Elamur Research⌕
☰
home / papers / 2024 / Cyber Risks of Machine Translation Critical Errors : Arabic Mental Health Tweets as a Case Study

Cyber Risks of Machine Translation Critical Errors : Arabic Mental Health Tweets as a Case Study

Summary

Free neural MT tools often produce fluent output that is quietly wrong. Neither readers who don't know the source language nor standard quality metrics tend to notice. The authors study this risk for Arabic tweets about mental health. Here a mistranslation can hide signs of depression or suicidal thinking from a human reader or from an English-language detection model.

They translated the AraDepSu tweet corpus into English with Google Translate. An English mental-health classifier was trained on Reddit posts and run on the translations. Tweets whose predicted label disagreed with the Arabic gold label went to bilingual annotators. The annotators marked mistranslations, singled out "critical" ones, assigned an error type and wrote a corrected English reference.

The result is an openly shared dataset of roughly 800 critical errors. Dialectal Arabic is a major cause. The authors also stress-test common MT quality metrics on this dataset and show that they fail to penalise these errors.

The work matters because MT is used routinely to bring low-resource social media text into English pipelines. Without error-aware evaluation, urgent cases may turn into false negatives.

Key points

  • Source data: AraDepSu, about 20K tweets in MSA and the Gulf, Egyptian and Levantine dialects. Each tweet is labelled depression, depression with suicidal ideation, or non-depression. All tweets were translated with Google Translate, which is also the back-end of Twitter's MT tool.
  • Filtering classifier: XLM-RoBERTa-base (written "XML-Roberta" in the paper), fine-tuned on about 150k posts from the Reddit Mental Health Dataset.
    • Labels: anxiety, bipolar, depression, non-mental.
    • Training: 4 epochs, AdamW at learning rate 1e-5, 90/10 split, about 92% validation accuracy.
  • Candidates were tweets with a gold depressive or suicidal label that the classifier predicted as "non-mental". Manual analysis found about 2200 mistranslations, of which about 800 met the critical-error definition.
  • A critical error is defined by three properties:
    • it is a dangerous departure from the intended meaning;
    • it is fluent enough to go unnoticed;
    • its repetition across a corpus can bias detectors toward false negatives.
  • Roughly a third of critical errors come from dialect. For example, the Gulf form of "want" is consistently rendered as "father", so "I want to die now" becomes "My father is now dead". The authors attribute this to out-of-vocabulary words being replaced by contextually likely ones.

Interesting ideas

  • I1. Mine translation errors by running a downstream classifier on the MT output and keeping cases where its label contradicts the source's gold label. This cheaply shrinks a large corpus to a manageable set for human review. — For Mova (learn): when silver data is produced, compare the frame or answer derived from a parse against a known label, and route only the contradictions to the LLM teacher or human repair.
  • I2. Define an error as critical by its consequence (a missed alarm, a skewed detector), not by surface distance from a reference. This ties severity to what actually matters downstream. — For Mova (back): judge the text → graph → text compressor by whether key frames survive regeneration, such as who wants what and negation, instead of by overlap.
  • I3. Stress-test evaluation metrics with a curated set of fluent-but-wrong outputs to expose metrics that reward fluency over meaning. — For Mova (learn): add meaning-flipping fluent paraphrases as negative controls, and require that any evaluator scores them low.
  • I4. Systematic dialect errors arise because the model silently swaps an unknown word for a plausible known one, and the fluency hides the swap. Flagging such substitutions instead of masking them would make the risk visible. — For Mova (tokens): when an unknown form is mapped to a guessed lemma, mark it as guessed so that later stages can propagate the uncertainty and explain it.
  • I5. Each annotated item carries three layers: an error label, an error type, and a corrected reference. This makes one dataset useful for detection, analysis and metric testing at once. — For Mova (frames): annotate frame errors the same way (wrong role, error category, corrected frame) so that repair rules can be induced from the corrections.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages · Machine translation

Topics: Machine translation

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2405.11668 · arXiv 2405.11668

@misc{saadany2024cyber,
  title = {{Cyber Risks of Machine Translation Critical Errors : Arabic Mental Health Tweets as a Case Study}},
  author = {Saadany, Hadeel and Tantawy, Ashraf and Orasan, Constantin},
  year = {2024},
  doi = {10.48550/arxiv.2405.11668},
  eprint = {2405.11668},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2405.11668}
}

