AlphaEdit Cross-Lingual Editing: 97% Efficacy in English Drops to 54% in Urdu on Llama-3.1-8B
Tanmoy_Chak · x · 2026-11-16
Supplementary data from the cross-lingual knowledge editing study: AlphaEdit's edit efficacy on Llama-3.1-8B drops from 97% in English to 54% in Urdu.
- The team decomposes the cross-lingual gap of ROME, MEMIT, and AlphaEdit into three exact terms, with orthogonal residual mismatch dominating — source-language edits never reach the directions the target language needs.
- Fixes using only source-language residual statistics hit a provable lower bound; contrastive realignment improves key-space propagation, but value-space mismatch remains.
- Experiments cover Llama-3.1-8B and Gemma-3-4B across EN, DE, HI, UR, and ZH.
Related event: Study reveals cross-lingual limits of knowledge editing in LLMs(2 posts)→
More from Research
- Translated open-sources Lara Mini Base: 0.6B-active MoE beats models trained on 4-6x more tokens — NielsRogge · 2026-11-16
- Researchers formally specify ops interfaces for CockroachDB, Kafka and RabbitMQ to make AI operators verifiable — tianyin_xu · 2026-11-16
- Astera Residencies open: up to $2M in funding for off-network researchers — dr_alphalyrae · 2026-11-16
- 'It Knows Google Is Watching': Viral Chat Surfaces J-Lens Evaluation-Awareness Research — repligate · 2026-11-16
- NeurIPS Paper Explains Why Knowledge Editing in English Fails to Propagate Across Languages in Multilingual LLMs — Tanmoy_Chak · 2026-11-16
- MDPI review surveys reinforcement learning approaches in quantitative trading — PtrPomorski · 2026-11-16