EMNLP 2026 paper: whether multilingual LLMs 'work in English' depends on your probing method
ABosselut · x · 2026-10-26
The EMNLP 2026 paper 'Lingua Franca or Probing Artifact?' examines whether multilingual LLM hidden states are fundamentally English-based. The authors probe the same hidden states with three latent language identification methods and get different answers—suggesting prior claims about English-dominant internal representations may be probing artifacts rather than properties of the models.
More from Research
- MemTensor's MemT-1.0-Nano trends on Hugging Face, a memory-tuned Qwen3.6-35B-A3B finetune — MemTensor · 2026-10-26
- Mathematicians dissect OpenAI's fluid singularity paper: unprecedented smooth-force construction — Singularitarian · 2026-10-26
- EMNLP 2026 Findings: "Ask or Assume?" turns coding agents into proactive help-seekers — najoungkim · 2026-10-26
- Subliminal learning transfers hidden traits across models without a shared base model — burny_tech · 2026-10-26
- Implementing SynthID-Text From Scratch: Watermarking Doesn't Hurt Output Quality — ell-hol1 · 2026-10-26
- Two teams independently discovered the same interpretability method—a sign the idea is robust — sedielem · 2026-10-26