Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Psychology-Driven Enhancement of Humour Translation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a three-step prompt decomposition into topic, angle, and punchline improves LLM humour translation quality by an average of 7.75% in humour.

desk verdict The prompt idea is sensible and the experimental breadth is real, but the evaluation never checks whether the output still conveys the source joke's meaning, so the central claim about improving translation quality is not supported. read the letter →

arxiv 2507.09259 v1 pith:7ERGC6IK submitted 2025-07-12 cs.CL

Yuchen Su , Yonghua Zhu , Yang Chen , Diana Benavides-Prado , Michael Witbrock This is my paper
classification cs.CL
keywords humourtranslationlargelanguagemodelsChain-of-ThoughttheoryGEMBAevaluationpromptengineeringlinguisticinterference
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

This paper tries to establish that a three-step Chain-of-Thought prompting strategy, the Humour Decomposition Mechanism (HDM), makes large language models translate jokes better across languages. The authors argue that translating a joke is not a direct word-for-word task; the model should first decompose the joke into topic, angle, and punchline, translate that analysis, and then rebuild a new joke in the target language. On English-to-Chinese translation of 500 short jokes, they report average gains of 7.75% in humour, 2.81% in fluency, and 6.13% in coherence over strong translation baselines, using a GPT-4-based evaluation metric. If true, the method offers a simple, prompt-level fix for a known weakness of LLMs: humour loss and linguistic interference in translated texts.

What carries the argument

The Humour Decomposition Mechanism (HDM) is a three-step Chain-of-Thought prompt paradigm: (1) humour decomposition, where the model is prompted to analyze a joke into topic, angle, and punchline following a standard monologue joke theory; (2) a translation module that renders that analysis into the target language; and (3) humour composition, where the model generates a new joke from the translated analysis. This decomposition is meant to mimic a human translator's thought process, preserving the joke's semantic core while reconstructing natural expression in the target language. The humour theory component supplies the structured labels that force the model to separate background knowledge from the surprise element.

What would settle it

Take the same 500 jokes and translations, have native speakers rate humour, fluency, and coherence, then correlate their scores with the modified GEMBA scores; if the correlation is weak or HDM no longer beats the baselines under human ratings, the reported improvement is an artifact of the metric.

Watch

Extended reading notes

Core claim

The central claim is that HDM improves humour translation quality across multiple LLMs, datasets, and target languages. On the Short Jokes dataset translated from English to Chinese, HDM outperforms the DUAL-REFLECT and MAPS baselines on every model tested (Gemini1.5-Pro, Yi-Large, GPT3.5-Turbo, GPT4-Turbo) on all six GEMBA metrics, with humour score gains reaching an average of 11.2% for GPT4-Turbo. Generality experiments extend the gains to question-answer jokes and SemEval-2021 datasets, to Spanish and German translations, and to open-source models, with smaller models benefiting less. The paper also shows through ablation that both the decomposition mechanism and the humour theory component contribute, and that the effect is not tied to one prompt phrasing.

Load-bearing premise

The paper's central claim rests on the assumption that its automatic evaluation scores, produced by an AI judge with added humour keywords but never checked against human judgement, faithfully measure how funny, fluent, and coherent a translation is.

Editorial extensions

If this is right

  • HDM can be applied as a drop-in prompt strategy on top of existing translation LLMs without fine-tuning, making it easy to adopt in practice.
  • Humour translation quality improves for multiple language pairs (at least EN-ZH, EN-ES, EN-DE) and multiple datasets, so the method generalizes beyond a single test set.
  • The ablation shows that the decomposition mechanism is the main driver of improvement, while humour theory adds smaller but consistent gains.
  • Prompt variation experiments indicate that the method is not an artifact of one specific prompt template.

