IHLC@LT-EDI 2026: Steering Toward Inclusivity - A Representation Engineering for Gender-Neutral Rewriting
Year: 2026
Type: conference-paper
Abstract: This paper describes the IHLC team's submission to the LT-EDI 2026 shared task on genderinclusive language generation.For Subtask A (gender-neutral rewriting), we applied Low-Rank Adaptation (LoRA) fine-tuning, achieving an 80.00% evaluation score and placing 3rd.Our primary methodological focus, however, is Subtask B (counter-narrative generation), where we propose a compute-efficient representation engineering approach.We compute a PCA-derived steering direction from counterfactual activations and inject it into the Gemma-3-4B-it model at inference time, shifting behavior toward inclusivity without weight updates.Paired with constrained prompting, this yielded polite, context-aware responses and a 78.12% score (Rank 6).We conclude with a manual evaluation of steering failure modes, detailing critical trade-offs in semantic preservation and over-steering instability.
Authors: Akhil Rajeev P, Manoj Balaji Jagadeeshan
Institution: Indian Institute of Technology Kharagpur
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166825976