Pith. sign in

REVIEW 4 major objections 5 minor 73 references

Culturally loaded translation is a distinct failure mode for frontier LLMs, and the usual tools for judging translations misjudge it.

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 10:21 UTC pith:NGRXYJ4G

load-bearing objection Useful new corpus and a likely-true task-difficulty story, but the headline human-disagreement claim rests on a protocol contradiction and unmeasured rater variance. the 4 major comments →

arxiv 2607.20241 v1 pith:NGRXYJ4G submitted 2026-07-22 cs.CL cs.AI

On the Systematic Challenges of Culturally Loaded Machine Translation: Dream of the Red Chamber as the Cultural Lens

Yiming Wang , Jiayuan Di This is my paper
classification cs.CL cs.AI
keywords culturally loaded translationmachine translation evaluationDream of the Red Chamberlarge language modelshuman evaluationautomatic metricscultural categoriesChinese-Japanese translation
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.

This paper tries to establish that translating culturally charged passages remains an unsolved problem for today's best language models, even in a language pair as culturally close as Chinese and Japanese. Using five hundred segments from Dream of the Red Chamber, it finds that the strongest model still trails a professional human translation by roughly a third of a point on a five-point scale, and that the gap is concentrated in cultural dimensions—appropriateness and native readability—rather than in accuracy or fluency. It further claims that human evaluation of such translations is systematically colored by the evaluator's own cultural background and expertise: students score higher than professors, and source-culture and target-culture judges weigh different things. Finally, it claims that common automatic metrics, including lexical, semantic, and LLM-as-judge metrics, correlate only weakly with human rankings and fail entirely at the level of individual sentences. The reason to care: if these claims hold, progress on culturally aware translation cannot be measured with current benchmarks, and the bottleneck is not linguistic fluency but cultural transfer.

Core claim

On its own terms, the paper's central discovery is a three-part empirical finding. First, frontier LLMs underperform on culturally loaded translation: the best model scores 3.89 versus 4.27 for the human reference overall, and scores on the two cultural dimensions are consistently lower than on general accuracy and fluency, with different models ranking differently on cultural versus general competence. Second, human judgment of these translations is not stable across evaluator backgrounds: students are more lenient than professors, Japanese and Chinese evaluators emphasize target appropriateness versus source faithfulness respectively, and the native-readability dimension shows disagreement

What carries the argument

The load-bearing object is a purpose-built bilingual evaluation corpus: five hundred culturally loaded segments from a single canonical Chinese novel, balanced across five cultural categories (ecology, religion, material culture, linguistics, society), paired with a reference translation by a native Japanese literary translator, and judged on four dimensions—content accuracy, language fluency, cultural appropriateness, and native readability—plus an overall score. The protocol that carries the argument is a four-group human evaluation design (source-culture and target-culture evaluators, crossed with students and professors) that is treated as the gold standard for all model comparisons, and

Load-bearing premise

The load-bearing premise is that the four-evaluator-per-group split design measures genuine differences between evaluator backgrounds; if within-group judgment variability is comparable to the observed between-group gaps, the human-disagreement findings and, through them, every model comparison built on those human scores could be sampling noise.

What would settle it

Re-run the human evaluation with all 500 segments scored by every evaluator in each group and compute standard inter-annotator agreement within each group and between groups. If within-group agreement is low or between-group differences shrink below the reported effect sizes, the evaluator-background claims collapse. Separately, if any culture-aware automatic metric (for instance, one that rewards parenthetical explanation of cultural terms) achieves a strong sample-level correlation with human scores on this dataset, the 'metrics totally fail' claim would be refuted.

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

If this is right

  • Benchmarks for culture-aware MT should include human evaluation stratified by evaluator background; a single evaluator pool can produce systematically biased rankings.
  • Automatic metric scores cannot be used to select or rank outputs on culturally loaded content; culture-aware metrics or evaluation designs need to be built.
  • Model development should target cultural dimensions directly: gains in accuracy and fluency do not transfer to cultural appropriateness or readability.
  • Reasoning (chain-of-thought) is not a shortcut: reasoning and non-reasoning models perform similarly, and reasoning models over-explain without improving cultural transfer.
  • The category difficulty hierarchy (material and ecology easier than linguistics and society) provides a concrete diagnostic axis for future MT evaluation.