References

52 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Khetam Al Sharou and Lucia Specia. 2022. A taxonomy and study of critical errors in machine translation. In Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, pages 171–180. find
  2. Haifa K Aldayel and Aqil M Azmi. 2016. Arabic tweets sentiment analysis–a hybrid scheme. Journal of Information Science, 42(6):782–797. find
  3. Samah Alhazmi, William Black, and John Mc-Naught. 2013. Arabic SentiWordNet in relation to SentiWordNet 3.0. 2180, 1266(4):1. find
  4. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116. ↗
  5. Marta R Costa-jussà, Eric Smith, Christophe Ropers, Daniel Licht, Javier Ferrando, and Carlos Escolano. 2022. Toxicity in multilingual machine translation at scale. arXiv preprint arXiv:2210.03070. ↗
  6. Salma Almouzini, Asem Alageel, et al. 2019. Detecting arabic depressed users from twitter data. Procedia Computer Science, 163:257–265. find
  7. Sariah Daouk, Mina Dailami, Suzanne Barakat, Rania Awaad, Ricardo F Muñoz, and Yan Leykin. 2023. Suicidality in the arab world: results from an online screener. Community mental health journal, pages 1–8. find
  8. Satanjeev Banerjee and Alon Lavie. 2005. ME-TEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72, Ann Arbor, Michigan. Association for Computational Linguistics. find
  9. Jaleh Delfani, Constantin Orasan, Hadeel Saadany, Ozlem Temizoz, Eleanor Taylor-Stilgoe, Diptesh Kanojia, Sabine Braun, and Barbara Schouten. 2024. Google translate error analysis for mental healthcare information: Evaluating accuracy, comprehensibility, and implications for multilingual healthcare communication. arXiv preprint arXiv:2402.04023. ↗
  10. Valerio Basile, Cristina Bosco, Elisabetta Fersini, Nozza Debora, Viviana Patti, Francisco Manuel Rangel Pardo, Paolo Rosso, Manuela Sanguinetti, et al. 2019. Semeval-2019 task 5: Multilingual detection of hate speech against immigrants and women in Twitter. In 13th International Workshop on Semantic Evaluation, pages 54–63. Association for Computational Linguistics. find our card
  11. Michael Denkowski and Alon Lavie. 2014. Meteor universal: Language specific translation evaluation for any target language. In Proceedings of the ninth workshop on statistical machine translation, pages 376–380. find our card
  12. BSI. 1973a. Natural Fibre Twines, 3rd edition. British Standards Institution, London. BS 2570. find
  13. Umberto Eco. 1990. The Limits of Interpretation. Indian University Press. find
  14. BSI. 1973b. Natural fibre twines. BS 2570, British Standards Institution, London. 3rd. edn. find
  15. Alvin Grissom et al. 2022. Rare but severe neural machine translation errors induced by minimal deletion: An empirical study on chinese and english. International Committee on Computational Linguistics. find
  16. Michael Carl and M Cristina Toledo Báez. 2019. Machine translation errors and the translation process: a study across different languages. Journal of Specialised Translation, 31:107–132. find
  17. A. Castor and L. E. Pollux. 1992. The use of user modelling to guide inference and learning. Applied Intelligence, 2(1):37–53. find
  18. J.L. Chercheur. 1994. Case-Based Reasoning, 2nd edition. Morgan Kaufman Publishers, San Mateo, CA. find
  19. N. Chomsky. 1973. Conditions on transformations. In A festschrift for Morris Halle, New York. Holt, Rinehart & Winston. find
  20. Bart Desmet and Véronique Hoste. 2013. Emotion detection in suicide notes. Expert Systems with Applications, 40(16):6351–6358. find
  21. Yinuo Guo and Junfeng Hu. 2019. Meteor++ 2.0: Adopt syntactic level paraphrase knowledge into machine translation evaluation. In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 501–506. find
  22. Mariam Hassib, Nancy Hossam, Jolie Sameh, and Marwan Torki. 2022. Aradepsu: Detecting depression and suicidal ideation in arabic tweets using transformers. In Proceedings of the The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 302–311. find
  23. Paul Gerhard Hoel. 1971a. Elementary Statistics, 3rd edition. Wiley series in probability and mathematical statistics. Wiley, New York, Chichester. ISBN 0 471 40300. find
  24. Paul Gerhard Hoel. 1971b. Elementary Statistics, 3rd edition, Wiley series in probability and mathematical statistics, pages 19–33. Wiley, New York, Chichester. ISBN 0 471 40300. find
  25. Tucker Ian. 2010. Twitter spreads regional slang. https://www.theguardian.com/science/2010/sep/ 05/tv-not-twitter-spreads-slang. ↗
  26. Marwan Jarrah and Nimer Abusalim. 2021. In favour of the low ip area in the arabic clause structure: Evidence from the vso word order in jordanian arabic. Natural Language & Linguistic Theory, 39:123–156. find
  27. Otto Jespersen. 1922. Language: Its Nature, Development, and Origin. Allen and Unwin. find
  28. Shaoxiong Ji, Shirui Pan, Xue Li, Erik Cambria, Guodong Long, and Zi Huang. 2020. Suicidal ideation detection: A review of machine learning methods and applications. IEEE Transactions on Computational Social Systems, 8(1):214–226. find
  29. Philipp Koehn and Rebecca Knowles. 2017. Six challenges for neural machine translation. ACL 2017, page 28. find
  30. Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics. find
