The DOI matches its Crossref record; 38 of the 43 references it cites were found in a registry or an index.
Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation
Open accessAbstract
Human evaluation of modern high-quality machine translation systems is a difficult problem, and there is increasing evidence that inadequate evaluation procedures can lead to erroneous conclusions. While there has been considerable research on human evaluation, the field still lacks a commonly accepted standard procedure. As a step toward this goal, we propose an evaluation methodology grounded in explicit error analysis, based on the Multidimensional Quality Metrics (MQM) framework. We carry out the largest MQM research study to date, scoring the outputs of top systems from the WMT 2020 shared task in two language pairs using annotations provided by professional translators with access to full document context. We analyze the resulting data extensively, finding among other results a substantially different ranking of evaluated systems from the one established by the WMT crowd workers, exhibiting a clear preference for human over machine output. Surprisingly, we also find that automatic metrics based on pre-trained embeddings can outperform human crowd workers. We make our corpus publicly available for further research.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“These metrics offer notable advantages in that they are (theoretically) fully reproducible, far easier and cheaper to compute, and can avoid some of the issues faced by human studies (Wei and Jia, 2021; Freitag et al., 2021; Amidei et al., 2020).”
From Lessons from the Trenches on Reproducible Evaluation of Language Models (2024) · cites this paper as Freitag et al., 2021
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“While COMET generally aligns better with human judgement than BLEU [2]”
From Metric Inversion in Generative AI Translation: Divergence Between Automated Evaluation Metrics and Human Expert Judgement in the LRM Era (2026) · cites this paper as [2]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
chrF: character n-gram F-score for automatic MT evaluation
Popović, M. (2015) · Proceedings of the Tenth Workshop on Statistical Machine Translation · doi:10.18653/v1/w15-3049 · Cited by 1,220 · cited 1×
“In general, metrics that are not solely based on surface characteristics do somewhat better, though this pattern is not consistent (for example, chrF (Popović, 2015) has a high correlation of 0.”
BackgroundCOMET: A Neural Framework for MT Evaluation
Rei, R., Stewart, C., Farinha, A. C., & Lavie, A. (2020) · Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.213 · Cited by 617
Findings of the 2016 Conference on Machine Translation
Bojar, O., Chatterjee, R., Federmann, C., Graham, Y., Haddow, B., Huck, M., Jimeno Yepes, A., Koehn, P., Logacheva, V., Monz, C., et al. (2016) · Proceedings of the First Conference on Machine Translation: Volume 2, Shared Task Papers · doi:10.18653/v1/w16-2301 · Cited by 539
Achieving Human Parity on Automatic Chinese to English News Translation
Hassan, H., Aue, A., Chen, C., Chowdhary, V., Clark, J. H., Federmann, C., Huang, X., Junczys-Dowmunt, M., Lewis, W. D., Li, M., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1803.05567 · Cited by 521
Findings of the 2017 Conference on Machine Translation (WMT17)
Bojar, O., Chatterjee, R., Federmann, C., Graham, Y., Haddow, B., Huang, S., Huck, M., Koehn, P., Liu, Q., Logacheva, V., et al. (2017) · Proceedings of the Second Conference on Machine Translation · doi:10.18653/v1/w17-4717 · Cited by 395
Has Machine Translation Achieved Human Parity? A Case for Document-level Evaluation
Läubli, S., Sennrich, R., & Volk, M. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1512 · Cited by 283 · cited 1×
“Recent papers have argued that poor human evaluation practices have led to misleading results, including erroneous claims that MT has achieved human parity (Toral, 2020; Läubli et al., 2018).”
BackgroundManual and automatic evaluation of machine translation between European languages
Koehn, P., & Monz, C. (2006) · Proceedings of the Workshop on Statistical Machine Translation - StatMT '06 · doi:10.3115/1654650.1654666 · Cited by 269 · cited 1×
“The first WMT evaluation campaign (Koehn and Monz, 2006) used adequacy and fluency ratings on a 5-point scale acquired from participants as their main metric.”
MethodLommel, A., Uszkoreit, H., & Burchardt, A. (2014) · Tradumàtica tecnologies de la traducció · doi:10.5565/rev/tradumatica.77 · Cited by 233 · cited 1×
“eu) to address the shortcomings of previous quality evaluation methods (Lommel et al., 2014).”
BackgroundBLEU might be Guilty but References are not Innocent
Freitag, M., Grangier, D., & Caswell, I. R. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.5 · Cited by 201
Self-citationAttaining the Unattainable? Reassessing Claims of Human Parity in Neural Machine Translation
Toral, A., Castilho, S., Hu, K., & Way, A. (2018) · Proceedings of the Third Conference on Machine Translation: Research Papers · doi:10.18653/v1/w18-6312 · Cited by 195 · cited 1×
“The impact of translationese (Koppel and Ordan, 2011) on human evaluation of MT has recently received attention (Toral et al., 2018; Zhang and Toral, 2019; Freitag et al., 2019; Graham et al., 2020).”