Where Pith is reading between the lines

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

  • An omitted check that would test the paper's human-disagreement claim: rerun the evaluation with every translation scored by all evaluators and report inter-annotator agreement; without that, the observed student-versus-professor and Chinese-versus-Japanese gaps could be sampling noise.
  • The strong preference for domestication suggests a plausible fix worth probing: prompting models to preserve source imagery with in-text explanation (like the reference's parenthetical annotations) could reduce cultural loss without hurting readability.
  • The failure of general metrics at sample level implies that any culture-aware metric must be trained or evaluated on category-specific items; the five-category taxonomy offers a natural split for building such a metric.
  • Because the language pair is culturally close, the same protocol applied to a distant pair (e.g., Chinese-English) would likely show larger model-human gaps; the paper's limitation section acknowledges this, and its open-sourced data makes that extension feasible.

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

4 major / 5 minor

Summary. The paper constructs a Chinese–Japanese bilingual dataset of 500 culturally loaded segments from Dream of the Red Chamber, classified into Nida's five cultural categories, and evaluates eight LLMs plus a human reference translation on four MQM-inspired dimensions (Content Accuracy, Language Fluency, Cultural Appropriateness, Native Readability) plus an overall score. Human evaluators are drawn from four groups defined by native language (Chinese/Japanese) and academic level (student/professor). The authors report three main challenges: (1) task difficulty, with frontier LLMs underperforming relative to human references and showing specific weaknesses on cultural dimensions; (2) human evaluation disagreement, with evaluator background affecting scores, especially on Native Readability and between students and professors; and (3) automatic evaluation unreliability, with BLEU, xCOMET, and LLM-as-a-judge showing weak system-level or sample-level correlations with human judgments. Extended analyses examine error types and domestication/foreignization strategies.

Significance. If the empirical claims hold, the paper offers a useful, culturally grounded MT benchmark for a non-Western language pair and draws attention to real issues in human and automatic evaluation of culture-loaded translation. The choice of an authoritative bilingual edition, the five-category taxonomy, the inclusion of both reasoning and non-reasoning model families, and the close reading of translation strategies in the case studies are genuine strengths. However, the central claims currently rest on human group-mean scores from a split-sample design with no inter-annotator agreement or significance testing, and the Native Readability protocol contains an apparent inconsistency with the reported data. The paper's significance is therefore conditional on resolving these evaluation-protocol issues.

major comments (4)
  1. [§4.1 / Fig.2 / §5.2] Section 4.1 states: 'For Read., scores are assigned only by target-language natives.' Since the target language is Japanese, Chinese/Student and Chinese/Professor raters should not produce Native Readability scores. Yet Fig.2's readability subplot and its caption ('scored by four evaluator groups') present scores for all four groups, and §5.2's headline disagreement example ('students and professors gave average scores of 3.70 and 4.52' on Read.) is not restricted to Japanese raters. This creates a direct provenance conflict in the gold-standard data. The authors must either confirm that Chinese groups scored Native Readability and amend §4.1, or confirm they did not and replot/restrict that part of the analysis. As printed, the reader cannot determine which, and this affects the main human-disagreement claim.
  2. [§4.1 / §5.2] The split-sample design confounds rater identity with sample identity: within each group, four evaluators each rate a disjoint 125-sample set, so every sample receives exactly one score per group. The paper reports no inter-annotator agreement, no within-group variance, and no significance tests for the claims that students 'consistently' score higher than professors or that the evaluator-group gap 'approaches 1 point.' The verbal statement that pilot assessments verified 'low judgment variance' does not substitute for final-rating reliability statistics or a mixed-effects analysis. Because these human scores are the gold standard for every model comparison in §5.1 and every metric correlation in §5.3, the evaluator-background finding and the reliability of the overall gold standard are not established at the stated strength.
  3. [§5.3 / Fig.3] With only eight models, the reported system-level correlations are not statistically distinguishable from zero: for xCOMET Spearman ρ=0.43 (p≈0.29 by a standard Spearman test) and for the reasoning LLM-judge ρ=0.48 (p≈0.23). No p-values or confidence intervals are reported. The text says xCOMET and LLM judges show 'modest positive correlations' that are 'somewhat more informative' than BLEU, but the evidence supports only the weaker claim that no reliable ranking signal was observed. This overstates the strength of the system-level result and should be reanalyzed or rephrased.
  4. [§5.3 / Fig.4] The sample-level conclusion that automatic metrics 'totally fail to distinguish sample-level quality differences' is too absolute. Many Pearson r values are below 0.15, but several are nontrivial with n=500 (e.g., DS-r1/xCOMET r=0.34; Claude-4/reasoning LLM-judge r=0.27). A fairer statement is that the metrics carry weak, model-dependent signal. 'Total failure' would require a decision-theoretic or variance-decomposition argument that the manuscript does not provide. Please soften or quantify this claim.
minor comments (5)
  1. [§2.3] Typo: 'thoery' should be 'theory'.
  2. [Title/headings] 'Dream of the Red Chamberas the Cultural Lens' has a spacing issue; should be 'Dream of the Red Chamber as the Cultural Lens.'
  3. [§3.4 / §C.2] The LLM-as-a-judge prompt in §C.2 reuses the same four evaluation dimensions and overall scoring rubric as the human guidelines in Table 12. This should be explicitly acknowledged as a potential source of correlation between LLM-judge and human scores; it is not an entirely independent automatic judge.
  4. [§7/Limitations] The Limitations section promises to 'open-source our data sources and provide raw multilingual corpus,' but no repository link or access information is provided in the manuscript. Please include it in the final version.
  5. [Tables/References] Some table and reference formatting is inconsistent (e.g., model abbreviation 'Qwen3-235B-A22-Non-Thinking' vs 'A22B'; author lists truncated with 'and 1 others'). Clean up for the camera-ready version.