  31. Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. In: 7th International Conference on Learning Representations (2017). Http://arxiv.org/abs/1711.05101. ↗
  32. Daniel M Low, Laurie Rumker, John Torous, Guillermo Cecchi, Satrajit S Ghosh, and Tanya Talkar. 2020. Natural language processing reveals vulnerable mental health support groups and heightened health anxiety on reddit during covid-19: Observational study. Journal of medical Internet research, 22(10):e22635. find
  33. Marianna Martindale and Marine Carpuat. 2018. Fluency Over Adequacy: A Pilot Study in Measuring User Trust in Imperfect MT. In Proceedings of the 13th Conference of the Association for Machine Translation in the Americas (Volume 1: Research Track), pages 13–25. find
  34. Marianna Martindale, Marine Carpuat, Kevin Duh, and Paul McNamee. 2019. Identifying fluently inadequate output in neural and statistical machine translation. In Proceedings of Machine Translation Summit XVII: Research Track, pages 233–243. find
  35. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318. ↗ our card
  36. Ricardo Rei, Nuno M Guerreiro, Marcos Treviso, Luisa Coheur, Alon Lavie, and André FT Martins. 2023. The inside story: Towards better understanding of machine translation neural evaluation metrics. arXiv preprint arXiv:2305.11806. ↗
  37. Hadeel Saadany, Constantin Orasan, Rocio Caro Quintana, Felix do Carmo, and Leonardo Zilio. 2021a. Challenges in translation of emotions in multilingual user-generated content: Twitter as a case study. arXiv preprint arXiv:2106.10719. ↗
  38. Hadeel Saadany, Constantin Orašan, Rocío Caro Quintana, Félix Do Carmo, and Leonardo Zilio. 2023. Analysing mistranslation of emotions in multilingual tweets by online mt tools. In Proceedings of the 24th Annual Conference of the European Association for Machine Translation, pages 275–284. find
  39. Hadeel Saadany and Constantin Orăsan. 2020. Is it Great or Terrible? Preserving Sentiment in Neural Machine Translation of Arabic Reviews. In Proceedings of the Fifth Arabic Natural Language Processing Workshop, pages 24–37. find
  40. Hadeel Saadany and Constantin Orǎsan. 2021. Bleu, meteor, bertscore: Evaluation of metrics performance in assessing critical translation errors in sentiment-oriented text. In Proceedings of the Translation and Interpreting Technology Online Conference, pages 48–56. find
  41. Hadeel Saadany, Constantin Orǎsan, Emad Mohamed, and Ashraf Tantavy. 2021b. Sentiment-aware measure (sam) for evaluating sentiment transfer by machine translation systems. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021), pages 1217–1226. find
  42. Charles Joseph Singer, E. J. Holmyard, and A. R. Hall, editors. 1954–58. A history of technology. Oxford University Press, London. 5 vol. find
  43. Matthew Snover, Bonnie Dorr, Richard Schwartz, Linnea Micciulla, and John Makhoul. 2006. A study of translation edit rate with targeted human annotation. In Proceedings of association for machine translation in the Americas. Cambridge, MA. find our card
  44. Lucia Specia, Frédéric Blain, Marina Fomicheva, Chrysoula Zerva, Zhenhao Li, Vishrav Chaudhary, and André Martins. 2021. Findings of the WMT 2021 shared task on quality estimation. In Association for Computational Linguistics. find our card
  45. Jannik Strötgen and Michael Gertz. 2012. Temporal tagging on different domains: Challenges, strategies, and gold standards. In Proceedings of the Eight International Conference on Language Resources and Evaluation (LREC’12), pages 3746– 3753, Istanbul, Turkey. European Language Resource Association (ELRA). find our card
  46. Katsuhito Sudoh, Kosuke Takahashi, and Satoshi Nakamura. 2021. Is this translation error critical?: Classification-based human and automatic machine translation evaluation focusing on critical errors. In Proceedings of the Workshop on Human Evaluation of NLP Systems (HumEval), pages 46–55. find
  47. S. Superman, B. Batman, C. Catwoman, and S. Spiderman. 2000. Superheroes experiences with books, 20th edition. The Phantom Editors Associates, Gotham City. find
  48. Kasturi Dewi Varathan and Nurhafizah Talib. 2014. Suicide detection system based on twitter. In 2014 Science and Information Conference, pages 785–788. find
  49. Lucas Nunes Vieira, Minako O’Hagan, and Carol O’Sullivan. 2021. Understanding the societal impacts of machine translation: a critical review of the literature on medical and legal use cases. Information, Communication & Society, 24(11):1515–1532. find
  50. Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, and Jeffrey Dean. 2016. Google’s neural machine translation system: Bridging the gap between human and machine translation. find
  51. Rabih Zbib, Erika Malchiodi, Jacob Devlin, David Stallard, Spyros Matsoukas, Richard Schwartz, John Makhoul, Omar Zaidan, and Chris Callison-Burch. 2012. Machine translation of Arabic dialects. In Proceedings of the 2012 conference of the north american chapter of the association for computational linguistics: Human language technologies, pages 49–59. find our card
  52. Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saadany-2024-cyber-risks-machine-translation-critical/