Pith. sign in

REVIEW 3 major objections 4 minor 26 references

Combining the Best of Both Worlds: A Method for Hybrid NMT and LLM Translation

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read One small classifier picks NMT or LLM per sentence, beats both systems

desk verdict Practical routing paper with consistent average gains; the headline overstates the tables and the COMET label/eval overlap is a real, untested weakness, but it deserves a serious referee. read the letter →

arxiv 2505.13554 v1 pith:2NASEBRU submitted 2025-05-19 cs.CL cs.AI

Zhanglin Wu , Daimeng Wei , Xiaoyu Chen , Hengchao Shang , Jiaxin Guo , Zongyao Li , Yuanchang Luo , Jinlong Yang
show 2 more authors
Zhiqiang Rao Hao Yang
This is my paper · ORCID
classification cs.CLcs.AI
keywords machinetranslationLLMNMThybridsource-sentencedeciderroutingqualityestimationCOMET
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

The paper claims that a translation system does not need to score every output of its neural machine translation (NMT) model to decide when a large language model (LLM) should take over: a small classifier that looks only at the source sentence can route each sentence to either model. The authors compare their Joint Decision-making (JDM) decider with QE-threshold routing (QET), which calls the LLM whenever an automatic quality estimate of the NMT output is low, and with a perplexity threshold on the source text (PPLT). On Zh2En, En2Zh, De2En, and Ja2En test suites, JDM reports the best average COMET-DA and BLEURT scores among the integrated methods while using the LLM on roughly 22% to 30% of sentences. The practical payoff is cost: most traffic stays on a fast NMT model, and the expensive LLM is reserved for the sentences where it actually helps.

What carries the argument

JDM (Joint Decision-making) is the central object: a binary classifier trained on positive samples where the NMT output is bad and the LLM is clearly better, meaning the NMT's COMET-DA score is below a first threshold and the LLM-minus-NMT DA gap is above a second threshold, plus negative samples drawn from the rest of the data. At inference the decider consumes only source-sentence features, so it replaces a quality-estimation model over the NMT output with a small classifier over the input text. The same metric, wmt22-comet-da, supplies the training labels, the reported DA scores, and the oracle ceiling, which is what makes the paper's quality claims internally coherent.

What would settle it

Run a human preference study on the Literary, Tech, Subtitle, and Travel test sets: if human raters do not consistently prefer JDM outputs over the better single system, or if human preferences disagree with wmt22-comet-da on the sentences JDM routes to the LLM, the central claim of optimal translation quality at minimal LLM usage is not supported.

Watch

Extended reading notes

Core claim

The central claim is that a binary classifier trained offline on triples of source sentences with NMT and LLM quality scores can learn to predict, from the source sentence alone, when the LLM will beat the NMT model. At inference time the JDM decider runs before translation: if it says LLM, the sentence goes to the LLM, otherwise to NMT. The paper reports that this routing policy achieves the best average DA and BLEURT among all compared hybrid methods on four language-pair test suites with an average LLM usage between 22% and 30%, and that the call rate adapts by domain, rising on Literary and News where the LLM is strong and falling on Tech and Flores where the NMT model is stronger. The oracle upper bound, defined by choosing the better of the two outputs with wmt22-comet-da, shows that the decider leaves some headroom and is not selecting optimally in every case.

Load-bearing premise

The load-bearing premise is that wmt22-comet-da, the metric used to label the decider's training data, to score all reported results, and to define the oracle ceiling, agrees with what human judges would prefer; if COMET's rankings diverge from human preferences, the training labels, the reported quality gains, and the oracle are all biased in the same direction.

Editorial extensions

If this is right

  • Production MT services can capture most of the quality benefit of LLM translation while sending only about a quarter of sentences to the LLM, without running a per-sentence quality estimator.
  • A single fixed decider can serve mixed traffic because its LLM call rate rises in domains where the LLM is strong and falls where the NMT model is strong.
  • The generalization experiment suggests the decider can be reused with fine-tuned NMT and LLM models without retraining, as long as the two systems keep their complementary strengths.
  • The gap between JDM and the oracle upper bound implies that better routing or a confidence-aware fallback could improve hybrid quality further without increasing LLM usage.

