Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation
Open accessThe DOI matches its Crossref record; 38 of the 43 references it cites were found in a registry or an index.
Abstract
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
Continuous 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 353
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).”
Background
Research works printed but not linked: 1
- ALPAC, 1966, Language and Machines: Computers in Translation and Linguistics; a Report
Cited by
Items 101–200 of 280
Document Flattening: Beyond Concatenating Context for Document-Level Neural Machine Translation
Wu, M., Foster, G. F., Qu, L., & Haffari, G. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.33 · Cited by 8
Self-citationMetricX-23: The Google Submission to the WMT 2023 Metrics Shared Task
Juraska, J., Finkelstein, M., Deutsch, D., Siddhant, A., Mirzazadeh, M., & Freitag, M. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.63 · Cited by 147
“The two data sources that are primarily used to train and meta-evaluate MT metrics are the direct assessment (DA) data and Multi-dimensional Quality Metrics (MQM; Lommel et al., 2014; Freitag et al., 2021) that have been collected by WMT over the years, and both of which are publicly available.”
BackgroundMethodSelf-citationFreitag, M., Ghorbani, B., & Fernandes, P. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.617 · Cited by 82
Self-citationTOWARDS BUILDING A TAXONOMY OF MACHINE TRANSLATION ERRORS: VERSATILITY VS TARGETED APPROACHES
Grebenyuk, A. V. (2024) · Lomonosov Translation Studies Journal · doi:10.55959/msu2074-6636-22-2024-17-2-7-25 · Cited by 7
Collaborative Ontology Matching With Dual Population Genetic Programming and Active Meta-Learning
Xue, X., Lin, J. C., & Jiang, Z. (2025) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2025.3569336 · Cited by 11
“However, the reliability of collaborative validation is debatable, primarily because it cannot distinguish between “genuine” errors due to user inexperience and “false” errors arising from expert disagreements [14].”
Measuring Uncertainty in Translation Quality Evaluation (TQE)
Gladkoff, S., Sorokina, I., Han, L., & Alekseeva, A. (2022) · Language Resources and Evaluation Conference (LREC) · doi:10.63317/5oye6bgfxmey · Cited by 9
Toward More Effective Human Evaluation for Machine Translation
Saldías-Fuentes, B., Foster, G. F., Freitag, M., & Tan, Q. (2022) · doi:10.18653/v1/2022.humeval-1.7 · Cited by 9
Self-citationOn nature and causes of observed MT errors.
Popović, M. (2021) · Dublin City University Open Access Institutional Repository (Dublin City University) · Cited by 7
Arabic and English Relative Clauses and Machine Translation Challenges
Nagi, K. A. S. K. (2023) · Mağallaẗ al-dirāsāt al-iğtimāʿiyyaẗ · doi:10.20428/jss.v29i3.2180 · Cited by 7
Training and Meta-Evaluating Machine Translation Evaluation Metrics at the Paragraph Level
Deutsch, D., Juraska, J., Finkelstein, M., & Freitag, M. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.96 · Cited by 31
Self-citationDelgado, D., Burgueño, L., & Clarisó, R. (2026) · Journal of Systems and Software · doi:10.1016/j.jss.2026.112871 · Cited by 10
“It should be noted that in complex AI / ML tasks such as machine translation or code generation, some level of disagreement is expected even among human experts [37].”
Vayá, A. R., & Enamorado, M. Á. J. (2022) · Tradumàtica tecnologies de la traducció · doi:10.5565/rev/tradumatica.306 · Cited by 6
DivEMT: Neural Machine Translation Post-Editing Effort Across Typologically Diverse Languages
Sarti, G., Bisazza, A., Arenas, A. G., & Toral, A. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.532 · Cited by 31
“…as WMT (Koehn and Monz, 2006; Akhbardeh et al., 2021) and IWSLT (Cettolo et al., 2016, 2017) among others, but their focus is on language-and domain-specific ranking of MT systems — often leveraging non-professional anno-tators (Freitag et al., 2021) — rather than cross-lingual quality comparisons.”