BackgroundResults of the WMT20 Metrics Shared Task
Mathur, N., Wei, J. T.-Z., Freitag, M., Ma, Q., & Bojar, O. (2020) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2020.wmt-1.77 · Cited by 180
Self-citationFindings of the 2020 Conference on Machine Translation (WMT20)
Barrault, L., Biesialska, M., Bojar, O., Costa-jussà, M. R., Federmann, C., Graham, Y., Grundkiewicz, R., Haddow, B., Huck, M., Joanis, E., et al. (2020) · Proceedings of the Fifth Conference on Machine Translation · doi:10.18653/v1/2020.wmt-1.1 · Cited by 178
Can machine translation systems be evaluated by the crowd alone
Graham, Y., Baldwin, T., Moffat, A., & Zobel, J. (2015) · Natural Language Engineering · doi:10.1017/s1351324915000339 · Cited by 156
The ARPA MT Evaluation Methodologies: Evolution, Lessons, and Future Approaches
White, J. S., O'Connell, T. A., & O'Mara, F. E. (1994) · Conference of the Association for Machine Translation in the Americas · Cited by 156
Translationese and Its Dialects
Koppel, M., & Ordan, N. (2011) · Cited by 148
Comparing Bayesian Models of Annotation
Paun, S., Carpenter, B., Chamberlain, J., Hovy, D., Kruschwitz, U., & Poesio, M. (2018) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00040 · Cited by 147 · cited 1×
“Differences in error rates among raters are not severe but could be reduced with corrections from an annotation model (Paun et al., 2018), especially when working with larger anno-tator pools.”
BackgroundContinuous Measurement Scales in Human Evaluation of Machine Translation
Graham, Y., Baldwin, T., Moffat, A., & Zobel, J. (2013) · Trinity's Access to Research Output (TARA) (Trinity College Dublin) · Cited by 146
Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse
Banarescu, L., Bonial, C., Cai, S., Georgescu, M., Griffitt, K., Hermjakob, U., Knight, K. K., Koehn, P., Palmer, M. S., & Schneider, N. (2013) · The Association for Computational Linguistics · Cited by 133
A Set of Recommendations for Assessing Human–Machine Parity in Language Translation
Läubli, S., Castilho, S., Neubig, G., Sennrich, R., Shen, Q., & Toral, A. (2020) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.11371 · Cited by 130
Statistical Power and Translationese in Machine Translation Evaluation
Graham, Y., Haddow, B., & Koehn, P. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.6 · Cited by 89
APE at Scale and Its Implications on MT Evaluation Biases
Freitag, M., Caswell, I. R., & Roy, S. (2019) · Conference on Machine Translation (WMT) · doi:10.18653/v1/w19-5204 · Cited by 85 · cited 1×
“The impact of translationese (Koppel and Ordan, 2011) on human evaluation of MT has recently received attention (Toral et al., 2018; Zhang and Toral, 2019; Freitag et al., 2019; Graham et al., 2020).”
BackgroundSelf-citationThe Effect of Translationese in Machine Translation Test Sets
Zhang, M., & Toral, A. (2019) · Conference on Machine Translation (WMT) · doi:10.18653/v1/w19-5208 · Cited by 84 · cited 1×
“The impact of translationese (Koppel and Ordan, 2011) on human evaluation of MT has recently received attention (Toral et al., 2018; Zhang and Toral, 2019; Freitag et al., 2019; Graham et al., 2020).”
BackgroundTranslationese in Machine Translation EvaluationPreprint
Graham, Y., Haddow, B., & Koehn, P. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.09833 · Cited by 69 · cited 1×
“The impact of translationese (Koppel and Ordan, 2011) on human evaluation of MT has recently received attention (Toral et al., 2018; Zhang and Toral, 2019; Freitag et al., 2019; Graham et al., 2020).”
BackgroundA Comparative Quality Evaluation of PBSMT and NMT using Professional Translators
Castilho, S., Moorkens, J., Gaspari, F., Sennrich, R., Sosoni, V., Georgakopoulou, Y., Lohar, P., Way, A., Barone, A. M., & Gialama, M. (2017) · Edinburgh Research Explorer (University of Edinburgh) · Cited by 58
Klubička, F., Toral, A., & Sánchez-Cartagena, V. M. (2018) · Machine Translation · doi:10.1007/s10590-018-9214-x · Cited by 54 · cited 1×
“Klubiˇcka et al. (2018) designed an MQM-compliant error taxonomy tailored to the relevant linguistic phenomena of Slavic languages to run a case study for 3 MT systems for English → Croatian.”
BackgroundA Set of Recommendations for Assessing Human–Machine Parity in Language Translation
Läubli, S., Castilho, S., Neubig, G., Sennrich, R., Shen, Q., & Toral, A. (2020) · Zurich Open Repository and Archive (University of Zurich) · doi:10.5167/uzh-188217 · Cited by 54
Reassessing Claims of Human Parity and Super-Human Performance in Machine Translation at WMT 2019Preprint
Toral, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.05738 · Cited by 44 · cited 1×
“Recent papers have argued that poor human evaluation practices have led to misleading results, including erroneous claims that MT has achieved human parity (Toral, 2020; Läubli et al., 2018).”