Reading between the lines

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

  • A natural next test is replacing the COMET-based training labels with human preference judgments; if human rankings disagree with COMET on the routed sentences, JDM's reported gains may not transfer to how readers perceive quality.
  • The paper's own category-3 failures, sentences that need context to translate, are a hard limit of sentence-level routing; a document-level decider that looks beyond the current sentence would be a direct extension.
  • Because the decider's signal appears to be largely about sentence complexity and domain, one could train it on one language pair and test it zero-shot on another to see how much of the routing signal is language-independent.
  • The threshold-setting procedure is a fixed quantile rule; learning the T1 and T2 thresholds per domain or per language pair from small labeled sets would likely improve the trade-off between quality and LLM usage.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper addresses the problem of combining NMT and LLM translation to improve quality while limiting LLM use. It proposes two routing methods: PPLT, which uses source-sentence perplexity from a small LM to decide when to invoke the LLM, and JDM, a binary classifier trained on source-sentence features to predict when the LLM output is better than the NMT output. The JDM training labels are derived from wmt22-comet-da scores using the conditions in Equation (1). The paper compares these methods against QET (based on wmt22-cometkiwi-da quality estimation) and single-system baselines on four language pairs (Zh2En, En2Zh, Ja2En, De2En) and several test sets (News, Flores, self-constructed Literary and Tech, plus Subtitle and Travel). The main finding is that JDM achieves the best average DA and BLEURT among the compared integration methods, with LLM usage around 22-30%, and outperforms QET on average across language pairs. The paper also contributes manually constructed literary and technical test sets.

Significance. The contribution is practically relevant: the JDM router uses only the source sentence, so it avoids running a QE model on every NMT output and can skip NMT when the LLM is chosen, reducing inference cost relative to QET. The experimental coverage is broad for a paper of this type (four directions, multiple domains, two metrics), and the reported pattern that JDM wins on average while using comparable LLM calls is consistent across language pairs. The analysis in Appendix A, showing that the router preferentially sends informal expressions to the LLM, gives some qualitative insight. However, the core quantitative claim is currently supported only by automatic metrics, and the overlap between the training metric and the evaluation metric is a genuine concern; the paper's significance would be materially strengthened by a human evaluation and by error bars on the headline comparisons.

major comments (3)
  1. [§3.2, §4] The JDM decider is trained on positive/negative labels computed from wmt22-comet-da scores via Equation (1), and the same metric is used as the primary evaluation (DA) and as the oracle upper bound in Tables 2 and 3. If wmt22-comet-da's ranking of translations diverges from human preferences on Literary, Tech, Subtitle, or Travel sentences, then the training labels, the reported quality gains, and the oracle ceiling are all biased in the same direction. The paper reports no human evaluation. BLEURT is a partially independent proxy because it is not used in training, but it is also a learned metric and the paper does not report whether the JDM gains on BLEURT are statistically reliable. Please add a human evaluation on a sample of sentences, or at least a sensitivity analysis that varies the evaluation metric, so that the central quality claim does not rest on a single metric that also generated the training signal.
  2. [Abstract and §4 (Tables 2-3, 7-8)] The abstract's claim of 'optimal translation performance with minimal LLM usage' is too strong. In Table 3, JDM's DA is below NMT alone on En2Zh Tech (85.39 vs 86.30) and Flores (87.76 vs 87.88). In Tables 7 and 8, JDM's LLM usage on Ja2En and De2En is slightly higher than QET (22.42% vs 21.95% and 23.32% vs 22.06%). Thus JDM does not dominate all baselines on every domain or use the fewest LLM calls in all settings. Please revise the abstract and conclusion to say that JDM achieves the best average performance among the compared policies with LLM usage comparable to or slightly higher than the QET baseline, rather than claiming optimality and minimality in an unqualified sense.
  3. [§4, Tables 2, 3, 7, 8] The paper does not report confidence intervals or significance tests for the DA/BLEURT differences. For example, in Table 2 the Zh2En average DA difference between JDM (78.81) and NMT (77.29) is 1.52 points, and the Literary test set has only 500 sentences. Without a bootstrap or other significance test, it is unclear whether the observed differences are within noise. This matters for the central claim that JDM improves over both single systems and QET. Please add significance tests or confidence intervals for the main comparisons, especially for the average scores across domains.
minor comments (4)
  1. [Table 9] The LLM-SFT row appears to be missing the BLEURT score on the Tech test set; the entry reads '78.20 100.00%' with no BLEURT value, whereas the other columns in that row show DA, BLEURT, and LLMp.
  2. [References] The citation 'A Conneau. 2019. Unsupervised cross-lingual representation learning at scale' is incomplete; the XLM-R paper is by Conneau et al. and was published at ACL 2020. Please correct the reference.
  3. [§1 and §4] The phrase 'obtain best in results for MT' in the introduction is grammatically awkward and should be rephrased.
  4. [Appendix F] The threshold sensitivity discussion is qualitative; please clarify how the thresholds were validated across domains and whether the same thresholds are used for all test sets within a language pair, since Table 5 lists only one threshold per language pair.