BackgroundLessons from the Trenches on Reproducible Evaluation of Language Models
Biderman, S., Schoelkopf, H., Sutawika, L., Gao, L., Tow, J., Abbasi, B., Aji, A. F., Ammanamanchi, P. S., Black, S., Clive, J., et al. (2024) · arXiv · doi:10.48550/arxiv.2405.14782 · Cited by 9
“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).”
Jung, D., Park, C., Eo, S., & Lim, H. (2023) · Mathematics · doi:10.3390/math11194169 · Cited by 7
“During the fine-tuning, we use the expert-based human evaluation datasets for the submissions of WMT 2020, 2021, and 2022 for En-De and Zh-En [25].”
Learning Evaluation Models From Large Language Models for Sequence Generation
Wang, C., Zhou, H., Chang, K., Liu, T., Zhang, C., Yang, M., Du, Q., Xiao, T., Zhang, Y., & Zhu, J. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3587460 · Cited by 12
“Inspired by [77], we hypothesize that increasing the number of output sequences might improve the performance of the evaluation model learned by CSEM.”
Onception: Active Learning with Expert Advice for Real World Machine Translation
Mendonça, V., Rei, R., Coheur, L., & Sardinha, A. (2022) · Computational Linguistics · doi:10.1162/coli_a_00473 · Cited by 9
SESCORE2: Learning Text Generation Evaluation via Synthesizing Realistic Mistakes
Xu, W., Qian, X., Wang, M., Li, L., & Wang, W. Y. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.283 · Cited by 22
“We are inspired by the human evaluation protocol MQM (Freitag et al., 2021a) which assesses transla-a tion quality by identifying errors with two levels of severity.”
BackgroundMethodResultInfluentialQuality Estimation Using Minimum Bayes Risk
Naskar, S., Deutsch, D., & Freitag, M. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.67 · Cited by 5
“It is trained on a combination of direct assessment and MQM (Lommel et al., 2014; Freitag et al., 2021a) data that was collected by WMT.”
MethodSelf-citationMendonça, V., Rei, R., Coheur, L., Sardinha, A., & Santos, A. L. (2021) · doi:10.18653/v1/2021.acl-long.242 · Cited by 5
“Moreover, this metric correlates better with ratings by professional translators than the WMT scores (Freitag et al., 2021).”
BackgroundBulut, S. Ö. (2021) · transLogos Translation Studies Journal · doi:10.29228/translogos.33 · Cited by 5
Are Experts Needed? On Human Evaluation of Counselling Reflection Generation
Wu, Z., Balloccu, S., Reiter, E., Helaoui, R., Recupero, D. R., & Riboni, D. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.382 · Cited by 22
“Some human evaluation studies have compared expert and non-expert NLG evaluations, such as for summarisation (Gillick and Liu, 2010; Fabbri et al., 2021), machine translation (Freitag et al., 2021), story generation (Karpinska et al., 2021) and others (e.g., Snow et al., 2008).”
BackgroundDisentangling Uncertainty in Machine Translation Evaluation
Zerva, C., Glushkova, T., Rei, R., & Martins, A. F. T. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.591 · Cited by 18
“In recent studies it has become increasingly evident that supervised metrics exhibit higher correlations with human judgements (Mathur et al., 2020; Freitag et al., 2021a) and produce more reliable assessments of MT quality (Kocmi et al., 2021).”
BackgroundJansen, P. A., Smith, K. J., Moreno, D., & Ortiz, H. (2021) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2021.emnlp-main.596 · Cited by 14
“This parallels the disparity in automatic versus manual evaluations in other fields, such as machine translation (Freitag et al., 2021).”