BackgroundInformative Manual Evaluation of Machine Translation Output
Popović, M. (2020) · International Conference on Computational Linguistics · doi:10.18653/v1/2020.coling-main.444 · Cited by 41 · cited 1×
“Popovi´c (2020) proposed a new method for manual evaluation based on marking actual issues in the translated text.”
MethodA Gold Standard Methodology for Evaluating Accuracy in Data-To-Text Systems
Thomson, C., & Reiter, E. (2020) · Proceedings of the 13th International Conference on Natural Language Generation · doi:10.18653/v1/2020.inlg-1.22 · Cited by 38
Proceedings of the Third Workshop on Statistical Machine Translation
Callison-Burch, C., & Association for Computing Machinery-Digital Library (2008) · Cited by 36
Human evaluation of machine translation through binary system comparisons
Vilar, D., Leusch, G., Ney, H., & Banchs, R. E. (2007) · Conference on Machine Translation (WMT) · doi:10.3115/1626355.1626368 · Cited by 33
Forcada, M. L., Scarton, C., Specia, L., Haddow, B., & Birch, A. (2018) · Conference on Machine Translation (WMT) · doi:10.18653/v1/w18-6320 · Cited by 21 · cited 1×
“Forcada et al. (2018) proposed gap-filling, where certain words are removed from reference translations and readers are asked to fill the gaps left using the machine-translated text as a hint.”
MethodInvolving language professionals in the evaluation of machine translation
Popović, M., Avramidis, E., Burchardt, A., Hunsicker, S., Schmeier, S., Tscherwinka, C., Vilar, D., & Uszkoreit, H. (2014) · Language Resources and Evaluation · doi:10.1007/s10579-014-9286-z · Cited by 19 · cited 1×
“Avramidis et al. (2012) used professional trans-lators to rate MT output on three different tasks: ranking, error classification and post-editing.”
Bentivogli, L., Cettolo, M., Federico, M., & Federmann, C. (2018) · Cited by 18
A Reading Comprehension Corpus for Machine Translation Evaluation
Scarton, C., & Specia, L. (2016) · Language Resources and Evaluation Conference (LREC) · doi:10.63317/4zt8wddg63ct · Cited by 9
The Role of human reference translation in machine translation evaluation
Fomicheva, M. (2017) · Repositori digital de la UPF (Universitat Pompeu Fabra) · Cited by 4
Fischer, L., & Läubli, S. (2020) · Zurich Open Repository and Archive (University of Zurich) · doi:10.48550/arxiv.2006.04781 · Cited by 2
Research works printed but not linked: 1
- ALPAC, 1966, Language and Machines: Computers in Translation and Linguistics; a Report
Cited by
Anil, R., Dai, A. M., Firat, O., Johnson, M., Lepikhin, D., Passos, A., Shakeri, S., Taropa, E., Bailey, P., Chen, Z., et al. (2023) · arXiv · doi:10.48550/arxiv.2305.10403 · Cited by 834
“MQM represents the average errors per segment, with lower numbers indicating better results.”
MethodInfluentialSelf-citationSurvey of Low-Resource Machine Translation
Haddow, B., Bawden, R., Barone, A. V. M., Helcl, J., & Birch, A. (2022) · Computational Linguistics · doi:10.1162/coli_a_00446 · Cited by 352
One Embedder, Any Task: Instruction-Finetuned Text Embeddings
Su, H., Shi, W., Kasai, J., Wang, Y., Hu, Y., Ostendorf, M., Yih, W.-T., Smith, N. A., Zettlemoyer, L., & Yu, T. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.71 · Cited by 404
Large Language Models Are State-of-the-Art Evaluators of Translation Quality
Kocmi, T., & Federmann, C. (2023) · arXiv · doi:10.48550/arxiv.2302.14520 · Cited by 122
“(2020) and Freitag et al. (2021b) changed this back to the 2011 version.”
BackgroundMethodInfluentialRepairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Text
Gehrmann, S., Clark, E., & Sellam, T. (2023) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.13715 · Cited by 123
Metadata
- Authors
- Freitag, M., Foster, G., Grangier, D., Ratnakar, V., Tan, Q., & Macherey, W.
- Affiliations
- Google (United States) (US)
- Corresponding author
- Freitag, Markus
- Foster, George
- Grangier, David
- Ratnakar, Viresh
- Tan, Qijun
- Macherey, Wolfgang
- ORCID
- Grangier, David
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2021
- Volume
- 9
- Pages
- 1460-1474
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W3159892921
- Cited by
- 271
- References
- 43
- Citations by year
- 2026: 45 · 2025: 48 · 2024: 36 · 2023: 73 · 2022: 45 · 2021: 24
- FWCI
- 15.59
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Transactions of the Association for Computational Linguistics (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Social Sciences › Arts and Humanities › Language and Linguistics › Translation Studies and Practices
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- human evaluation of machine translation, Multidimensional Quality Metrics, error analysis, evaluation methodology, professional translators, crowd workers, automatic evaluation metrics, pre-trained embeddings, human-machine comparison
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.