nexttext
Lightweight Python toolkit for neutralizing gendered language in AI text.
Episodes & demos
- S01E01 Playhttps://jpeckenpaugh.github.io/nexttext/ ► Watch (opens in a new tab)
About
nexttext is a lightweight Python toolkit for neutralizing and re-framing text to reduce noise, hidden skew, and unwanted artifacts in AI pipelines. It works on both human-annotated input labels and model-generated labels, using simple, fast heuristics to neutralise gendered or stereotypical language (debias) and optionally reintroduce controlled skew for testing and validation (rebias), including a context-aware strategy. It ships with no heavy dependencies, requires Python 3.9+, and is released under the MIT license, targeting tasks like training data cleanup, inference filtering, and standardization of labels across datasets.
Summary
A dependency-free Python library that debiases and rebias gendered terms in text labels using fast heuristics, for cleaning training data and filtering model outputs.
Tech
- Python
- hatchling
- pytest
Tags
- ai-ethics
- data-cleaning
- debias
- nlp
- python-library
- text-processing