ResultRegressive Ensemble for Machine Translation Quality EvaluationPreprint
Štefánik, M., Novotný, V. S., & Sojka, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.07242 · Cited by 9
QE4PE: Word-level Quality Estimation for Human Post-Editing
Sarti, G., Zouhar, V., Chrupała, G., Arenas, A. G., Nissim, M., & Bisazza, A. (2025) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl.a.46 · Cited by 5
“Using multiple severity levels follows the current MT evaluation practices (Freitag et al., 2021, 2024), and previous results showing that users tend to prefer more informative word-level highlights (Shenoy et al., 2021; Vasconcelos et al., 2024).”
Alzain, E., Nagi, K. A. S. K., & Algobaei, F. (2024) · Forum for Linguistic Studies · doi:10.30564/fls.v6i3.6799 · Cited by 26
“The study employed the SQM which uses a 0-6 Likert-like scale, whose rankings are as follows.”
BackgroundMethodInfluentialHuman-centered pedagogies in the age of generative AI
Ayvazyan, N. (2025) · INContext Studies in Translation and Interculturalism · doi:10.54754/incontext.v5i1.114 · Cited by 4
The DO-KB knowledgebase 2026 update: expanding programmatic and language access
Baron, J. A., Johnson, C. M. S.-B., Schor, M. A., Olley, D., Nickel, L., Felix, V., Bello, S. M., Greene, C. L., Lichenstein, R., Bisordi, K., et al. (2025) · Nucleic Acids Research · doi:10.1093/nar/gkaf1213 · Cited by 9
“…short order and provide verified, high-quality translations, we devised a multistep protocol for translation, beginning with professional translation, which has been found to be less error prone than machine translation but not free from error [ 21 ], followed by semi-automated and manual review.”
Lou, L., Yin, X., Xie, Y., & Xiang, Y. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.682 · Cited by 4
“It is very costly to do MQM/DA annotation of different systems’ performance on CCEval , since this must be done by senior linguistic experts mastering both Chinese and our 11 languages.”
InfluentialInitial exploration into sarcasm and irony through machine translation
Chia, Z. L., Ptaszynski, M. E., Karpinska, M., Eronen, J., & Masui, F. (2024) · Natural Language Processing Journal · doi:10.1016/j.nlp.2024.100106 · Cited by 15
Multilingual Document-Level Translation Enables Zero-Shot Transfer From Sentences to Documents
Zhang, B., Bapna, A., Johnson, M., Dabirmoghaddam, A., Arivazhagan, N., & Fırat, O. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.287 · Cited by 16
Li, Z., Qu, L., Cohen, P. R., Tumuluri, R., & Haffari, G. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.529 · Cited by 5
Near-Negative Distinction: Giving a Second Life to Human Evaluation Datasets
Laban, P., Wu, C.-S., Liu, W., & Xiong, C. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.135 · Cited by 5
eBLEU: Unexpectedly Good Machine Translation Evaluation Using Simple Word Embeddings
ElNokrashy, M., & Kocmi, T. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.61 · Cited by 4
“Human labels are produced via three methods: • MQM - annotated by professionals who mark individual errors in each translation, as described in (Freitag et al., 2021) • DA+SQM - professional annotators are asked to rate each translation on a scale 0-100 (Kocmi et al., 2022) • MTurk DA - low paid…”
MethodUDAAN - Machine Learning based Post-Editing tool for Document Translation
Maheshwari, A., Ravindran, A., Subramanian, V., & Ramakrishnan, G. (2023) · Joint International Conference on Data Science & Management of Data (CODS-COMAD) · doi:10.1145/3570991.3571068 · Cited by 4
Calibration and context in human evaluation of machine translation
Knowles, R., & Lo, C.-K. (2024) · Natural language processing. · doi:10.1017/nlp.2024.5 · Cited by 4
Zhao, J., Li, X., & Liao, H. (2025) · Language Teaching Research · doi:10.1177/13621688251352261 · Cited by 7
Ashirova, A. I., Khurramov, N. K., Ravikumar, R. N., Aarthi, S., & Sindhuja, S. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-4202-3.ch008 · Cited by 3
Ramadhan, T. I., Ramadhan, N. G., & Supriatman, A. (2022) · Building of Informatics Technology and Science (BITS) · doi:10.47065/bits.v4i3.2614 · Cited by 3
“The manual approach is simply using human experts in the language field to evaluate the result of the model [16].”
