IIT Delhi's FIRE framework at EMNLP 2026: memory-efficient multi-agent counterspeech generation
Tanmoy_Chak · x · 2026-10-25
An IIT Delhi team (LCS2 Lab) unveiled an EMNLP 2026 main-conference paper, 'Counter with Evidence!', presenting FIRE, a framework for hate category-informed counterspeech that aims to understand the underlying claims of hate speech and respond with the right strategy and evidence.
The two-phase framework:
- HEAL: a Hate Speech Analyst built on Qwen 3-1.7B uses a trained contrasting encoder to retrieve top-k memory exemplars, then predicts structured latent variables — target group, hate type, reasoning trace, and search query; when the hate type contains factual claims, a web search tool is triggered to fetch supporting evidence.
- CARE: a Counterspeech Generator synthesizes the final response from the input, structured analysis, retrieved evidence, and stylistic patterns of exemplars.
The design targets memory-efficient reasoning, using small models plus retrieval and web search for evidence-based counterspeech.
More from Research
- Training data-poisoning victim speaks out; ByteDance's Xinggang Wang says VAR路线皆 bitter lesson — zibuyu9 · 2026-10-25
- JevBench v1.7 adds a 10-second rule: late decisions count as wrong in agent benchmark — airesearch12 · 2026-10-25
- Open-source virtual human body: 700 muscle actuators, 206 joints, 90 segments — mtizard · 2026-10-25
- EMNLP paper: users trust model verdicts over reasoning, confident tone hides errors — MaartenSap · 2026-10-25
- Prime Intellect merges IcePop score centering with truncated sampling replay in prime-rl — samsja19 · 2026-10-25
- EMBER Targets Token Embeddings: New Method Blocks LLM Knowledge Relearning — megamor2 · 2026-10-25