Learning Representations , 2022

work page 2022

[4] [4]

Continual test-time domain adaptation,

Q. Wang, O. Fink, L. Van Gool, and D. Dai, “Continual test-time domain adaptation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 7201–7211

2022

[5] [5]

Test time adaptation via conjugate pseudo-labels,

S. Goyal, M. Sun, A. Raghunathan, and J. Z. Kolter, “Test time adaptation via conjugate pseudo-labels,” Advances in Neural Information Processing Systems, vol. 35, pp. 6204–6218, 2022

work page 2022

[6] [6]

Listen, Adapt, Better WER: Source-free Single-utterance Test-time Adaptation for Automatic Speech Recognition,

G.-T. Lin, S.-W. Li, and H. yi Lee, “Listen, Adapt, Better WER: Source-free Single-utterance Test-time Adaptation for Automatic Speech Recognition,” in Proc. Interspeech 2022 , 2022, pp. 2198–2202

work page 2022

[7] [7]

SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization,

C. Kim, J. Park, H. Shim, and E. Yang, “SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization,” in Proc. INTERSPEECH 2023, 2023, pp. 3367– 3371

work page 2023

[8] [8]

Awmc: Online test-time adap- tation without mode collapse for continual adaptation,

J.-H. Lee, D.-H. Kim, and J.-H. Chang, “Awmc: Online test-time adap- tation without mode collapse for continual adaptation,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) . IEEE, 2023, pp. 1–8

work page 2023