Metraipolis › archive › dc_f37d6e8d4bf4

The Null Token Knows: Reducing Message-Free Hallucination in ASR and NMT — arXiv:2608.15940

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.15940
License (as archived)
CC0-1.0
Archived
2026-10-18T18:58:52.208Z
SHA-256 of the archived text
068af22fa159b11a3317cab75726c81afc04b2299e0db55b69490eaefd7ae516
Stable URL of this page
https://metraipolis.com/documento/dc_f37d6e8d4bf4

How to cite

The Null Token Knows: Reducing Message-Free Hallucination in ASR and NMT — arXiv:2608.15940 — arXiv (en), https://arxiv.org/abs/2608.15940 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f37d6e8d4bf4 (Metraipolis archive dc_f37d6e8d4bf4, sha256 068af22fa159b11a3317cab75726c81afc04b2299e0db55b69490eaefd7ae516)
BibTeX
@misc{dcf37d6e8d4bf4,
  title = {The Null Token Knows: Reducing Message-Free Hallucination in ASR and NMT — arXiv:2608.15940},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.15940},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f37d6e8d4bf4; sha256 068af22fa159b11a3317cab75726c81afc04b2299e0db55b69490eaefd7ae516},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15940. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Modern encoder-decoder systems can produce fluent text even when their input contains no recoverable message. We study this failure in ASR and NMT through the models' reserved null tokens, asking whether the score for ending generation already carries a usable abstention signal. Across speech recognizers and translation models, we audit native null-token scores and scalar logit shifts. In Whisper, we additionally probe decoder states and compare supervised row edits with conventional external gates. The evaluated models often expose a useful abstention signal, but stock decoding does not reliably act on it. Raising the null-token score can sharply suppress fabrication, but aggressive intervention also deletes valid speech or shortens legitimate translations. These findings turn the null token into a diagnostic lens on hallucination and motivate evaluating abstention methods by both suppression and deletion costs, rather than by hallucination reduction alone.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG); Sound (cs.SD).
Pagina del lavoro: https://arxiv.org/abs/2608.15940 — 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 2026-10-18T18:06:57.873Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).