Circularity Check

0 steps flagged

No significant circularity; the paper is an empirical evaluation against an external human gold standard.

full rationale

The paper's three central claims (task difficulty, human-evaluation disagreement, automatic-evaluation unreliability) are empirical measurements, not derivations from fitted inputs. No parameter is fitted to a subset of data and then relabeled as a prediction, and no load-bearing argument reduces to a self-citation. Human judgments serve as an external gold standard, and model outputs are compared against a published human translation (Ito Sohei's version), so the main comparisons are self-contained. The only self-referential element is the LLM-as-a-judge prompt in §C.2, which reuses the same four dimensions and overall-scale wording as the human evaluation criteria; however, this does not make the metric correlation circular—if anything, sharing the rubric makes the observed low correlations a stronger, not weaker, indication of unreliability. One internal-consistency concern is noted but is not circularity: §4.1 states that 'For Read., scores are assigned only by target-language natives,' while Fig. 2 displays Native Readability scores for all four evaluator groups, including Chinese/Student and Chinese/Professor raters, and §5.2's disagreement analysis leans on Read. gaps. This is a data-provenance/validity issue that could affect the human-disagreement and downstream metric-reliability conclusions, but it is not a case of a prediction reducing to its inputs by construction. No circular chain of reasoning was found, so the score is 0.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

No free parameters are fitted; the central load is the annotation and evaluation protocol rather than a model. The axioms are domain assumptions about taxonomy validity, human evaluation as gold standard, reference quality, and representativeness.

axioms (4)
  • domain assumption Nida's five-category cultural taxonomy (ecology, religion, material, linguistics, society) cleanly applies to Dream of the Red Chamber segments.
    Adopted in §2.2 as the annotation scheme; if the categories are not exhaustive or separable, category-level difficulty conclusions shift.
  • domain assumption Human MQM-style evaluation is a valid gold standard for culturally loaded translation even though each translation is scored by one evaluator per group.
    §4.1 treats human scores as ground truth; the reliability evidence is pilot-based and does not cover the actual split-scoring design.
  • domain assumption Ito Sohei's Japanese translation is an appropriate high-quality reference for culturally loaded content.
    §3.3 argues Ito's balance of fidelity and annotation is the most authoritative reference; if it is idiosyncratic, model-versus-reference gaps partly reflect translation style rather than model deficit.
  • domain assumption Chinese-to-Japanese 'intermediate cultural distance' makes this pair representative of culturally loaded translation challenges.
    §3.2 proposes this balance; the authors themselves note in the Limitations that findings may not generalize to other language pairs.

pith-pipeline@v1.3.0-alltime-deepseek · 26076 in / 9486 out tokens · 77889 ms · 2026-08-01T10:21:52.774981+00:00 · methodology

0 comments
read the original abstract

Culturally loaded translation poses unique challenges for machine translation (MT), as meanings are deeply embedded in socio-cultural contexts beyond surface linguistic forms. Although large language models (LLMs) have enabled MT systems to achieve human-like quality in many scenarios, their ability to handle culturally loaded expressions remains underexplored. In this study, we systematically investigate the challenges posed by culturally loaded translation in LLM-based MT systems. We construct a Chinese-Japanese bilingual dataset from the culturally representative corpus Dream of the Red Chamber, containing 500 segments across diverse cultural categories. Using a comprehensive evaluation protocol, we reveal three main challenges: (1) task challenges, where frontier LLMs exhibit notable performance gaps and struggle with culturally loaded content; (2) human evaluation challenges, where evaluator backgrounds lead to substantial disagreement in translation judgments; and (3) automatic evaluation challenges, where widely used metrics fail to reliably assess translation quality for this task. These findings may offer valuable insights for culture-oriented translation research in both computational science and linguistics.

