Paper Image

Expanding toxicity mitigation to multiple languages

Published on:

6 March 2024

Primary Category:

Computation and Language

Paper Authors:

Luiza Pozzobon,

Patrick Lewis,

Sara Hooker,

Beyza Ermis


Key Details

Examines toxicity mitigation across 9 languages from 5 scripts and multiple resource levels

Employs translated data to evaluate/enhance techniques due to lack of annotated cross-lingual datasets

Compares finetuning vs. retrieval methods for toxicity mitigation in static and continual learning settings

Explores effects of translation quality, model size, and data quantity on multilingual toxicity mitigation

AI generated summary

Expanding toxicity mitigation to multiple languages

This paper expands conventional toxicity mitigation techniques to address the complexities of working with multiple languages. Using translated datasets and comparing finetuning and retrieval methods, it examines translation quality, model size, data quantity, and more. Through experiments on 9 languages, it provides insights into multilingual toxicity mitigation.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up