mlsanityData to 7 Oct 2026

A New Generation of Perspective API: Efficient Multilingual Character-level Transformers

arXiv 2202.11176 ↗First posted 22 Feb 2022cs.CL cs.AI cs.CYcited by 181

Not on the map. Why?

Cited by 40 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

On the world wide web, toxic content detectors are a crucial line of defense against potentially hateful and offensive messages. As such, building highly effective classifiers that enable a safer internet is an important research area. Moreover, the web is a highly multilingual, cross-cultural community that develops its own lingo over time. As such, it is crucial to develop models that are effective across a diverse range of languages, usages, and styles. In this paper, we present the fundamentals behind the next version of the Perspective API from Google Jigsaw. At the heart of the approach is a single multilingual token-free Charformer model that is applicable across a range of languages, domains, and tasks. We demonstrate that by forgoing static vocabularies, we gain flexibility across a variety of settings. We additionally outline the techniques employed to make such a byte-level model efficient and feasible for productionization. Through extensive experiments on multilingual toxic comment classification benchmarks derived from real API traffic and evaluation on an array of code-switching, covert toxicity, emoji-based hate, human-readable obfuscation, distribution shift, and bias evaluation settings, we show that our proposed approach outperforms strong baselines. Finally, we present our findings from deploying this system in production.

Key facts. Cited by 181 papers in the map (data through 7 Oct 2026). Builds on 21 of its 38 references that are in the map. First posted on arXiv 22 Feb 2022.

Builds on

21 of 38 references.

References · 21

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  4. 2018SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processingcited by 1,973
  5. 2020mT5: A massively multilingual pre-trained text-to-text transformercited by 1,813
  6. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,366
  7. 2018Adafactor: Adaptive Learning Rates with Sublinear Memory Costcited by 896
  8. 2017Automated Hate Speech Detection and the Problem of Offensive Languagecited by 724
  9. 2021ByT5: Towards a token-free future with pre-trained byte-to-byte modelscited by 504
  10. 2020TweetEval: Unified Benchmark and Comparative Evaluation for Tweet Classificationcited by 414
  11. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 396
  12. 2020BERTweet: A pre-trained language model for English Tweetscited by 346
9 more
  1. 2016Ex Machina: Personal Attacks Seen at Scalecited by 208
  2. 2019Racial Bias in Hate Speech and Abusive Language Detection Datasetscited by 208
  3. 2018Mesh-TensorFlow: Deep Learning for Supercomputerscited by 168
  4. 2021Charformer: Fast Character Transformers via Gradient-based Subword Tokenizationcited by 158
  5. 2021Towards generalisable hate speech detection: a review on obstacles and solutionscited by 66
  6. 2018All You Need is "Love": Evading Hate-speech Detectioncited by 56
  7. 2021Hatemoji: A Test Suite and Adversarially-Generated Dataset for Benchmarking and Detecting Emoji-based Hatecited by 50
  8. 2021Overview of the HASOC track at FIRE 2020: Hate Speech and Offensive Content Identification in Indo-European Languagescited by 38
  9. 2019Towards Robust Toxic Content Classificationcited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 3; 2023: 35; 2024: 62; 2025: 47; 2026: 34.

181 papers.

Sort:

By area:

  1. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,124
  2. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,106
  3. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 828
  4. 2024WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMscited by 460
  5. 2022A Holistic Approach to Undesired Content Detection in the Real Worldcited by 337
  6. 2022UL2: Unifying Language Learning Paradigmscited by 292
  7. 2023PaLI-X: On Scaling up a Multilingual Vision and Language Modelcited by 254
  8. 2024SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Modelscited by 225
  9. 2024PaliGemma 2: A Family of Versatile VLMs for Transfercited by 204
  10. 2023A Pretrainer's Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, & Toxicitycited by 180
  11. 2024AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Expertscited by 126
  12. 2023PaLI-3 Vision Language Models: Smaller, Faster, Strongercited by 125
38 more
  1. 2025Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrailscited by 122
  2. 2024GuardAgent: Safeguard LLM Agents by a Guard Agent via Knowledge-Enabled Reasoningcited by 113
  3. 2024Representation Noising: A Defence Mechanism Against Harmful Finetuningcited by 93
  4. 2023Language Model Behavior: A Comprehensive Surveycited by 92
  5. 2025GuardReasoner: Towards Reasoning-based LLM Safeguardscited by 89
  6. 2024Feedback Loops With Language Models Drive In-Context Reward Hackingcited by 70
  7. 2024RigorLLM: Resilient Guardrails for Large Language Models against Undesired Contentcited by 65
  8. 2024Fairness in Large Language Models: A Taxonomic Surveycited by 57
  9. 2024ShieldLM: Empowering LLMs as Aligned, Customizable and Explainable Safety Detectorscited by 48
  10. 2024Towards Human-centered Proactive Conversational Agentscited by 48
  11. 2023On the Challenges of Using Black-Box APIs for Toxicity Evaluation in Researchcited by 44
  12. 2024R2-Guard: Robust Reasoning Enabled LLM Guardrail via Knowledge-Enhanced Logical Reasoningcited by 41
  13. 2023Are Personalized Stochastic Parrots More Dangerous? Evaluating Persona Biases in Dialogue Systemscited by 37
  14. 2023Lost in Translation: Large Language Models in Non-English Content Analysiscited by 34
  15. 2024Gradient-Based Language Model Red Teamingcited by 33
  16. 2024Preference Tuning For Toxicity Mitigation Generalizes Across Languagescited by 31
  17. 2024LLM-jp: A Cross-organizational Project for the Research and Development of Fully Open Japanese LLMscited by 31
  18. 2024A Review of Multi-Modal Large Language and Vision Modelscited by 30
  19. 2024ChatBug: A Common Vulnerability of Aligned LLMs Induced by Chat Templatescited by 29
  20. 2023ToViLaG: Your Visual-Language Generative Model is Also An Evildoercited by 27
  21. 2023Unveiling Safety Vulnerabilities of Large Language Modelscited by 26
  22. 2023People Make Better Edits: Measuring the Efficacy of LLM-Generated Counterfactually Augmented Data for Harmful Language Detectioncited by 25
  23. 2024"They are uncultured": Unveiling Covert Harms and Social Threats in LLM Generated Conversationscited by 24
  24. 2024Jill Watson: A Virtual Teaching Assistant powered by ChatGPTcited by 24
  25. 2024Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixturecited by 22
  26. 2024Adversarial Attacks of Vision Tasks in the Past 10 Years: A Surveycited by 22
  27. 2025Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safetycited by 22
  28. 2024On Calibration of LLM-based Guard Models for Reliable Content Moderationcited by 21
  29. 2024Teuken-7B-Base & Teuken-7B-Instruct: Towards European LLMscited by 20
  30. 2025VLM as Policy: Common-Law Content Moderation Framework for Short Video Platformcited by 19
  31. 2024From One to Many: Expanding the Scope of Toxicity Mitigation in Language Modelscited by 18
  32. 2023Data Management For Training Large Language Models: A Surveycited by 17
  33. 2025SoK: Evaluating Jailbreak Guardrails for Large Language Modelscited by 17
  34. 2024PolygloToxicityPrompts: Multilingual Evaluation of Neural Toxic Degeneration in Large Language Modelscited by 16
  35. 2024AcTracer: Active Testing of Large Language Model via Multi-Stage Samplingcited by 16
  36. 2024Search, Verify and Feedback: Towards Next Generation Post-training Paradigm of Foundation Models via Verifier Engineeringcited by 16
  37. 2023What Makes it Ok to Set a Fire? Iterative Self-distillation of Contexts and Rationales for Disambiguating Defeasible Social and Moral Situationscited by 15
  38. 2024Contextualized Counterspeech: Strategies for Adaptation, Personalization, and Evaluationcited by 15

50 most cited of 181 All 181 (JSON)