Figures

Figures reproduced from arXiv: 2607.20241 by Jiayuan Di, Yiming Wang.

Figure 1
Figure 1. Figure 1: Dataset examples from our Dream of the Red Chamber corpus across the five cultural categories. Detailed category definitions are in §2.2, and more examples are provided in §B.1. alized nature. Evaluating such translations adds another layer of difficulty. Translation quality is already subjective, and cultural elements amplify this subjectivity: evaluators from different cultural backgrounds may judge the … view at source ↗
Figure 2
Figure 2. Figure 2: Human evaluation results for eight models across four dimensions, scored by four evaluator groups: [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: System-level correlation between human overall scores and automatic metrics. In each subplot, models on [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sample-level correlation between human overall scores and automatic metrics. Each subplot corresponds [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Original PDF format of our corpus with direct bilingual comparisons (Case 1). [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Original PDF format of our corpus with direct bilingual comparisons (Case 2). [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of translation error types across eight models. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Distribution of culture-oriented translation strategy types across eight models. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] 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

73 extracted references · 10 linked inside Pith

  1. [1]

    Nature , volume=

    DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning , author=. Nature , volume=. 2025 , publisher=

    2025

  2. [2]

    arXiv preprint arXiv:2412.19437 , year=

    Deepseek-v3 technical report , author=. arXiv preprint arXiv:2412.19437 , year=

    Pith/arXiv arXiv

  3. [3]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    Pith/arXiv arXiv

  4. [4]

    arXiv preprint arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities , author=. arXiv preprint arXiv:2507.06261 , year=

    Pith/arXiv arXiv

  5. [5]

    2025 , url =

    Introducing GPT-4.1 in the API , author =. 2025 , url =

    2025

  6. [6]

    2025 , url =

    Introducing o3 and o4-mini , author =. 2025 , url =

    2025

  7. [7]

    2025 , url =

    Introducing Claude 4 , author =. 2025 , url =

    2025

  8. [8]

    2025 , url =

    Gemini 3 Pro: the frontier of vision AI , author =. 2025 , url =

    2025

  9. [9]

    arXiv preprint arXiv:2601.03267 , year=

    Openai gpt-5 system card , author=. arXiv preprint arXiv:2601.03267 , year=

    Pith/arXiv arXiv

  10. [10]

    Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

    Bleu: a method for automatic evaluation of machine translation , author=. Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

  11. [11]

    Advances in neural information processing systems , volume=

    Judging llm-as-a-judge with mt-bench and chatbot arena , author=. Advances in neural information processing systems , volume=

  12. [12]

    Transactions of the Association for Computational Linguistics , volume=

    xcomet: Transparent machine translation evaluation through fine-grained error detection , author=. Transactions of the Association for Computational Linguistics , volume=. 2024 , publisher=

    2024

  13. [13]

    Word , volume=

    Linguistics and ethnology in translation-problems , author=. Word , volume=. 1945 , publisher=

    1945

  14. [14]

    (No Title) , year=

    Language, culture, and translating , author=. (No Title) , year=

  15. [15]

    A companion to translation studies , volume=

    Culture and translation , author=. A companion to translation studies , volume=. 2007 , publisher=

    2007

  16. [16]

    1964 , publisher=

    Toward a science of translating: With special reference to principles and procedures involved in Bible translating , author=. 1964 , publisher=

    1964

  17. [17]

    1997 , publisher=

    Meaning-based translation: A guide to cross-language equivalence , author=. 1997 , publisher=

    1997

  18. [18]

    Culturally loaded words and English language teaching , author=

  19. [19]

    IRAL: International Review of Applied Linguistics in Language Teaching , volume=

    Mona Baker, In Other Words: A Coursebook on Translation (Book Review) , author=. IRAL: International Review of Applied Linguistics in Language Teaching , volume=. 1994 , publisher=

    1994

  20. [20]

    1988 , publisher=

    A textbook of translation , author=. 1988 , publisher=

    1988

  21. [21]

    London and New York: Routledge , year=

    In other words: A coursebook on translation , author=. London and New York: Routledge , year=

