多模态和大规模多语言翻译中推理时新增毒性缓解
Added Toxicity Mitigation at Inference Time for Multimodal and Massively Multilingual Translation
AI总结：
提出MinTox流程，在推理时通过多模态毒性检测分类器识别并缓解翻译新增毒性，应用于SEAMLESSM4T系统，可过滤25%-95%毒性且保持翻译质量。
摘要：
在翻译语境中，新增毒性（added toxicity）指的是生成的翻译输出比输入包含更多毒性的现象。本文提出了MinTox，一种新颖的流程，用于识别新增毒性并在推理时缓解该问题。MinTox使用一个毒性检测分类器，该分类器是多模态的（语音和文本），并能在大规模语言中工作。缓解方法直接应用于大规模语言的文本输出。MinTox被应用于SEAMLESSM4T，这是最新的多模态和大规模多语言机器翻译系统。对于该系统，MinTox在多个领域、模态和语言方向上实现了显著的新增毒性缓解。MinTox能够大致过滤掉25%至95%的新增毒性（取决于模态和领域），同时保持翻译质量。
英文摘要：
Added toxicity in the context of translation refers to the fact of producing a translation output with more toxicity than there exists in the input. In this paper, we present MinTox which is a novel pipeline to identify added toxicity and mitigate this issue which works at inference time. MinTox uses a toxicity detection classifier which is multimodal (speech and text) and works in languages at scale. The mitigation method is applied to languages at scale and directly in text outputs. MinTox is applied to SEAMLESSM4T, which is the latest multimodal and massively multilingual machine translation system. For this system, MinTox achieves significant added toxicity mitigation across domains, modalities and language directions. MinTox manages to approximately filter out from 25% to 95% of added toxicity (depending on the modality and domain) while keeping translation quality.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
