Pith. sign in

REVIEW 3 major objections 5 minor 28 references

Reinforcement learning with verifiable rewards improves legal machine translation beyond supervised fine-tuning, the paper reports.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 13:10 UTC pith:QZFDARWG

load-bearing objection Solid empirical comparison with a useful dataset, but the RL-vs-SFT claim rests on automatic metrics that barely correlate with expert judgment and a possibly undertrained SFT baseline. the 3 major comments →

arxiv 2607.19181 v1 pith:QZFDARWG submitted 2026-07-21 cs.CL cs.AIcs.LG

Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

Aixiu An , Michael Jungo , Eloi Eynard , Mark Drenhaus , Andreas Fischer , Jean Hennebert , S\'ebastien Rumley This is my paper
classification cs.CL cs.AIcs.LG
keywords legal machine translationreinforcement learningGRPOsupervised fine-tuningreasoningChrFCOMETSwiss legal domain
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that for legal translation, reinforcement learning using simple verifiable rewards improves quality beyond supervised fine-tuning, while explicit reasoning steps in SFT actually hurt. Across three small models, RL with GRPO consistently outperforms both Simple SFT and Reasoning SFT on Swiss legal translation, narrowing the gap to frontier reasoning models. The sympathetic reader should care because this suggests a cost-effective path to high-quality domain translation with open, deployable models.

Core claim

The authors report that, on the SwissLegalTranslations test set, GRPO-trained models using a format reward for reasoning tags plus a ChrF-based translation-quality reward surpass both Simple SFT and Reasoning SFT on all four evaluation metrics and across all three model sizes. The best RL model, Gemma 3 12B, reaches COMET 83.08, within about two points of frontier models, while Reasoning SFT consistently degrades quality relative to Simple SFT.

What carries the argument

The central mechanism is a GRPO training recipe with two rule-based rewards: a format reward that enforces reasoning inside <think> tags before a final translation inside <translation> tags, and a continuous translation-quality reward equal to the ChrF character n-gram F-score against the reference. This connects the verifiable-reward approach from math reasoning to neural machine translation.

Load-bearing premise

The entire RL advantage rests on ChrF against a single test-set reference being a trustworthy proxy for legal translation quality; the paper's own human evaluation shows ChrF correlates only 0.05 with a legal expert, the weakest of the four metrics tested.

What would settle it

Collect expert ratings on a larger sample of RL versus SFT outputs large enough to detect a difference; if expert-preferred translations do not favor the RL models—or if replacing the ChrF reward with a more expert-correlated metric like COMET erases the gap—the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • RL-trained small models come within roughly two COMET points of frontier reasoning models on Swiss legal translation, using far fewer parameters.
  • Supervised fine-tuning on automatically generated reasoning steps not only fails to help; it systematically hurts translation quality.
  • RL gains are largest for the smallest model and shrink with scale, suggesting the main beneficiaries are compact, deployable models.
  • The released 40k reasoning-augmented legal translation dataset provides a resource for future research on reasoning in domain-specific NMT.
  • The GRPO + ChrF reward recipe offers a straightforward blueprint for applying RL to other terminology-dense translation domains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the paper's own human evaluation finds ChrF correlates only r=0.05 with a legal expert, the RL advantage may be an artifact of optimizing a weak proxy; an expert-based evaluation could rank methods differently.
  • The failure of Reasoning SFT suggests the reasoning traces themselves are not the active ingredient; the reward signal is. A testable extension is RL with only a translation-quality reward and no explicit thinking format.
  • The released reasoning-augmented dataset could be repurposed for other legal NLP tasks, such as summarization or legal error detection, beyond translation.
  • The same GRPO + ChrF recipe may transfer to other high-stakes, terminology-dense domains like medical or patent translation, where a reference-based lexical metric is available.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper investigates whether structured reasoning can improve legal machine translation for Swiss legal texts. The authors compare three strategies: prompting frontier reasoning models, supervised fine-tuning (Simple SFT and Reasoning SFT) of small language models (Qwen3.5 4B/9B, Gemma 3 12B), and GRPO-based reinforcement learning with rule-based rewards that include a ChrF term. They release a dataset of 40k automatically generated reasoning-augmented translation pairs and report that RL consistently outperforms both SFT variants across ChrF, COMET, METEOR, and MetricX, while noting diminishing returns with model scale. They also present a small human evaluation showing that COMET correlates best with a legal expert. The central claim is that RL with verifiable rewards surpasses SFT for legal translation and narrows the gap to frontier models.