  22. [22]

    1995 , publisher =

    Venuti, Lawrence , title =. 1995 , publisher =

    1995

  23. [23]

    1998 , publisher=

    More paragraphs on translation , author=. 1998 , publisher=

    1998

  24. [24]

    Spirit Food

    Culture through Keywords in Kenneth Wong's Translation of" Spirit Food" by Nu Nu Yi (Inwa) , author=. Journal of English Language and Linguistics , volume=

  25. [25]

    Journal of Literature and Art Studies , year=

    English Translation of Culture-Loaded Words—A Corpus Based Study , author=. Journal of Literature and Art Studies , year=

  26. [26]

    Rajapark Journal , volume=

    A Study of Culture-Specific Items (CSIs) and Translation Strategies in The Blind Earthworm in the Labyrinth , author=. Rajapark Journal , volume=

  27. [27]

    International Journal of Language, Literacy and Translation , volume=

    Cultural Representation in Children’s Cartoon Programmes: Insights from the Nida/Newmark Typology , author=. International Journal of Language, Literacy and Translation , volume=

  28. [28]

    Computational linguistics , volume=

    The mathematics of statistical machine translation: Parameter estimation , author=. Computational linguistics , volume=

  29. [29]

    COLING 1996 Volume 2: The 16th International Conference on Computational Linguistics , year=

    HMM-based word alignment in statistical translation , author=. COLING 1996 Volume 2: The 16th International Conference on Computational Linguistics , year=

    1996

  30. [30]

    Proceedings of the 40th Annual meeting of the Association for Computational Linguistics , pages=

    Discriminative training and maximum entropy models for statistical machine translation , author=. Proceedings of the 40th Annual meeting of the Association for Computational Linguistics , pages=

  31. [31]

    arXiv preprint arXiv:1409.0473 , year=

    Neural machine translation by jointly learning to align and translate , author=. arXiv preprint arXiv:1409.0473 , year=

    Pith/arXiv arXiv

  32. [32]

    Advances in neural information processing systems , volume=

    Sequence to sequence learning with neural networks , author=. Advances in neural information processing systems , volume=

  33. [33]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

  34. [34]

    OpenAI Blog , year=

    Improving language understanding by generative pre-training , author=. OpenAI Blog , year=

  35. [35]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

  36. [36]

    arXiv preprint arXiv:2307.09288 , year=

    Llama 2: Open foundation and fine-tuned chat models , author=. arXiv preprint arXiv:2307.09288 , year=

    Pith/arXiv arXiv

  37. [37]

    arXiv preprint arXiv:2309.16609 , year=

    Qwen technical report , author=. arXiv preprint arXiv:2309.16609 , year=

    Pith/arXiv arXiv

  38. [38]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Prompting palm for translation: Assessing strategies and performance , author=. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  39. [39]

    Transactions of the Association for Computational Linguistics , volume=

    Exploring human-like translation strategy with large language models , author=. Transactions of the Association for Computational Linguistics , volume=. 2024 , publisher=

    2024

  40. [40]

    arXiv preprint arXiv:2301.08745 , year=

    Is ChatGPT a good translator? Yes with GPT-4 as the engine , author=. arXiv preprint arXiv:2301.08745 , year=

    Pith/arXiv arXiv

  41. [41]

    Proceedings of the sixth conference on machine translation , pages=

    Findings of the 2021 conference on machine translation (WMT21) , author=. Proceedings of the sixth conference on machine translation , pages=

    2021

  42. [42]

    Proceedings of the Seventh Conference on Machine Translation (WMT) , pages=

    Findings of the 2022 conference on machine translation (WMT22) , author=. Proceedings of the Seventh Conference on Machine Translation (WMT) , pages=

    2022

  43. [43]

    Proceedings of Translating and the Computer 35 , year=

    Multidimensional quality metrics: a flexible system for assessing translation quality , author=. Proceedings of Translating and the Computer 35 , year=

  44. [44]

    arXiv preprint arXiv:2504.21318 , year=

    Phi-4-reasoning technical report , author=. arXiv preprint arXiv:2504.21318 , year=

    Pith/arXiv arXiv

  45. [45]

    arXiv preprint arXiv:2505.05410 , year=

    Reasoning models don't always say what they think , author=. arXiv preprint arXiv:2505.05410 , year=

    Pith/arXiv arXiv

  46. [46]

    arXiv preprint arXiv:2504.18428 , year=

    Polymath: Evaluating mathematical reasoning in multilingual contexts , author=. arXiv preprint arXiv:2504.18428 , year=

    arXiv

  47. [47]