Circularity Check

1 steps flagged · score 4.0 of 10

Moderate circularity: the JDM router is trained on wmt22-comet-da labels, and the same metric defines the headline DA scores and the oracle upper bound; BLEURT provides partial independent support.

  1. fitted input called prediction [Section 3.2 (Eq. 1) training labels; Section 4 (Tables 2-3) evaluation and oracle]
    "When selecting positive and negative samples for training according to Equation 1, two thresholds need to be set. Specifically, we use one million bilingual data, obtain the NMT translation and LLM translation corresponding to the source text, and calculate the QNMT and QLLM scores using wmt22-comet-da ... wmt22-comet-da is used for reporting DA score (%) ... oracle system that selects the best translation results based on wmt22-comet-da, representing the upper bound that the fusion method can achieve."

    Equation 1 defines positive and negative training samples for the JDM decider from wmt22-comet-da scores of NMT and LLM outputs. The same wmt22-comet-da metric is then the primary DA evaluation and defines the oracle upper bound. The decider is therefore trained to reproduce COMET's preferences and is subsequently judged by that same preference function; the reported quality gain is partly an artifact of optimizing the evaluation metric. This is a training-evaluation loop rather than an independent test of translation quality. BLEURT is not used in training and still improves, so the circularity is partial, not total.

full rationale

The paper's derivation chain is otherwise self-contained: the PPLT baseline uses source-side perplexity with no overlap with the evaluation metric; JDM's core innovation is a source-feature classifier that is tested on held-out multilingual test sets and on unseen model combinations (Appendix I); and no load-bearing argument rests on self-citation. The one genuine circularity concern is the metric overlap: the JDM training labels (Eq. 1, Section 3.2), the reported DA scores, and the oracle ceiling are all computed with wmt22-comet-da. If COMET's ranking diverges from human judgement on Literary, Tech, Subtitle, or Travel sentences, the training objective, the headline gains, and the oracle upper bound are all biased in the same direction, and the claim of 'optimal translation performance' is not independently validated. BLEURT is a partially independent learned metric and does improve, which lowers the severity. The authors also do not report human evaluation, and on the En2Zh Tech and Flores test sets JDM is slightly below NMT alone (Table 3), so 'optimal' is domain-dependent; this is a correctness and risk issue rather than circularity. Overall score 4: a real training-evaluation metric loop, but the central claim retains independent content through BLEURT and held-out source-feature generalization.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on four domain assumptions: COMET is a trustworthy quality proxy, the calibration samples match the test distribution, source features can predict model preference, and sentence-level evaluation is fair for context-dependent sentences. The free parameters are the thresholds used to define positive/negative training examples and to control LLM usage; they are fitted to data rather than derived from first principles. No new physical or theoretical entities are introduced.

