JustGen@LT-EDI 2026: Controlled Gender Inclusive and Bias-Aware Language Generation using LLMs

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

The JustGen@LT-EDI 2026 framework addresses gender biases in large language models (LLMs) by enabling controlled gender-inclusive and bias-aware language generation. Developed for the LT-EDI shared task on Gender-Inclusive Language Generation, this system aims to rewrite gender-biased sentences into neutral alternatives while preserving original meaning. LLMs often amplify gender stereotypes, androcentric language, and misgendering, impacting critical applications in education, healthcare, and legal systems. JustGen proposes a retrieval-augmented approach that integrates lexical replacement, semantic retrieval, and instruction-tuned generation. It incorporates an edit-distance constraint and a self-evaluation step to ensure minimal, coherent, and bias-free outputs. The framework also features zero-shot adaptation capabilities for low-resource languages, with implementation code available on GitHub.

Key takeaway

For NLP Engineers developing language generation systems, you should integrate controlled bias mitigation techniques to prevent LLMs from amplifying gender stereotypes. Your systems can adopt a retrieval-augmented framework, combining lexical replacement and instruction-tuned generation with edit-distance constraints and self-evaluation. This approach ensures gender-inclusive outputs while preserving meaning, especially crucial for applications in sensitive domains like healthcare or legal systems. Consider its zero-shot adaptation for low-resource language support.

Key insights

A retrieval-augmented framework enables LLMs to generate gender-inclusive, bias-aware language through controlled rewriting.

Principles

• LLMs inherit and amplify gender biases.
• Bias mitigation requires controlled generation.
• Preserve meaning in bias rewriting.

Method

The proposed method combines lexical replacement, semantic retrieval, and controlled instruction-tuned generation. It uses an edit-distance constraint and self-evaluation for minimal, coherent, bias-free outputs, supporting zero-shot adaptation.

In practice

• Implement retrieval-augmented generation.
• Apply edit-distance for minimal changes.
• Use self-evaluation for bias detection.

Topics

• Large Language Models
• Gender Bias Mitigation
• Natural Language Generation
• Retrieval-Augmented Generation
• Instruction Tuning
• Low-Resource Languages

Code references

• SupriyaChanda/gilg-ltedi-acl2026

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, NLP Engineer, AI Ethicist

Related on AIssential

• Gender Disambiguation in Machine Translation: Diagnostic Evaluation in Decoder-Only Architectures — Decoder-only MT models exhibit gender bias, but post-training improves contextual awareness and reduces masculine "Prior Bias."
• Retrieval-Augmented Generation as a Bridge Between Classical NLP Pipelines and Modern LLM Systems — Retrieval-augmented generation (RAG) addresses the limitation of static knowledge in large language models (LLMs) by connecting them to external, up-to-date knowledge sources.
• Exposing biases, moods, personalities, and abstract concepts hidden in large language models — A new method can identify and manipulate abstract concepts hidden within large language models to enhance or minimize their expression.
• Isolating LLM Lexical Bias: A Curation-Free Triangulated Metric for Preference-Stage Learning — A new metric quantifies LLM lexical bias from preference learning without manual curation.
• Opportunities and Challenges of Large Language Models for Low-Resource Languages in Humanities Research — LLMs offer transformative potential for low-resource language research in humanities, despite significant data and ethical challenges.
• SciText2Eq: Assessing LLMs for Explainable Equation Generation for Scientific Creativity — LLMs moderately generate equations from text but struggle with semantic accuracy and reliable self-evaluation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.