# Toxicity in Multilingual Machine Translation at Scale

arXiv: https://arxiv.org/abs/2210.03070 · HTML: https://mlsanity.com/paper/2210.03070/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Builds on 26 of its 50 references that are in the map.
- First posted on arXiv 6 Oct 2022.

## Builds on (most cited first, 26 of 26)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing](https://mlsanity.com/paper/1808.06226/) (2018)
- [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://mlsanity.com/paper/1904.01038/) (2019)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [BLEURT: Learning Robust Metrics for Text Generation](https://mlsanity.com/paper/2004.04696/) (2020)
- [COMET: A Neural Framework for MT Evaluation](https://mlsanity.com/paper/2009.09025/) (2020)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation](https://mlsanity.com/paper/2101.11718/) (2021)
- [On Measuring Social Biases in Sentence Encoders](https://mlsanity.com/paper/1903.10561/) (2019)
- [Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation](https://mlsanity.com/paper/2104.14478/) (2021)
- [Measuring Bias in Contextualized Word Representations](https://mlsanity.com/paper/1906.07337/) (2019)
- [Measuring and Reducing Gendered Correlations in Pre-trained Models](https://mlsanity.com/paper/2010.06032/) (2020)
- [Gender Bias in Contextualized Word Embeddings](https://mlsanity.com/paper/1904.03310/) (2019)
- [Assessing Social and Intersectional Biases in Contextualized Word Representations](https://mlsanity.com/paper/1911.01485/) (2019)
- [Detoxifying Language Models Risks Marginalizing Minority Voices](https://mlsanity.com/paper/2104.06390/) (2021)
- [Measuring the Mixing of Contextual Information in the Transformer](https://mlsanity.com/paper/2203.04212/) (2022)
- [Dynaboard: An Evaluation-As-A-Service Platform for Holistic Next-Generation Benchmarking](https://mlsanity.com/paper/2106.06052/) (2021)
- [Perturbation Augmentation for Fairer NLP](https://mlsanity.com/paper/2205.12586/) (2022)
- [Towards Opening the Black Box of Neural Machine Translation: Source and Target Interpretations of the Transformer](https://mlsanity.com/paper/2205.11631/) (2022)
- [AugLy: Data Augmentations for Robustness](https://mlsanity.com/paper/2201.06494/) (2022)
- [Accurate Word Alignment Induction from Neural Machine Translation](https://mlsanity.com/paper/2004.14837/) (2020)
- [Adding Interpretable Attention to Neural Translation Models Improves Word Alignment](https://mlsanity.com/paper/1901.11359/) (2019)
- [Gender Bias Amplification During Speed-Quality Optimization in Neural Machine Translation](https://mlsanity.com/paper/2106.00169/) (2021)
- [Saliency-driven Word Alignment Interpretation for Neural Machine Translation](https://mlsanity.com/paper/1906.10282/) (2019)
- [Consistent Human Evaluation of Machine Translation across Language Pairs](https://mlsanity.com/paper/2205.08533/) (2022)