free parameters (5)
  • QET threshold = Zh2En 70, En2Zh 72, De2En 67, Ja2En 73
    Chosen so that the proportion of LLM calls is about 25% on a 1M sentence development sample (Section 3.2, Appendix F).
  • PPLT threshold = Zh2En 5.6, En2Zh 5.5, De2En 5.7, Ja2En 5.8
    Chosen as the 250,000th largest perplexity value on 1M monolingual sentences, targeting roughly 25% LLM usage (Section 3.2, Appendix F).
  • JDM T1 threshold = Zh2En 73, En2Zh 76, De2En 79, Ja2En 64
    Set as the 100,000th lowest wmt22-comet-da score of NMT translations on 1M bilingual pairs (Section 3.2).
  • JDM T2 threshold = Zh2En 3.5, En2Zh 3.5, De2En 2.5, Ja2En 3.5
    Set as the 10,000th highest difference between LLM and NMT COMET scores among data with QNMT below T1 (Section 3.2).
  • Target LLM usage rate = 25%
    An author-selected operating point used to calibrate all thresholds (Section 3.2); the actual usage on test sets ranges from 0.3% to 80.4%.
assumptions (4)
  • domain assumption wmt22-comet-da is a valid and accurate measure of translation quality for the languages and domains tested.
    Used to create JDM training labels (Section 3.2), to report all DA scores, and to define the oracle (Section 4). The paper does not validate COMET against human judgments on the custom test sets.
  • domain assumption The 1M bilingual and 30M monolingual samples used for thresholds and training are representative of the test-time distribution across news, literary, technical, subtitle, and travel domains.
    Thresholds fitted on these samples transfer to the test sets; the large variation in LLM usage across domains (Section 4, Tables 2-3) shows the transfer is imperfect.
  • domain assumption Source-sentence features (learned representations or perplexity) contain enough signal to predict when the LLM will outperform the NMT model.
    The entire JDM/PPLT approach depends on this premise. Appendix A reports a small 50-case check per category, but no systematic analysis of the decider's errors.
  • domain assumption Sentence-level translation is a fair task even for sentences that require context (category 3 in Appendix B).
    The authors note LLM also fails on context-dependent sentences at sentence level, yet those sentences remain in the test sets and the decider routes only 5% of them to the LLM. The evaluation treats all sentences as context-free.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Combining the Best of Both Worlds: A Method for Hybrid NMT and LLM Translation." pith.science (2026). https://pith.science/paper/2NASEBRU

@misc{pith2026250513554,
  author       = {Pith},
  title        = {Pith review of: Combining the Best of Both Worlds: A Method for Hybrid NMT and LLM Translation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NASEBRU}},
  note         = {Machine review of arXiv:2505.13554}
}
read the original abstract

Large language model (LLM) shows promising performances in a variety of downstream tasks, such as machine translation (MT). However, using LLMs for translation suffers from high computational costs and significant latency. Based on our evaluation, in most cases, translations using LLMs are comparable to that generated by neural machine translation (NMT) systems. Only in particular scenarios, LLM and NMT models show respective advantages. As a result, integrating NMT and LLM for translation and using LLM only when necessary seems to be a sound solution. A scheduling policy that optimizes translation result while ensuring fast speed and as little LLM usage as possible is thereby required. We compare several scheduling policies and propose a novel and straightforward decider that leverages source sentence features. We conduct extensive experiments on multilingual test sets and the result shows that we can achieve optimal translation performance with minimal LLM usage, demonstrating effectiveness of our decider.

Figures

Figures reproduced from arXiv: 2505.13554 by the authors.

Figure 1
Figure 1. A comparison of translations done by an NMT [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Two approaches to integrate NMT model and [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. illustrates the translation prompt used for LLM. {source_language} and {target_language} denote the full names of the languages involved, for example, "Translate this from Chinese to En￾glish." {source_sentence} represents the content that actually needs to be translated. <|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n Translate this from {source_lang} to {target_lang}:<|eot_id|> <|start_header_id|>u… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 5 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2016. https://arxiv.org/abs/1409.0473 Neural machine translation by jointly learning to align and translate . ICLR

    arXiv 2016

  4. [4]

    A Conneau. 2019. Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116

    arXiv 2019

  5. [5]

    Marta R Costa-juss \`a , James Cross, Onur C elebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, et al. 2022. No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672

    arXiv 2022

  6. [6]

    Ant \'o nio Farinhas, Jos \'e de Souza, and Andre Martins. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.733 An empirical study of translation hypothesis ensembling with large language models . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 11956--11970, Singapore. Association for Computational Linguistics

    doi:10.18653/v1/2023.emnlp-main.733 2023

  7. [7]

    Marina Fomicheva, Shuo Sun, Lisa Yankovskaya, Frédéric Blain, Francisco Guzmán, Mark Fishel, Nikolaos Aletras, Vishrav Chaudhary, and Lucia Specia. 2020. https://arxiv.org/abs/2005.10608 Unsupervised quality estimation for neural machine translation . TACL

    work page Pith review arXiv 2020

  8. [8]

    Amr Hendy, Mohamed Abdelrehim, Amr Sharaf, Vikas Raunak, Mohamed Gabr, Hitokazu Matsushita, Young Jin Kim, Mohamed Afify, and Hany Hassan Awadalla. 2023. How good are gpt models at machine translation? a comprehensive evaluation. arXiv preprint arXiv:2302.09210

    arXiv 2023

Show all 26 references
  1. [9]

    Kelleher, and Andy Way

    Yasmin Moslem, Rejwanul Haque, John D. Kelleher, and Andy Way. 2023. https://arxiv.org/abs/2301.13294 Adaptive machine translation with large language models . EAMT

    2023 arXiv

  2. [10]

    Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, and Michael Auli. 2019. fairseq: A fast, extensible toolkit for sequence modeling. In Proceedings of NAACL-HLT 2019: Demonstrations

    2019

  3. [11]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and...