MethodAssessment of Computer-Aided Translation Quality Based on Large-Scale Corpora
Xu, H. (2024) · International Journal of e-Collaboration · doi:10.4018/ijec.357274 · Cited by 3
MT Use by professional translators in Poland – survey results
Paradowska, U., Sycz-Opoń, J., & Bądziński, A. (2025) · Perspectives · doi:10.1080/0907676x.2025.2502937 · Cited by 3
Siu, S. C. (2022) · Corpora and intercultural studies · doi:10.1007/978-981-19-6680-4_5 · Cited by 3
Thresh: A Unified, Customizable and Deployable Platform for Fine-Grained Text Evaluation
Heineman, D., Dou, Y., & Xu, W. H. (2023) · Conference on Empirical Methods in Natural Language Processing: System Demonstrations (EMNLP) · doi:10.18653/v1/2023.emnlp-demo.30 · Cited by 13
Fine Tuning Language Models: A Tale of Two Low-Resource Languages
Oida-Onesa, R., & Ballera, M. A. (2024) · Data Intelligence · doi:10.3724/2096-7004.di.2024.0016 · Cited by 3
Yan, Y., Song, J., Fu, B., Ye, N., & Shi, X. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-5672-8_26 · Cited by 3
Automatic Correction of Human Translations
Lin, J., Kovacs, G., Shastry, A., Wuebker, J., & DeNero, J. (2022) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2022.naacl-main.36 · Cited by 3
Ogrodniczuk, M., & Kryńska, K. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-21756-2_34 · Cited by 3
On reporting scores and agreement for error annotation tasks
Popović, M., & Belz, A. (2022) · Workshop on Generation, Evaluation and Metrics (GEM) · doi:10.18653/v1/2022.gem-1.26 · Cited by 3
“Freitag et al. (2021) apply the MQM error scheme on large amounts of texts for two language pairs.”
MethodOn the Effectiveness of Automated Metrics for Text Generation Systems
von Däniken, P., Deriu, J., Tuggener, D., & Cieliebak, M. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.108 · Cited by 10
Han, L., Sorokina, I., Erofeev, G., & Gladkoff, S. (2021) · Research Explorer (The University of Manchester) · doi:10.48550/arxiv.2108.09484 · Cited by 7
“We refer to the paper (Freitag et al., 2021) for detailed MT system names and offering institutions.”
BackgroundMethodInfluentialConstruction and Evaluation of Context Aware Machine Translation System
Jiang, C., & He, Y. P. (2025) · Procedia Computer Science · doi:10.1016/j.procs.2025.04.242 · Cited by 2
The impact of MT as a writing tool on EFL academic writing: A qualitative linguistic analysis
Comelles, E., & Laso, N. J. (2025) · Journal of Second Language Writing · doi:10.1016/j.jslw.2025.101231 · Cited by 6
Munková, D., Benková, L., Munk, M., Benko, Ľ., & Hájek, P. (2025) · Machine Learning with Applications · doi:10.1016/j.mlwa.2025.100810 · Cited by 3
Shalawati, S., Nasution, A. H., Monika, W., Derin, T., Onan, A., & Murakami, Y. (2026) · Digital · doi:10.3390/digital6010008 · Cited by 7
Yang, W., & Yang, M. (2026) · Humanities and Social Sciences Communications · doi:10.1057/s41599-026-06868-y · Cited by 2
Reward Gaming in Conditional Text Generation
Pang, R. Y., Padmakumar, V., Sellam, T., Parikh, A. P., & He, H. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.262 · Cited by 36
The Ethics of Machine Translation in The Era of Artificial Intelligence
Wen, P. P. (2024) · Advances in Social Science, Education and Humanities Research/Advances in social science, education and humanities research · doi:10.2991/978-2-38476-263-7_18 · Cited by 2
Gender Issues between Gemini and ChatGPT: The Case of English-Arabic Translation
Algobaei, F., Alzain, E., Naji, E., & Nagi, K. A. S. K. (2024) · World Journal of English Language · doi:10.5430/wjel.v15n1p9 · Cited by 4
Document-Level Machine Translation with Effective Batch-Level Context Representation
Zhong, K., Zhang, J., & Guo, W. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10651489 · Cited by 2
“It was recently shown in [4], [5] that NMT models still cannot perform as well as human translations at the document level.”