Significance. If the central result holds, the paper offers a practical recipe for improving small open-weight models on a high-stakes domain and releases a useful dataset and codebase. The comparison across three model families, four automatic metrics, and multiple training paradigms is systematic and reproducible. The main significance is the demonstration that GRPO with a continuous lexical reward can improve LLM-based translation beyond SFT, although this is less surprising given that the reward is also used as an evaluation metric. The human-evaluation evidence is too thin to establish that the RL gains reflect expert-valued quality, so the significance depends on additional validation.

major comments (3)
  1. [§4.3, §3.1, Table 3] The RL reward is ChrF and the training data are filtered by ChrF > median, while Table 2 reports ChrF as a headline metric; this creates a reward-fitting risk. The paper's own Table 3 shows ChrF correlates with a legal expert at r=0.05 (p=0.54), and even COMET correlates only at r=0.24. Thus the automatic-metric improvements do not establish that RL outputs are better in human-valued legal translation quality. The human evaluation covers only 240 segments with a single expert and no per-system means or pairwise RL-vs-SFT comparison. I recommend a direct human preference test between RL and SFT outputs, with multiple experts, before the central claim can be accepted.
  2. [§4.2, Table 2, Abstract] The abstract claims RL 'surpasses the translation quality of supervised fine-tuning', but the comparison is limited to LoRA-based SFT of instruction-tuned LLMs; no conventional NMT baseline (e.g., NLLB, M2M-100, or a standard transformer) is included. Simple SFT often underperforms the base model (e.g., Qwen3.5 4B ChrF 50.03 vs 50.53; Gemma 12B ChrF 55.86 vs 56.73), suggesting the SFT configuration may not be representative. Without a standard NMT baseline, the practical significance of the RL gains relative to widely deployed NMT systems is unclear.
  3. [§3.1] The training dataset is constructed by taking the first 40,000 sentence pairs and retaining only those with ChrF > median (64.19), yielding 19,979 pairs. There is no control experiment with a random subset of the same size. Because the same metric is used as the RL reward, the filtering may bias the training distribution toward examples that are particularly ChrF-favorable, weakening the generalization claims of the released dataset and the RL recipe. A random-subset control would disentangle data-selection effects from method effects.
minor comments (5)
  1. [Table 3] MetricX has a negative correlation with human judgments (r=-0.13), but the text only notes that COMET correlates best; the negative correlation is not discussed. This is relevant to the choice of metrics for the central comparison.
  2. [Table 4 / Evaluation] The training set is heavily skewed (e.g., fr→it has 8,152 pairs while en→de, en→fr, en→it each have only 63–65), and the test set excludes Romansh after filtering. The paper does not report the language-pair distribution of the final test set, so the aggregate results in Table 2 may not represent balanced legal translation performance.
  3. [Table 2 / Figure 2] For Gemma 3 12B, the base model technically surpasses RL on it→en by a small margin in Figure 2, and several RL-vs-SFT gaps are small (e.g., COMET 83.08 vs 82.45). No pairwise significance tests are reported; the confidence intervals alone do not establish significance for all differences.
  4. [Table 1] The table lists 'OpenAI 4o' while the text mentions 'o4-mini', 'o4', and 'o3'; the naming is inconsistent and could confuse readers about which OpenAI model is which.
  5. [§6.1] There is a typo, 'self-reaffine', presumably 'self-refinement'. Also, the claim that reasoning steps from DeepSeek-R1 'resemble the workflow of a human translator' is anecdotal and not supported by analysis.

Circularity Check

1 steps flagged

Partial circularity: ChrF is both the RL reward and a headline evaluation metric, but independent metrics keep the central claim from fully reducing to the training objective.

specific steps
  1. fitted input called prediction [Section 4.3 (Experiment III: Reinforcement Learning); Table 2; Section 3.1 (Datasets)]
    "We chose the ChrF score to assess the quality of the translation and used it directly as a reward, since the values are in the range [0,1]."

