Cross-Domain Toxic Spans Detection
This work addresses the challenge of maintaining performance in toxic language detection for online platforms when faced with domain shifts, though it is incremental as it compares existing methods without introducing new techniques.
The study tackled the problem of detecting toxic language spans under cross-domain distributional shift by evaluating three methods, finding that a simple lexicon-based approach performed best in cross-domain conditions, with lexicons recalling more explicitly toxic words than language models.
Given the dynamic nature of toxic language use, automated methods for detecting toxic spans are likely to encounter distributional shift. To explore this phenomenon, we evaluate three approaches for detecting toxic spans under cross-domain conditions: lexicon-based, rationale extraction, and fine-tuned language models. Our findings indicate that a simple method using off-the-shelf lexicons performs best in the cross-domain setup. The cross-domain error analysis suggests that (1) rationale extraction methods are prone to false negatives, while (2) language models, despite performing best for the in-domain case, recall fewer explicitly toxic words than lexicons and are prone to certain types of false positives. Our code is publicly available at: https://github.com/sfschouten/toxic-cross-domain.