K/DA: Automated Data Generation Pipeline for Detoxifying Implicitly Offensive Language in Korean

TL;DR

This paper presents K/DA, an automated pipeline for generating paired datasets of offensive and detoxified Korean language data, addressing challenges of manual annotation and language evolution, and enabling effective detoxification model training.

Contribution

K/DA introduces an automated data generation method that produces high-quality, trend-aligned offensive language pairs for detoxification, reducing reliance on manual annotation.

Findings

Generated datasets show high pair consistency.

Datasets exhibit greater implicit offensiveness.

Effective detoxification models trained with K/DA data.

Abstract

Language detoxification involves removing toxicity from offensive language. While a neutral-toxic paired dataset provides a straightforward approach for training detoxification models, creating such datasets presents several challenges: i) the need for human annotation to build paired data, and ii) the rapid evolution of offensive terms, rendering static datasets quickly outdated. To tackle these challenges, we introduce an automated paired data generation pipeline, called K/DA. This pipeline is designed to generate offensive language with implicit offensiveness and trend-aligned slang, making the resulting dataset suitable for detoxification model training. We demonstrate that the dataset generated by K/DA exhibits high pair consistency and greater implicit offensiveness compared to existing Korean datasets, and also demonstrates applicability to other languages. Furthermore, it enables effective training of a high-performing detoxification model with simple instruction fine-tuning.