    The same ChrF function is used to filter the training data (Section 3.1: retain samples with ChrF above the median 64.19), as the GRPO reward (Section 4.3), and as the first headline metric in Table 2. Thus the reported ChrF gains for RL are not an independent check of the method; they are the training objective itself evaluated on a held-out split. This is a partial reduction: the central claim that RL beats SFT is also supported by COMET, METEOR, and MetricX, which were not optimized, so the conclusion is not fully definitional. However, the paper's own human evaluation (Table 3) shows ChrF correlates only r=0.05 with expert judgment, indicating the ChrF-based component of the evidence is reward-fitting rather than a validated quality signal.

full rationale

The paper is an empirical comparison rather than a derivation from first principles, so most circularity categories do not apply. No uniqueness theorem is imported from the authors' prior work, and the one self-citation (Jungo & Fischer 2025) is related-work support, not load-bearing for the main conclusion. The concrete overlap between training and evaluation is ChrF: it is used to filter the training data (§3.1), as the RL reward (§4.3), and as an evaluation metric (Table 2). Because RL directly optimizes ChrF, reporting ChrF improvements is partly reporting the optimization target back; this is a genuine but partial circularity. The paper's central conclusion that RL surpasses SFT is not forced by this alone, because COMET, METEOR, and MetricX all move in the same direction and were not optimized. Still, the paper's own human evaluation shows ChrF has near-zero correlation with a legal expert (r=0.05, p=0.54), and even the best metric, COMET, correlates only r=0.24; the human sample (240 segments, one expert) is too thin to establish expert-valued legal quality. These are validity threats more than definitional circularity, so the score is moderate rather than extreme. The Limitations section also acknowledges reliance on cost-efficient automatic metrics, which further tempers the strength of the empirical claim.

Axiom & Free-Parameter Ledger

6 free parameters · 7 axioms · 0 invented entities

The paper's claims are empirical; there is no mathematical derivation. The central result depends on choices of evaluation metric, data filtering, and hyperparameters, each listed above.

free parameters (6)
  • ChrF filtering threshold (median = 64.19) = 64.19
    Retains roughly half of DeepSeek-R1 generated training data; chosen from the generated data's quality distribution, not from an independent criterion (§3.1).
  • SFT learning rate = 1e-4 (tuned in [2e-5, 1e-4])
    Manually tuned; RL used 2e-5 due to instability (§4.2, §4.3).
  • GRPO group size G = 4
    Chosen without ablation; affects reward baseline and variance (§4.3).
  • RL reward weights (format/order/ChrF) = 0.5 tag presence each, 0.5 order, ChrF continuous
    Hand-specified rule-based reward design (§4.3).
  • LoRA rank/alpha = r=16, alpha=32
    Standard values, not swept (§4.2).
  • Max sequence lengths = 512 (Simple SFT), 1024 (Reasoning SFT)
    Chosen to avoid repetitive outputs; truncates the final 5% of Reasoning SFT training examples, a confound between SFT variants (§4.2).
axioms (7)
  • domain assumption ChrF is a valid proxy for legal translation quality
    Used as RL reward and training filter; the paper's own human eval shows r=0.05, p=0.54 with an expert (§5, Table 3).
  • domain assumption The SwiLTra-Bench SwissLegalTranslations test split fairly represents Swiss legal translation quality
    All conclusions rest on this single test set (§3.1).
  • ad hoc to paper The first 40,000 training pairs are a representative sample of Swiss legal translation
    Subset chosen by position in the split, not by stratified sampling; heavily imbalanced across language pairs (§3.1, Table 4).
  • ad hoc to paper Filtering to ChrF > median removes bad translations without biasing task difficulty
    No unfiltered or random-subset control; may select easy sentences and inflate apparent training gains (§3.1).
  • domain assumption DeepSeek-R1 reasoning traces paired with ground-truth translations are a useful training signal
    Central to Reasoning SFT; results show it degrades quality, so the assumption fails empirically for this setup (§4.2, Table 2).
  • standard math GRPO as defined by Shao et al. 2024 works as assumed
    Algorithm adopted without modification or formal verification (§4.3).
  • domain assumption LoRA fine-tuning with rank 16 retains sufficient model capacity
    No full fine-tuning comparison; conclusions are about LoRA-fine-tuned models only (§4.2).