Reading between the lines

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

  • Beyond the paper's scope, HDM might transfer to other creative translation tasks where the punchline or effect must be re-created rather than literally translated, such as wordplay, poetry, or advertising slogans.
  • Because the paper's metric is a modified GEMBA without human validation, a natural next test is to compare HDM output against human humour ratings; the method's practical value will stand or fall on that correlation.
  • The failure case involving a pun on 'Fleece Navidad' suggests HDM currently handles puns that depend on cross-lingual sound similarity poorly; a hybrid that detects pun type before decomposition could be a testable extension.
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.

Referee Report

4 major / 5 minor

Summary. The paper proposes a Humour Decomposition Mechanism (HDM) for humour translation. HDM is a three-step Chain-of-Thought prompting procedure: (1) the model analyses a source joke into topic, angle, and punchline components; (2) the analysis is translated into the target language; and (3) a new target-language joke is generated from the translated analysis. The method is evaluated with GEMBA-SQM and GEMBA-STARS metrics for humour, fluency, and coherence, comparing HDM against DUAL-REFLECT and MAPS baselines on multiple LLMs, datasets, and language pairs. The central claim is that HDM significantly improves humour translation quality, with average gains of 7.75% in humour, 2.81% in fluency, and 6.13% in coherence.

Significance. If the reported gains were robust and the evaluation measured translation quality rather than only humorousness, HDM would be a simple, low-cost prompting strategy with broad potential applicability across models and languages. The paper's strengths are its consistent positive results across four API-based and five open-source models, its use of multiple datasets and language pairs, and its ablation and prompt-robustness analyses. These are useful empirical contributions. However, the evidence is weakened by the absence of any faithfulness/adequacy evaluation, the lack of validation of the modified GEMBA metric against human humour judgments, the unquantified discarding of invalid evaluator outputs, and the absence of statistical significance testing. These issues are load-bearing because the abstract claims improvements in translation quality, not merely in perceived humour.

major comments (4)
  1. [§3.1 and §4.2] The evaluation measures humour, fluency, and coherence but never measures adequacy or faithfulness to the source joke. The Humour Composition prompt in Eq. (3), 'Please generate a [target language] joke based on the analysis', imposes no constraint that the generated joke preserve the source joke's content. The success case P1 in Fig. 4 illustrates the consequence: the source joke about a snail riding on a turtle's back is rendered by HDM as a snail riding on a snail's back, an outright content error, and this is presented as a correct case. Because the reported gains are compatible with HDM discarding the source joke and generating a new, funnier one, the central claim that HDM improves translation quality is not established. I recommend adding a reference-based adequacy metric (e.g., COMET, chrF, or fine-grained human fidelity judgments) and/or constraining the composition step to preserve the source joke's semantic content.
  2. [§4.2 and §6] The paper modifies GEMBA prompts to incorporate humour keywords but provides no validation that the modified metric agrees with human humour judgments. The judge is GPT-4-based while many of the evaluated outputs are also produced by GPT-4-based systems, so evaluator bias is a real risk. The Limitation section concedes that human evaluation remains the gold standard, but this does not mitigate the absence of evidence that the modified GEMBA scores measure what they claim to measure. I request a correlation study against human humour ratings on a held-out subset, together with inter-annotator agreement statistics.
  3. [§4.2 and §4.3] No statistical significance testing is reported. Tables 1–4 present point estimates averaged over three runs, with no confidence intervals, variance, or paired significance tests. The abstract's claim that the method 'significantly improves' humour translation quality is therefore not supported statistically. I recommend reporting confidence intervals and conducting paired bootstrap or similar tests across the sampled jokes.
  4. [§4.2] Invalid evaluator outputs are discarded without reporting how many were omitted. The sentence 'we omit the invalid responses and retain only the valid results' gives no counts per model, metric, or condition. If the GPT-4 judge produced more invalid outputs for one condition than another, selective deletion could bias the comparison. Please report the number of discarded outputs and show that the main results are robust under alternative treatment of those cases.
minor comments (5)
  1. [§4.3 and Table 4] The text states that HDM performs consistently well across open-source models, but Table 4 shows a large drop for Qwen2.5-0.5B in SQM-H (39.80 to 28.70) and smaller drops for Llama3.1-8B in SQM-C (84.20 to 81.70). These failures should be acknowledged and discussed explicitly.
  2. [Fig. 3] The four prompt variants labelled V1–V4 are not defined anywhere in the text. Please specify the differences among them so the prompt-robustness claim can be assessed.
  3. [Fig. 2] The figure contains garbled labels and overlapping text; a clean, legible version is needed.
  4. [§4.1 and §5.1] The random selection of 500 and 100 samples is described without seeds or reproducible sample indices. Please provide this information for reproducibility.
  5. [Abstract and §4.2] The acronym GEMBA is expanded as 'Estimation Metric Based Assessment'; please verify this expansion against the cited reference, as it does not match the usual usage in the original paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: HDM is an empirical prompting intervention evaluated with an LLM-as-a-judge metric; no derivation or reported gain reduces by construction to its own inputs.

