NeuTral Rewriter: A Rule-Based and Neural Approach to Automatic Rewriting into Gender-Neutral Alternatives

TL;DR

This paper introduces NeuTral Rewriter, combining rule-based and neural methods for automatic gender-neutral rewriting in English, achieving high accuracy on synthetic and natural datasets.

Contribution

It presents a novel hybrid approach for gender-neutral rewriting, including curated datasets and comprehensive evaluation methods.

Findings

Word error rate below 0.18% on multiple datasets

Effective on synthetic and real-world data

Outperforms previous approaches in accuracy

Abstract

Recent years have seen an increasing need for gender-neutral and inclusive language. Within the field of NLP, there are various mono- and bilingual use cases where gender inclusive language is appropriate, if not preferred due to ambiguity or uncertainty in terms of the gender of referents. In this work, we present a rule-based and a neural approach to gender-neutral rewriting for English along with manually curated synthetic data (WinoBias+) and natural data (OpenSubtitles and Reddit) benchmarks. A detailed manual and automatic evaluation highlights how our NeuTral Rewriter, trained on data generated by the rule-based approach, obtains word error rates (WER) below 0.18% on synthetic, in-domain and out-domain test sets.