pith-pipeline@v1.3.0-alltime-deepseek · 12656 in / 12847 out tokens · 117371 ms · 2026-08-01T13:10:06.423417+00:00 · methodology

0 comments
read the original abstract

Neural machine translation (NMT) in the legal domain is a linguistically and conceptually demanding task, primarily due to the complexity of legal language and the high level of precision it requires. The recent emergence of reasoning-capable language models opens new possibilities for tackling such challenges. They add to a set of other previously proposed techniques to enhance the translation quality, which includes supervised fine-tuning and reinforcement learning. In this work, we perform a comparison between these various approaches. More particularly, we evaluate small language models such as Qwen3.5 4B, Qwen3.5 9B, and Gemma 3 12B enhanced with various re-training paradigms and compare their performances against frontier reasoning models. We focus on the Swiss legal system, which -- with its unique multilingual statutes -- offers a particularly challenging testbed for reasoning-augmented models. Our results show that the quality of small ``base'' models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning. The performance of enhanced small models is close to the one of state-of-the-art reasoning models yet remains inferior. We also note that re-training paradigms yield diminishing returns as model size increase. The code and models are publicly available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL.

Figures

Figures reproduced from arXiv: 2607.19181 by Aixiu An, Andreas Fischer, Eloi Eynard, Jean Hennebert, Mark Drenhaus, Michael Jungo, S\'ebastien Rumley.

Figure 1
Figure 1. Figure 1: Comparison of translation quality (mean of COMET score) across language pairs for different frontier models evaluated on the test dataset. translation is paramount to guide the model to produce better responses. The final translation is extracted from the response, which is then compared to the ground truth. We chose the ChrF score to assess the qual￾ity of the translation and used it directly as a reward,… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of translation quality (mean of COMET score) for Qwen3.5 9B (left) and Gemma 3 12B (right) across different language pairs and training procedures. tween small models and state-of-the-art LLMs. When comparing RL to SFT, we observe that RL consis￾tently outperforms both SFT variants across all models. Ex￾amining the two SFT approaches more closely, we find that Reasoning SFT degrades the quality … view at source ↗
Figure 3
Figure 3. Figure 3: Performance comparison of RL enhanced small models with frontier models, using the COMET score as a benchmark. of model for the moment), performance of small enhanced models could potentially be raised by fine-tuning parameters such as the value of the reward, the group size, etc. Hence, enhanced small models might be competitive for NMT. It might be tempting to try to fill the performance gap by us￾ing sl… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 20 linked inside Pith

  1. [2]

    Evaluating o1-like llms: Unlocking reasoning for translation through comprehensive analysis.arXiv preprint arXiv:2502.11544,

    Chen, A., Song, Y ., Zhu, W., Chen, K., Yang, M., Zhao, T., et al. Evaluating o1-like llms: Unlocking reasoning for translation through comprehensive analysis.arXiv preprint arXiv:2502.11544,

    Pith/arXiv arXiv

  2. [4]

    Dettmers, T., Lewis, M., Shleifer, S., and Zettlemoyer, L

    URL https://openreview.net/ forum?id=d3E3LWmTar. Dettmers, T., Lewis, M., Shleifer, S., and Zettlemoyer, L. 8- bit optimizers via block-wise quantization.arXiv preprint arXiv:2110.02861,

    Pith/arXiv arXiv

  3. [6]

    Mt-r1-zero: Advancing llm- based machine translation via r1-zero-like reinforcement learning.arXiv preprint arXiv:2504.10160,

    Feng, Z., Cao, S., Ren, J., Su, J., Chen, R., Zhang, Y ., Xu, Z., Hu, Y ., Wu, J., and Liu, Z. Mt-r1-zero: Advancing llm- based machine translation via r1-zero-like reinforcement learning.arXiv preprint arXiv:2504.10160,

    Pith/arXiv arXiv

  4. [9]

    R1-t1: Fully incentivizing translation capability in llms via reasoning learning.arXiv preprint arXiv:2502.19735,