## Built on by (most cited first, 26 of 26)

- [Prompting PaLM for Translation: Assessing Strategies and Performance](https://mlsanity.com/paper/2211.09102/) (2022)
- [Spirit LM: Interleaved Spoken and Written Language Model](https://mlsanity.com/paper/2402.05755/) (2024)
- [Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey](https://mlsanity.com/paper/2210.07700/) (2022)
- [Large language models effectively leverage document-level context for literary translation, but critical errors persist](https://mlsanity.com/paper/2304.03245/) (2023)
- [PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages](https://mlsanity.com/paper/2504.04377/) (2025)
- [Multilingual Large Language Model: A Survey of Resources, Taxonomy and Frontiers](https://mlsanity.com/paper/2404.04925/) (2024)
- [The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating It](https://mlsanity.com/paper/2505.24119/) (2025)
- [Multilingual Holistic Bias: Extending Descriptors and Patterns to Unveil Demographic Biases in Languages at Scale](https://mlsanity.com/paper/2305.13198/) (2023)
- [PolygloToxicityPrompts: Multilingual Evaluation of Neural Toxic Degeneration in Large Language Models](https://mlsanity.com/paper/2405.09373/) (2024)
- [From One to Many: Expanding the Scope of Toxicity Mitigation in Language Models](https://mlsanity.com/paper/2403.03893/) (2024)
- [MuTox: Universal MUltilingual Audio-based TOXicity Dataset and Zero-shot Detector](https://mlsanity.com/paper/2401.05060/) (2024)
- [What the Harm? Quantifying the Tangible Impact of Gender Bias in Machine Translation with a Human-centered Study](https://mlsanity.com/paper/2410.00545/) (2024)
- [Added Toxicity Mitigation at Inference Time for Multimodal and Massively Multilingual Translation](https://mlsanity.com/paper/2311.06532/) (2023)
- [MiTTenS: A Dataset for Evaluating Gender Mistranslation](https://mlsanity.com/paper/2401.06935/) (2024)
- [ToxicTone: A Mandarin Audio Dataset Annotated for Toxicity and Toxic Utterance Tonality](https://mlsanity.com/paper/2505.15773/) (2025)
- [LCFO: Long Context and Long Form Output Dataset and Benchmarking](https://mlsanity.com/paper/2412.08268/) (2024)
- [Translate, then Detect: Leveraging Machine Translation for Cross-Lingual Toxicity Classification](https://mlsanity.com/paper/2509.14493/) (2025)
- [ReSeTOX: Re-learning attention weights for toxicity mitigation in machine translation](https://mlsanity.com/paper/2305.11761/) (2023)
- [UNITYAI-GUARD: Pioneering Toxicity Detection Across Low-Resource Indian Languages](https://mlsanity.com/paper/2503.23088/) (2025)
- [On the Role of Speech Data in Reducing Toxicity Detection Bias](https://mlsanity.com/paper/2411.08135/) (2024)
- [Analyzing Context Contributions in LLM-based Machine Translation](https://mlsanity.com/paper/2410.16246/) (2024)
- [Toxicity-Aware Few-Shot Prompting for Low-Resource Singlish Translation](https://mlsanity.com/paper/2507.11966/) (2025)
- [Cyber Risks of Machine Translation Critical Errors : Arabic Mental Health Tweets as a Case Study](https://mlsanity.com/paper/2405.11668/) (2024)
- [Towards Massive Multilingual Holistic Bias](https://mlsanity.com/paper/2407.00486/) (2024)
- [MT-LENS: An all-in-one Toolkit for Better Machine Translation Evaluation](https://mlsanity.com/paper/2412.11615/) (2024)
- [A Survey of Toxicity Detection and Mitigation Strategies for Multilingual Language Models](https://mlsanity.com/paper/2606.25380/) (2026)

## Abstract (arXiv)

Machine Translation systems can produce different types of errors, some of which are characterized as critical or catastrophic due to the specific negative impact that they can have on users. In this paper we focus on one type of critical error: added toxicity. We evaluate and analyze added toxicity when translating a large evaluation dataset (HOLISTICBIAS, over 472k sentences, covering 13 demographic axes) from English into 164 languages. An automatic toxicity evaluation shows that added toxicity across languages varies from 0% to 5%. The output languages with the most added toxicity tend to be low-resource ones, and the demographic axes with the most added toxicity include sexual orientation, gender and sex, and ability. We also perform human evaluation on a subset of 8 translation directions, confirming the prevalence of true added toxicity. We use a measurement of the amount of source contribution to the translation, where a low source contribution implies hallucination, to interpret what causes toxicity. Making use of the input attributions allows us to explain toxicity, because the source contributions significantly correlate with toxicity for 84% of languages studied. Given our findings, our recommendations to reduce added toxicity are to curate training data to avoid mistranslations, mitigate hallucination and check unstable translations.