    2022 arXiv

  4. [12]

    Ricardo Rei, Jos \'e GC De Souza, Duarte Alves, Chrysoula Zerva, Ana C Farinha, Taisiya Glushkova, Alon Lavie, Luisa Coheur, and Andr \'e FT Martins. 2022 a . Comet-22: Unbabel-ist 2022 submission for the metrics shared task. In Proceedings of the Seventh Conference on Machine...

    2022

  5. [13]

    Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.213 COMET : A neural framework for MT evaluation . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2685--2702, ...

    2020 doi

  6. [15]

    Guerreiro, Chrysoula Zerva, Ana C

    Ricardo Rei, Marcos Treviso, Nuno M. Guerreiro, Chrysoula Zerva, Ana C. Farinha, Christine Maroti, José G. C. de Souza, Taisiya Glushkova, Duarte M. Alves, Alon Lavie, Luisa Coheur, and André F. T. Martins. 2022 c . https://arxiv.org/abs/2209.06243 Cometkiwi: Ist-unbabel 2022 ...

    2022 arXiv

  7. [16]

    Thibault Sellam, Dipanjan Das, and Ankur P Parikh. 2020. Bleurt: Learning robust metrics for text generation. In Proceedings of ACL

    2020

  8. [17]

    Jungha Son and Boyoung Kim. 2023. https://doi.org/10.3390/info14100574 Translation performance from the user’s perspective of large language models and neural machine translation systems . Information, 14(10)

    2023 doi

  9. [18]

    Ilya Sutskever, Oriol Vinyals, and Quoc V. Le. 2014. https://arxiv.org/abs/1409.3215 Sequence to sequence learning with neural networks . NIPS

    2014 arXiv

  10. [19]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. https://arxiv.org/abs/2302.13971 Llama:...

    2023 arXiv

  11. [20]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. https://arxiv.org/abs/1706.03762 Attention is all you need . NIPS

    2017 arXiv

  12. [21]

    Wong, and Lidia S

    Qiang Wang, Bei Li, Tong Xiao, Jingbo Zhu, Changliang Li, Derek F. Wong, and Lidia S. Chao. 2019. https://arxiv.org/abs/1906.01787 Learning deep transformer models for machine translation . ACL

    2019 arXiv

  13. [22]

    Xiaohui Wang, Ying Xiong, Xian Qian, Yang Wei, Lei Li, and Mingxuan Wang. 2021 a . Lightseq2: Accelerated training for transformer-based models on gpus. arXiv preprint arXiv:2110.05722

    2021 arXiv

  14. [23]

    Xiaohui Wang, Ying Xiong, Yang Wei, Mingxuan Wang, and Lei Li. 2021 b . L ight S eq: A high performance inference library for transformers. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech...

    2021

  15. [24]

    Daimeng Wei, Zhiqiang Rao, Zhanglin Wu, Shaojun Li, Yuanchang Luo, Yuhao Xie, Xiaoyu Chen, Hengchao Shang, Zongyao Li, Zhengzhe Yu, et al. 2022. Hw-tsc’s submissions to the wmt 2022 general machine translation shared task. In Proceedings of the Seventh Conference on Machine Tr...

    2022

  16. [25]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mari...

    2020

  17. [26]

    Jiali Zeng, Fandong Meng, Yongjing Yin, and Jie Zhou. 2024. https://arxiv.org/abs/2311.02851 Improving machine translation with large language models: A preliminary study with cooperative decoding . Findings in ACL

    2024 arXiv

  18. [27]

    Biao Zhang, Barry Haddow, and Alexandra Birch. 2023. https://arxiv.org/abs/2301.07069 Prompting large language model for machine translation: A case study . Preprint, arXiv:2301.07069

    2023 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.