    He, M., Liu, Y ., Tao, S., Luo, Y ., Zeng, H., Su, C., Zhang, L., Ma, H., Wei, D., Meng, W., et al. R1-t1: Fully incentivizing translation capability in llms via reasoning learning.arXiv preprint arXiv:2502.19735,

    Pith/arXiv arXiv

  5. [10]

    J., Afify, M., and Awadalla, 9 Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning H

    Hendy, A., Abdelrehim, M., Sharaf, A., Raunak, V ., Gabr, M., Matsushita, H., Kim, Y . J., Afify, M., and Awadalla, 9 Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning H. H. How good are gpt models at machine trans- lation? a comprehensive evaluation.arXiv preprint arXiv:2302.09210,

    Pith/arXiv arXiv

  6. [12]

    Is chatgpt a good translator? yes with gpt-4 as the engine.arXiv preprint arXiv:2301.08745,

    Jiao, W., Wang, W., Huang, J.-t., Wang, X., Shi, S., and Tu, Z. Is chatgpt a good translator? yes with gpt-4 as the engine.arXiv preprint arXiv:2301.08745,

    Pith/arXiv arXiv

  7. [13]

    MetricX-24: The Google submission to the WMT 2024 metrics shared task

    Juraska, J., Deutsch, D., Finkelstein, M., and Freitag, M. MetricX-24: The Google submission to the WMT 2024 metrics shared task. In Haddow, B., Kocmi, T., Koehn, P., and Monz, C. (eds.),Proceedings of the Ninth Conference on Machine Translation, pp. 492–504, Miami, Florida, USA, November

    2024

  8. [14]

    URL https://aclanthology.org/ 2024.wmt-1.35

    Association for Computational Linguistics. URL https://aclanthology.org/ 2024.wmt-1.35. Kalajdzievski, D. A rank stabilization scaling factor for fine-tuning with lora.arXiv preprint arXiv:2312.03732,

    Pith/arXiv arXiv 2024

  9. [15]

    and Federmann, C

    Kocmi, T. and Federmann, C. Gemba-mqm: Detecting translation quality error spans with gpt-4.arXiv preprint arXiv:2310.13988,

    Pith/arXiv arXiv

  10. [16]

    and Knowles, R

    Koehn, P. and Knowles, R. Six challenges for neural ma- chine translation.arXiv preprint arXiv:1706.03872,

    Pith/arXiv arXiv

  11. [17]

    ISBN 9798400702297

    Associa- tion for Computing Machinery. ISBN 9798400702297. doi: 10.1145/3600006.3613165. URL https://doi. org/10.1145/3600006.3613165. Liu, S., Lyu, C., Wu, M., Wang, L., Luo, W., Zhang, K., and Shang, Z. New trends for modern machine translation with large reasoning models.arXiv preprint arXiv:2503.10351,

    arXiv

  12. [18]

    and Hutter, F

    Loshchilov, I. and Hutter, F. Sgdr: Stochastic gradient descent with warm restarts. InInternational Conference on Learning Representations, 2017a. Loshchilov, I. and Hutter, F. Decoupled weight decay regu- larization.arXiv preprint arXiv:1711.05101, 2017b. Mart´ınez, E., Mollica, F., and Gibson, E. Even lawyers do not like legalese.Proceedings of the nati...

    Pith/arXiv arXiv

  13. [23]

    Deep reasoning translation via reinforcement learning.arXiv preprint arXiv:2504.10187,

    10 Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning Wang, J., Meng, F., and Zhou, J. Deep reasoning translation via reinforcement learning.arXiv preprint arXiv:2504.10187,

    Pith/arXiv arXiv

  14. [24]

    Emergent abilities of large language models

    Wei, J., Tay, Y ., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Met- zler, D., et al. Emergent abilities of large language models. arXiv preprint arXiv:2206.07682,

    Pith/arXiv arXiv

  15. [26]

    ISBN 979- 8-89176-332-6

    As- sociation for Computational Linguistics. ISBN 979- 8-89176-332-6. doi: 10.18653/v1/2025.emnlp-main

    doi:10.18653/v1/2025.emnlp-main 2025

  16. [28]

    Med-rlvr: Emerging medical reasoning from a 3b base model via reinforcement learning.arXiv preprint arXiv:2502.19655,