full rationale

The claimed derivation chain in HDM is an empirical prompting pipeline, not a formal derivation: Humour Decomposition produces an analysis Ai = arg max p(a | Li), the Translation Module produces A'i = arg max p(a' | Ai, S, T), and Humour Composition produces F = arg max p(f | A'i, T). These are conditional-generation steps with no identity between input and output, so the 'predicted' outputs are not equal to the inputs by construction. The evaluation in Section 4.2 uses GEMBA, an external LLM-based metric from Kocmi and Federmann [22], with prompts modified to include 'the keywords of humour, coherence and fluency based on [8]'. Reference [8] is co-authored by Yang Chen, a co-author of the present paper, but it is used only to supply keyword wording for the metric prompt; the metric's underlying validation is external, and the cited wording is not invoked as a uniqueness theorem or a fitted parameter. The absence of human evaluation and the lack of an adequacy/fidelity measure, including the P1 case in Figure 4 where HDM renders 'snail riding on turtle's back' as 'snail riding on a snail's back', are genuine construct-validity and evaluation-quality concerns, and the paper itself concedes that 'human evaluation remains the gold standard' and that the 'absence of human evaluation ... may introduce bias'. But those are validity limitations, not circular reductions: the reported gains do not become true by definition, and no fitted value is later relabelled as a prediction. The self-citation is minor and not load-bearing for the central claim, so the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the validity of the joke decomposition theory, on the unvalidated LLM-based evaluation metric, and on the representativeness of the selected samples. No fitted numeric parameters or invented entities are introduced. The main unrecognized risk is the evaluator bias.

assumptions (3)
  • domain assumption Toplyn's monologue joke theory (topic, angle, punchline) is a valid decomposition for humorous texts across languages.
    Section 3.2 states that the basic structure of humorous text is topic, angle, and punchline, but no empirical validation is provided for this decomposition across languages.
  • domain assumption GEMBA, an LLM-based metric, accurately measures humour, fluency, and coherence in translated jokes when prompts are modified to include these keywords.
    Section 4.2 says the original GEMBA prompts are modified using humour, coherence, and fluency keywords based on reference [8], but no validation against human judgment is provided for the humour dimension.
  • domain assumption The randomly selected 500 jokes from the Short Jokes Dataset are representative of the humour needed to support the claims.
    Section 4.1 states 'we randomly select 500 samples' with no stratification or description of humour types, which may limit generalizability across humour categories.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Psychology-Driven Enhancement of Humour Translation." pith.science (2026). https://pith.science/paper/7ERGC6IK

@misc{pith2026250709259,
  author       = {Pith},
  title        = {Pith review of: Psychology-Driven Enhancement of Humour Translation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ERGC6IK}},
  note         = {Machine review of arXiv:2507.09259}
}
read the original abstract