    1995 , publisher=

    The problem of a Chinese aesthetic , author=. 1995 , publisher=

    1995

  48. [48]

    Dream of the Red Chamber

    Archetype and Allegory in the" Dream of the Red Chamber" , author=. 2015 , publisher=

    2015

  49. [49]

    China Review International , year=

    Fictions of Enlightenment: Journey to the West, Tower of Myriad Mirrors and Dream of the Red Chamber, and: Androgyny in Late Ming and Early Qing Literature (review) , author=. China Review International , year=

  50. [50]

    Cross-cultural Communication , year=

    On English Translation of Culture-Specific Items in the Ancient Chinese Official System:A Descriptive and Comparative Study on Hawkes’ and Yangs’ English Translated Cases of Hong Lou Meng , author=. Cross-cultural Communication , year=

  51. [51]

    2022 , publisher=

    Dream of the Red Chamber: Literary and translation perspectives , author=. 2022 , publisher=

    2022

  52. [52]

    2019 , issn =

    Dan Song , title =. 2019 , issn =

    2019

  53. [53]

    1998 , publisher=

    Hiroaki Maruyama , journal=. 1998 , publisher=

    1998

  54. [54]

    1990 , url=

    Translation, History and Culture , author=. 1990 , url=

    1990

  55. [55]

    1995 , url=

    Descriptive translation studies and beyond , author=. 1995 , url=

    1995

  56. [56]

    Information, Communication & Society , volume=

    Understanding the societal impacts of machine translation: a critical review of the literature on medical and legal use cases , author=. Information, Communication & Society , volume=. 2021 , publisher=

    2021

  57. [57]

    Computational Linguistics , volume=

    Machine translation meta evaluation through translation accuracy challenge sets , author=. Computational Linguistics , volume=

  58. [58]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Liaozhai through the Looking-Glass: On Paratextual Explicitation of Culture-Bound Terms in Machine Translation , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  59. [59]

    How good are LLMs for literary translation, really? Literary translation evaluation with humans and LLMs , author=. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2025

  60. [60]

    Advances in neural information processing systems , volume=

    Chain-of-thought prompting elicits reasoning in large language models , author=. Advances in neural information processing systems , volume=

  61. [61]

    Translation, power, subversion , pages=

    Culture-specific items in translation , author=. Translation, power, subversion , pages=. 1996 , organization=

    1996

  62. [62]

    2019 , address =

    Wu, Jun , title =. 2019 , address =

    2019

  63. [63]

    Sun, Yuming , journal=

  64. [64]

    1993 , month =

    Hu, Wenbin , title =. 1993 , month =

    1993

  65. [65]

    Proceedings of the Ninth Conference on Machine Translation , pages=

    Findings of the WMT24 general machine translation shared task: The LLM era is here but MT is not solved yet , author=. Proceedings of the Ninth Conference on Machine Translation , pages=

  66. [66]

    Transactions of the Association for Computational Linguistics , volume=

    Salute the classic: Revisiting challenges of machine translation in the age of large language models , author=. Transactions of the Association for Computational Linguistics , volume=. 2025 , publisher=

    2025

  67. [67]

    Information , volume=

    Machine translation in the era of large language models: a survey of historical and emerging problems , author=. Information , volume=. 2025 , publisher=

    2025

  68. [68]

    Transactions of the Association for Computational Linguistics , volume=

    Frmt: A benchmark for few-shot region-aware machine translation , author=. Transactions of the Association for Computational Linguistics , volume=. 2023 , publisher=

    2023

  69. [69]

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

    Challenges and strategies in cross-cultural NLP , author=. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  70. [70]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    Benchmarking machine translation with cultural awareness , author=. Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    2024

  71. [71]

    Transactions of the Association for Computational Linguistics , volume=

    Cultural adaptation of recipes , author=. Transactions of the Association for Computational Linguistics , volume=. 2024 , publisher=

    2024

  72. [72]

    Proceedings of the Ninth Conference on Machine Translation , pages=

    Cultural adaptation of menus: A fine-grained approach , author=. Proceedings of the Ninth Conference on Machine Translation , pages=

  73. [73]

    Proceedings of the 31st International Conference on Computational Linguistics: System Demonstrations , pages=

    CULTURALLY YOURS: A reading assistant for cross-cultural content , author=. Proceedings of the 31st International Conference on Computational Linguistics: System Demonstrations , pages=