BackgroundDiBiMT : A Gold Evaluation Benchmark for Studying Lexical Ambiguity in Machine Translation
Martelli, F., Perrella, S., Campolungo, N., Munda, T., Koeva, S., Tiberius, C., & Navigli, R. (2024) · Computational Linguistics · doi:10.1162/coli_a_00541 · Cited by 9
Natural Language Processing RELIES on Linguistics
Opitz, J., Wein, S., & Schneider, N. (2025) · Computational Linguistics · doi:10.1162/coli_a_00560 · Cited by 10
“Gold standard evaluation.”
InfluentialClimate risk and low-carbon policies: implications for sports economics and global events
Fu, S.-Y., & Kong, C. (2025) · Frontiers in Environmental Science · doi:10.3389/fenvs.2025.1578634 · Cited by 3
Error Analysis of ToTTo Table-to-Text Neural NLG Models
Sundararajan, B., Sripada, S. G., & Reiter, E. (2022) · Workshop on Generation, Evaluation and Metrics (GEM) · doi:10.18653/v1/2022.gem-1.43 · Cited by 5
“…for error analysis predominantly focused on Machine Translation (MT) systems using a simple framework by (Stymne and Ahrenberg, 2012) and extensively using Multidimensional Quality Metrics (MQM) framework by Freitag et al. (2021), and on other NLG tasks (Cai et al., 2020; Thomson and Reiter, 2020).”
MethodStudent’s t-Distribution: On Measuring the Inter-Rater Reliability When the Observations are Scarce
Logrus Global, T. &. L., Gladkoff, S., Han, L., & Nenadić, G. (2023) · International conference Recent advances in natural language processing · doi:10.26615/978-954-452-092-2_047 · Cited by 3
IOL Research’s Submission for WMT 2023 Quality Estimation Shared Task
Yan, Z. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.76 · Cited by 2
“On account of the task similarity to QE, we also collect the MQM data (Freitag et al., 2021a,b) from previous WMT Metrics tasks 1 and APE data from QT21 (Specia et al., 2017) and APE-QUEST (Depraetere et al., 2020) to do further pretraining.”
MethodNorth Korean Neural Machine Translation through South Korean Resources
Kim, H., Hirasawa, T., Moon, S., Okazaki, N., & Komachi, M. (2023) · ACM Transactions on Asian and Low-Resource Language Information Processing · doi:10.1145/3608947 · Cited by 3
Rethinking the Word-level Quality Estimation for Machine Translation from Human Judgement
Yang, Z., Meng, F.-D., Yan, Y., & Zhou, J. (2023) · Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2023.findings-acl.126 · Cited by 7
“In the recent WMT22 word-level QE shared task, several language pairs began to use MQM-based annotation introduced from the Metrics shared task as the quality estimation (Freitag et al., 2021a,c).”
MethodBetter than Random: Reliable NLG Human Evaluation with Constrained Active Sampling
Ruan, J., Pu, X., Gao, M., Wan, X., & Zhu, Y. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i17.29857 · Cited by 14
The statistical advantage of automatic NLG metrics at the system levelPreprint
Wei, J. T.-Z., & Jia, R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.12437 · Cited by 2
Reward Gaming in Conditional Text Generation
Pang, R. Y., Padmakumar, V., Sellam, T., Parikh, A. P., & He, H. (2022) · arXiv · doi:10.48550/arxiv.2211.08714 · Cited by 2
“The WMT MQM dataset (Freitag et al., 2021a) is a high-quality human annotation dataset on translations, where each Zh-En translation is annotated with ≤ 5 most serious error spans by expert annotators according to the MQM metric (Lommel et al., 2014).”