Humour translation plays a vital role as a bridge between different cultures, fostering understanding and communication. Although most existing Large Language Models (LLMs) are capable of general translation tasks, these models still struggle with humour translation, which is especially reflected through linguistic interference and lacking humour in translated text. In this paper, we propose a psychology-inspired Humour Decomposition Mechanism (HDM) that utilises Chain-of-Thought (CoT) to imitate the ability of the human thought process, stimulating LLMs to optimise the readability of translated humorous texts. Moreover, we integrate humour theory in HDM to further enhance the humorous elements in the translated text. Our automatic evaluation experiments on open-source humour datasets demonstrate that our method significantly improves the quality of humour translation, yielding average gains of 7.75\% in humour, 2.81\% in fluency, and 6.13\% in coherence of the generated text.

Figures

Figures reproduced from arXiv: 2507.09259 by the authors.

Figure 1
Figure 1. An example of humour translation from English to Chinese. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The workflow of HDM. 2.3 Chain-of-Thought (CoT) CoT prompting involves either providing instruction or a few chain-of-thought examples [18]. Recently, a series of studies [40, 46] have proposed their respective prompting strategies, breaking down the entire task into smaller components and then systematically addressing, strategizing, and carrying out each of these components. With the improvement of model capabilit… view at source ↗
Figure 3
Figure 3. Performance comparisons of four various prompts of HDM in humour, [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Some correct Chinese cases generated by HDM. We present the original [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 28 canonical work pages

  1. [1]

    arXiv preprint arXiv:2303.08774 (2023)

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al.: Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  2. [2]

    AI, ., :, Young, A., Chen, B., Li, C., Huang, C., Zhang, G., Zhang, G., Li, H., Zhu, J., Chen, J., Chang, J., Yu, K., Liu, P., Liu, Q., Yue, S., Yang, S., Yang, S., Yu, T., Xie, W., Huang, W., Hu, X., Ren, X., Niu, X., Nie, P., Xu, Y., Liu, Y., Wang, Y., Cai, Y., Gu, Z., Liu, Z., Dai, Z.: Yi: Open foundation models by 01.ai (2024)

    2024

  3. [3]

    arXiv preprint arXiv:2211.01889 (2022)

    Alnajjar, K., Hämäläinen, M., Tiedemann, J., Laaksonen, J., Kurimo, M.: When to laugh and how hard? a multimodal approach to detecting humor and its intensity. arXiv preprint arXiv:2211.01889 (2022)

    arXiv 2022

  4. [4]

    arXiv preprint arXiv:1607.00976 (2016)

    Amir, S., Wallace, B.C., Lyu, H., Silva, P.C.M.J.: Modelling context with user embeddings for sarcasm detection in social media. arXiv preprint arXiv:1607.00976 (2016)

    arXiv 2016

  5. [5]

    In: Ku, L.W., Martins, 14 F

    Chen, A., Lou, L., Chen, K., Bai, X., Xiang, Y., Yang, M., Zhao, T., Zhang, M.: DUAL-REFLECT: Enhancing large language models for reflective trans- lation through dual learning feedback mechanisms. In: Ku, L.W., Martins, 14 F. Author et al. A., Srikumar, V. (eds.) Proceedings of the 62nd Annual Meeting of the As- sociation for Computational Linguistics (V...

    work page

  6. [6]

    In: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Chen, G., Ma, S., Chen, Y., Zhang, D., Pan, J., Wang, W., Wei, F.: Towards making the most of cross-lingual transfer for zero-shot neural machine translation. In: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). pp. 142–157 (2022)

    work page 2022

  7. [7]

    Prompt to GPT-3: Step-by-Step Thinking Instructions for Humor Generation

    Chen, Y., Shi, B., Si, M.: Prompt to gpt-3: Step-by-step thinking instructions for humor generation. arXiv preprint arXiv:2306.13195 (2023)

    work page Pith review arXiv 2023

  8. [8]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Chen, Y., Yuan, Y., Liu, P., Liu, D., Guan, Q., Guo, M., Peng, H., Liu, B., Li, Z., Xiao, Y.: Talk funny! a large-scale humor response dataset with chain-of-humor interpretation. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 38, pp. 17826–17834 (2024)

    work page 2024

Show all 49 references
  1. [9]

    arXiv preprint arXiv:2107.00061 (2021)

