Metraipolis › archive › dc_48adfff171a2

Lingua Franca or Probing Artifact? Rethinking Latent Language in Multilingual LLMs — arXiv:2609.00155

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.00155
License (as archived)
CC0-1.0
Archived
2026-10-17T06:32:47.305Z
SHA-256 of the archived text
e02a04f4b6718301cab3af55a764d2311f9fbad2d3f54b2400d89d8f3e426046
Stable URL of this page
https://metraipolis.com/documento/dc_48adfff171a2

How to cite

Lingua Franca or Probing Artifact? Rethinking Latent Language in Multilingual LLMs — arXiv:2609.00155 — arXiv (en), https://arxiv.org/abs/2609.00155 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_48adfff171a2 (Metraipolis archive dc_48adfff171a2, sha256 e02a04f4b6718301cab3af55a764d2311f9fbad2d3f54b2400d89d8f3e426046)
BibTeX
@misc{dc48adfff171a2,
  title = {Lingua Franca or Probing Artifact? Rethinking Latent Language in Multilingual LLMs — arXiv:2609.00155},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.00155},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_48adfff171a2; sha256 e02a04f4b6718301cab3af55a764d2311f9fbad2d3f54b2400d89d8f3e426046},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.00155. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Latent language identification is often used to argue that multilingual language models route computation through language-specific states, such as English pivots. However, existing probes infer latent language from different signals, such as the geometry of hidden states or what can be decoded from intermediate representations. Since such claims shape conclusions about how models share and route information across languages, we ask whether these probes measure the same phenomenon or expose distinct aspects of multilingual computation. We study this question across model families, training regimes, domains, tasks, checkpoints, and up to 27 languages. We find that identification probes systematically disagree: the GMM-based representation probe, which draws evidence from hidden state geometry, shows earlier cross-lingual mixing, whereas decoding-based probes, which rely on output-space decodability, retain sharper language-specific and more English-biased signals. These differences track model multilinguality and training progression, but are comparatively stable across domains. Our results suggest a more cautious interpretation of latent language identification, where current probes expose different aspects of multilingual processing, rather than directly revealing a single internal lingua franca.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.00155 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T06:15:12.745Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).