    Zhang, S., Liu, Q., Qin, G., Naumann, T., and Poon, H. Med-rlvr: Emerging medical reasoning from a 3b base model via reinforcement learning.arXiv preprint arXiv:2502.19655,

    Pith/arXiv arXiv

  17. [1031]

    emnlp-main.1031/

    URL https://aclanthology.org/2025. emnlp-main.1031/. Xia, S., Li, X., Liu, Y ., Wu, T., and Liu, P. Evaluating mathe- matical reasoning beyond accuracy. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp. 27723–27730,

    2025

  18. [1964]

    A., Humbel, C., G ¨osken, C., Tanzi, L., L¨uthi, T., et al

    Niklaus, J., Merane, J., Nenadic, L., Ahmadi, S., Gao, Y ., Chevalley, C. A., Humbel, C., G ¨osken, C., Tanzi, L., L¨uthi, T., et al. Swiltra-bench: The swiss legal translation benchmark.arXiv preprint arXiv:2503.01372,

    Pith/arXiv arXiv

  19. [2015]

    C., and Lavie, A

    Rei, R., Stewart, C., Farinha, A. C., and Lavie, A. Comet: A neural framework for mt evaluation.arXiv preprint arXiv:2009.09025,

    Pith/arXiv arXiv 2009

  20. [2016]

    Llms as models for analogical reasoning.arXiv preprint arXiv:2406.13803,

    Musker, S., Duchnowski, A., Milli`ere, R., and Pavlick, E. Llms as models for analogical reasoning.arXiv preprint arXiv:2406.13803,

    arXiv

  21. [2019]

    Please translate again: Two simple experiments on whether human- like reasoning helps translation

    Wu, D., Aycock, S., and Monz, C. Please translate again: Two simple experiments on whether human- like reasoning helps translation. In Christodoulopou- los, C., Chakraborty, T., Rose, C., and Peng, V . (eds.),Proceedings of the 2025 Conference on Em- pirical Methods in Natural Language Processing, pp. 20424–20440, Suzhou, China, November

    2025

  22. [2020]

    Deepseekmath: Push- ing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300,

    Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y ., Wu, Y ., et al. Deepseekmath: Push- ing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300,

    Pith/arXiv arXiv

  23. [2021]

    Competitive programming with large reasoning models.arXiv preprint arXiv:2502.06807,

    El-Kishky, A., Wei, A., Saraiva, A., Minaiev, B., Selsam, D., Dohan, D., Song, F., Lightman, H., Clavera, I., Pachocki, J., et al. Competitive programming with large reasoning models.arXiv preprint arXiv:2502.06807,

    Pith/arXiv arXiv

  24. [2022]

    K., and Shen, W

    Hu, J., Liu, J. K., and Shen, W. Reinforce++: An efficient rlhf algorithm with robustness to both prompt and reward models.arXiv preprint arXiv:2501.03262,

    Pith/arXiv arXiv

  25. [2023]

    Translategemma technical report

    Finkelstein, M., Caswell, I., Domhan, T., Peter, J.-T., Juraska, J., Riley, P., Deutsch, D., Kovacs, G., Dilanni, C., Cherry, C., et al. Translategemma technical report. arXiv preprint arXiv:2601.09012,

    arXiv

  26. [2024]

    doi: 10.18653/v1/2024

    Association for Computational Linguistics. doi: 10.18653/v1/2024. eacl-srw.17. URL https://aclanthology.org/ 2024.eacl-srw.17/. Banerjee, S. and Lavie, A. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. InProceedings of the acl workshop on in- trinsic and extrinsic evaluation measures for machine translation a...

    doi:10.18653/v1/2024 2024

  27. [2025]

    Iterative translation refinement with large language models.arXiv preprint arXiv:2306.03856,

    Chen, P., Guo, Z., Haddow, B., and Heafield, K. Iterative translation refinement with large language models.arXiv preprint arXiv:2306.03856,

    Pith/arXiv arXiv

  28. [2026]

    Deepseek-r1: In- centivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948,

    Guo, D., Yang, D., Zhang, H., Song, J., Zhang, R., Xu, R., Zhu, Q., Ma, S., Wang, P., Bi, X., et al. Deepseek-r1: In- centivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948,

    Pith/arXiv arXiv