    Clark, E., August, T., Serrano, S., Haduong, N., Gururangan, S., Smith, N.A.: All that’s’ human’is not gold: Evaluating human evaluation of generated text. arXiv preprint arXiv:2107.00061 (2021)

    2021 arXiv

  2. [10]

    International Journal of Communication10, 20 (2016)

    Gambier, Y.: Translations| rapid and radical changes in translation and translation studies. International Journal of Communication10, 20 (2016)

    2016

  3. [11]

    In: International Conference on Machine Learning

    Gao, L., Madaan, A., Zhou, S., Alon, U., Liu, P., Yang, Y., Callan, J., Neubig, G.: Pal: Program-aided language models. In: International Conference on Machine Learning. pp. 10764–10799. PMLR (2023)

    2023

  4. [12]

    In: Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021)

    García-Díaz, J.A., Valencia-García, R.: UMUTeam at SemEval-2021 task 7: De- tecting and rating humor and offense with linguistic features and word embed- dings. In: Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021). pp. 1096–1101. Association...

    2021

  5. [13]

    IEEE Transactions on Artificial Intelligence (2024)

    Hagos, D.H., Battle, R., Rawat, D.B.: Recent advances in generative ai and large language models: Current status, challenges, and perspectives. IEEE Transactions on Artificial Intelligence (2024)

    2024

  6. [14]

    In: Proceedings of the AAAI conference on artificial intelligence

    Hasan, M.K., Lee, S., Rahman, W., Zadeh, A., Mihalcea, R., Morency, L.P., Hoque, E.: Humor knowledge enriched transformer for understanding multimodal humor. In: Proceedings of the AAAI conference on artificial intelligence. vol. 35, pp. 12972– 12980 (2021)

    2021

  7. [15]

    Trans- actions of the Association for Computational Linguistics12, 229–246 (2024)

    He, Z., Liang, T., Jiao, W., Zhang, Z., Yang, Y., Wang, R., Tu, Z., Shi, S., Wang, X.: Exploring human-like translation strategy with large language models. Trans- actions of the Association for Computational Linguistics12, 229–246 (2024)

    2024

  8. [16]

    arXiv preprint arXiv:2302.09210 (2023)

    Hendy, A., Abdelrehim, M., Sharaf, A., Raunak, V., Gabr, M., Matsushita, H., Kim, Y.J., Afify, M., Awadalla, H.H.: How good are gpt models at machine trans- lation? a comprehensive evaluation. arXiv preprint arXiv:2302.09210 (2023)

    2023 arXiv

  9. [17]

    SKASE Journal of translation and interpretation2(1), 13–23 (2007)

    Hopkinson, C.: Factors in linguistic interference: A case study in translation. SKASE Journal of translation and interpretation2(1), 13–23 (2007)

    2007

  10. [18]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Ji, B., Liu, H., Du, M., Ng, S.K.: Chain-of-thought improves text generation with citations in large language models. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 38, pp. 18345–18353 (2024)

    2024

  11. [19]

    arXiv preprint arXiv:2301.087451(10) (2023)

    Jiao, W., Wang, W., Huang, J.t., Wang, X., Tu, Z.: Is chatgpt a good translator? a preliminary study. arXiv preprint arXiv:2301.087451(10) (2023)

    2023 arXiv

  12. [20]

    arXiv preprint arXiv:2304.03245 (2023) Psychology-Driven Enhancement of Humour Translation 15

    Karpinska, M., Iyyer, M.: Large language models effectively leverage document- level context for literary translation, but critical errors persist. arXiv preprint arXiv:2304.03245 (2023) Psychology-Driven Enhancement of Humour Translation 15