MethodZhang, Y., & Zhang, Y. (2026) · Proceedings of the 3rd International Conference on Educational Digitalization and Computer Science in the Guangdong-Hong Kong-Macao Greater Bay Area · doi:10.1145/3828183.3828226 · Cited by 1
“While COMETgenerallyalignsbetter withhumanjudgementthanBLEU[2], it is not a panacea.”
Improving BERTScore via Word Similarity Matrix Iterative Alignment Method
Meng, D., & Wang, L. (2023) · International Conference on Intelligent Computing and Signal Processing (ICSP) · doi:10.1109/icsp58490.2023.10248840 · Cited by 1
“Datasets and Evaluation Metrics for Algorithms In the experiments, the MQM dataset [12] was used to validate the improved BERTScore algorithm.”
MethodChen, X., Liu, Z., Ye, C., Guo, T., Song, D., & Gupta, A. C. (2023) · Proceedings of The World Conference on Social Sciences · doi:10.33422/worldcss.v2i1.101 · Cited by 1
Berro, A., Benatallah, B., Gaci, Y., & Benabdeslem, K. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-70341-6_20 · Cited by 2
“While quality control for text generated by NLG systems has been explored in a wide range of tasks, like machine translation (MT) [13], question generation [37], and open-ended generation with pre-trained language models [11], TPG has not been subjected to such scrutiny.”
Likelihood-based Mitigation of Evaluation Bias in Large Language Models
Ohi, M., Kaneko, M., Koike, R., Loem, M., & Okazaki, N. (2025) · Journal of Natural Language Processing · doi:10.5715/jnlp.32.480 · Cited by 1
Niu, S., Li, W.-H., & Lili (2025) · Frontiers in Public Health · doi:10.3389/fpubh.2025.1597381
“However, these methods faced significant challenges in dealing with the complexity and variability of real-world economic and health data (9).”
Strandvik, I. (2025) · Across Languages and Cultures · doi:10.1556/084.2025.01057 · Cited by 1
“The resulting hype – rarely but properly dispelled (see e.g. Freitag et al., 2021; Gladkoff, 2021, 2022; Läubli et al., 2018; Lommel, 2020; Melby & Kurz, 2021; Toral, 2020) – is having a dramatic impact on client expectations on savings.”
Cross-Lingual and Cross-Domain Evaluation of ChatGPT’s Translation Performance
Gašpar, A., & Knežević, M. (2025) · Communications in computer and information science · doi:10.1007/978-3-032-02801-3_13 · Cited by 4
Empowerment-AI-Enhanced System for Initiating and Sustaining Vocabulary Learning in EFL Learners
Oeda, S. (2025) · Procedia Computer Science · doi:10.1016/j.procs.2025.09.449 · Cited by 1
Карпіна, О., & Засєкін, С. В. (2025) · East European Journal of Psycholinguistics · doi:10.29038/kar · Cited by 1
Implementation of a Large Language Model (LLM) for Spanish-Nahuatl Translation
Lucero, E. S., Hernandez, J. J., Huerta, E. B., & Zistecatl, S. P. (2025) · Mexican International Conference on Computer Science · doi:10.1109/enc68268.2025.11311863 · Cited by 1
“That is why [7] conducts a large-scale meta-study on human evaluation of translations, crucial for validating the results of Transformer-based models, and finds high variability in human evaluations and underlines the importance of rigorous experimental designs.”
Brivá-Iglesias, V., & O’Brien, S. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2622588 · Cited by 11
“…the fact that there is no such thing as one agreed translation for any one sentence, that meaning is communicated also at a textual level and is not restricted to sentence level, is context-dependent, and the scores provided are meaningless to the general user (Freitag et al., 2021; Kenny, 2022).”