    2023 arXiv

  13. [21]

    arXiv preprint arXiv:2410.12842 (2024)

    Kenneth, M.O., Khosmood, F., Edalat, A.: A two-model approach for humour style recognition. arXiv preprint arXiv:2410.12842 (2024)

    2024 arXiv

  14. [22]

    arXiv preprint arXiv:2302.14520 (2023)

    Kocmi, T., Federmann, C.: Large language models are state-of-the-art evaluators of translation quality. arXiv preprint arXiv:2302.14520 (2023)

    2023 arXiv

  15. [23]

    NEJM AI1(8), AIoa2400196 (2024)

    Liang, W., Zhang, Y., Cao, H., Wang, B., Ding, D.Y., Yang, X., Vodrahalli, K., He, S., Smith, D.S., Yin, Y., et al.: Can large language models provide useful feedback on research papers? a large-scale empirical analysis. NEJM AI1(8), AIoa2400196 (2024)

    2024

  16. [24]

    Babel66(3), 434–456 (2020)

    Ma, X., Cheung, A.K.: Language interference in english-chinese simultaneous in- terpreting with and without text. Babel66(3), 434–456 (2020)

    2020

  17. [25]

    Brill Archive (1964)

    Nida, E.A.: Toward a science of translating: with special reference to principles and procedures involved in Bible translating. Brill Archive (1964)

    1964

  18. [26]

    Routledge (2023)

    Pym, A.: Exploring translation theories. Routledge (2023)

    2023

  19. [27]

    In: Annual Meeting of the Berkeley Linguistics Society

    Raskin, V.: Semantic mechanisms of humor. In: Annual Meeting of the Berkeley Linguistics Society. pp. 325–335 (1979)

    1979

  20. [28]

    arXiv preprint arXiv:2009.09025 (2020)

    Rei, R., Stewart, C., Farinha, A.C., Lavie, A.: Comet: A neural framework for mt evaluation. arXiv preprint arXiv:2009.09025 (2020)

    2020 arXiv

  21. [29]

    Available at: https://www.kaggle.com/datasets/jiriroz/qa-jokes

    Roznovjak, J.: Question-answer jokes (2016), kaggle, Data set. Available at: https://www.kaggle.com/datasets/jiriroz/qa-jokes

    2016

  22. [30]

    In: Jurafsky, D., Chai, J., Schluter, N., Tetreault, J

    Sellam, T., Das, D., Parikh, A.: BLEURT: Learning robust metrics for text generation. In: Jurafsky, D., Chai, J., Schluter, N., Tetreault, J. (eds.) Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 7881–7892. Association for Computat...

    2020 doi

  23. [31]

    arXiv preprint arXiv:2403.05530 (2024)

    Team, G., Georgiev, P., Lei, V.I., Burnell, R., Bai, L., Gulati, A., Tanzer, G., Vin- cent,D.,Pan,Z.,Wang,S.,etal.:Gemini1.5:Unlockingmultimodalunderstanding across millions of tokens of context. arXiv preprint arXiv:2403.05530 (2024)

    2024 arXiv

  24. [32]

    Twenty Lane Media, LLC (2014), https://books.google.co.nz/books?id=yYKooAEACAAJ

    Toplyn, J.: Comedy Writing for Late-night Tv: How to Write Mono- logue Jokes, Desk Pieces, Sketches, Parodies, Audience Pieces, Re- motes, and Other Short-form Comedy. Twenty Lane Media, LLC (2014), https://books.google.co.nz/books?id=yYKooAEACAAJ

    2014

  25. [33]

    Routledge (2016)

    Vandaele, J.: Translating humour. Routledge (2016)

    2016

  26. [34]

    arXiv preprint arXiv:2211.09102 (2022)

    Vilar, D., Freitag, M., Cherry, C., Luo, J., Ratnakar, V., Foster, G.: Prompt- ing palm for translation: Assessing strategies and performance. arXiv preprint arXiv:2211.09102 (2022)

    2022 arXiv

  27. [35]

    arXiv preprint arXiv:2204.06175 (2022)

    Wang, D., Fan, K., Chen, B., Xiong, D.: Efficient cluster-based k-nearest-neighbor machine translation. arXiv preprint arXiv:2204.06175 (2022)

    2022 arXiv

  28. [36]

    arXiv preprint arXiv:2203.11171 (2022)

    Wang, X., Wei, J., Schuurmans, D., Le, Q., Chi, E., Narang, S., Chowdhery, A., Zhou, D.: Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171 (2022)

    2022 arXiv

  29. [37]

    Advances in neural information processing systems35, 24824–24837 (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D., et al.: Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems35, 24824–24837 (2022)

    2022

  30. [38]

    Cadernos de Tradução43, e89705 (2023)

    Xia, C., Amini, M., Lee, K.F.: Humor translation: A case study on the loss of hu- morous loads in spongebob squarepants. Cadernos de Tradução43, e89705 (2023)

    2023

  31. [39]

    5 technical report

    Yang, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Li, C., Liu, D., Huang, F., Wei, H., et al.: Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115 (2024)

    2024 arXiv

  32. [40]

    arXiv preprint arXiv:2302.04813 (2023) 16 F

    Ye, X., Durrett, G.: Explanation selection using unlabeled data for chain-of- thought prompting. arXiv preprint arXiv:2302.04813 (2023) 16 F. Author et al

    2023 arXiv

  33. [41]

    Zabalbeascoa, P.: Humor and translation—an interdiscipline (2005)

    2005

  34. [42]

    In: International Conference on Machine Learning

    Zhang, B., Haddow, B., Birch, A.: Prompting large language model for machine translation: A case study. In: International Conference on Machine Learning. pp. 41092–41110. PMLR (2023)

    2023

  35. [43]

    arXiv preprint arXiv:2210.03493 (2022)

    Zhang, Z., Zhang, A., Li, M., Smola, A.: Automatic chain of thought prompting in large language models. arXiv preprint arXiv:2210.03493 (2022)

    2022 arXiv

  36. [44]

    arXiv preprint arXiv:2411.14405 (2024)

    Zhao, Y., Yin, H., Zeng, B., Wang, H., Shi, T., Lyu, C., Wang, L., Luo, W., Zhang, K.: Marco-o1: Towards open reasoning models for open-ended solutions. arXiv preprint arXiv:2411.14405 (2024)

    2024 arXiv

  37. [45]

    Zheng, L., Chiang, W.L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E.P., Zhang, H., Gonzalez, J.E., Stoica, I.: Judging llm-as-a-judge with mt-bench and chatbot arena (2023)

    2023

  38. [46]

    arXiv preprint arXiv:2205.10625 (2022)

    Zhou, D., Schärli, N., Hou, L., Wei, J., Scales, N., Wang, X., Schuurmans, D., Cui, C., Bousquet, O., Le, Q., et al.: Least-to-most prompting enables complex reasoning in large language models. arXiv preprint arXiv:2205.10625 (2022)

    2022 arXiv

  39. [47]

    arXiv preprint arXiv:2211.01910 (2022)

    Zhou, Y., Muresanu, A.I., Han, Z., Paster, K., Pitis, S., Chan, H., Ba, J.: Large lan- guage models are human-level prompt engineers. arXiv preprint arXiv:2211.01910 (2022)

    2022 arXiv

  40. [48]

    arXiv preprint arXiv:2304.04675 (2023)

    Zhu, W., Liu, H., Dong, Q., Xu, J., Huang, S., Kong, L., Chen, J., Li, L.: Mul- tilingual machine translation with large language models: Empirical results and analysis. arXiv preprint arXiv:2304.04675 (2023)

    2023 arXiv

  41. [704]

    Association for Computational Linguistics, Bangkok, Thailand (Aug 2024), https://aclanthology.org/2024.acl-short.64

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.