Enhancing Post-Editing of Kazakh Translations Using Fine-Tuned Large Language Models
Bekarystankyzy, A., Rakhimova, D., Zhiger, A., Sakatay, A., Zhumakhan, N., Yerimbetova, A., Оралбекова, Д., & Turdalyuly, M. (2026) · Algorithms · doi:10.3390/a19030199 · Cited by 2
“Freitag et al. [42] provide valuable guidance on best practices for human MT evaluation, but most studies focus on high-resource language pairs.”
Wang, F. (2026) · Journal of improved oil and gas recovery technology. · doi:10.53469/wjimt.2026.09(03).01 · Cited by 1
“In order to promote standardization, Freitag et al. proposed an evaluation method based on explicit error analysis under the framework of multidimensional quality measurement [2].”
Li, J., Zhu, S., & Xiong, D. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.104819 · Cited by 1
Beyond the Algorithm: Navigating Ethics, Quality, and Accountability in AI-Powered Translation
Aqad, M. H. A., Amirdabbaghian, A., & Alhasani, H. M. A. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-6420-9.ch010 · Cited by 1
Marchisio, K., Freitag, M., & Grangier, D. (2022) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2022.naacl-main.161 · Cited by 4
Self-citationInteresting Near-boundary Data: Inferring Model Ownership for DNNs
Sun, Z., Yang, Z., Huang, Z., Zhang, Y., & Zhang, J. (2023) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn54540.2023.10191063 · Cited by 1
“However, most models are often exposed to the public to provide services such as machine translation[8] or image recognition[14], which allows attackers to steal models through exposed interfaces[4].”
BackgroundNguyen, L., Bryant, C., Mayeux, O., & Yuan, Z. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.893 · Cited by 4
EvolveMT: an Ensemble MT Engine Improving Itself with Usage Only
Yüksel, K. A., Gündüz, A., Al-Badrashiny, M. A. S. A. A., & Sawaf, H. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-industry.33 · Cited by 1
Twist Decoding: Diverse Generators Guide Each Other
Kasai, J., Sakaguchi, K., Bras, R. J. L., Peng, H., Lu, X., Radev, D., Choi, Y., & Smith, N. A. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.326 · Cited by 4
Quality Scoring of Source Words in Neural Translation Models
Jain, P., Sarawagi, S., & Tomar, T. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.732 · Cited by 1
Trained MT Metrics Learn to Cope with Machine-translated References
Vamvas, J., Domhan, T., Trenous, S., Sennrich, R., & Hasler, E. (2023) · Conference on Machine Translation (WMT) · doi:10.18653/v1/2023.wmt-1.95 · Cited by 1
Evaluating optimal reference translations
Zouhar, V., Kloudová, V., Popel, M., & Bojar, O. (2024) · Natural language processing. · doi:10.1017/nlp.2024.3 · Cited by 1
Evaluation of deep neural network architectures for authorship obfuscation of Portuguese texts
Franco, A. M. R., Cunha, Í. F. S., & Oliveira, L. B. (2024) · Natural Language Processing Journal · doi:10.1016/j.nlp.2024.100107 · Cited by 2
He, Y. (2024) · Applied Mathematics and Nonlinear Sciences · doi:10.2478/amns-2024-3220 · Cited by 1
Deep Learning Techniques for Machine Translation: A Survey
Hameed, D. A., & Al‐Khateeb, B. (2025) · Procedia Computer Science · doi:10.1016/j.procs.2025.04.339 · Cited by 3
Multi-Corpus Alignment and Cross-Language Semantic Analysis for Chinese-English Machine Translation
Wang, W. (2025) · International Conference on Inventive Computing and Informatics (ICICI) · doi:10.1109/icici65870.2025.11069758 · Cited by 1
“Freitag et al. [5] studied the manual evaluation method of machine translation systems.”
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.