Metraipolis › archive › dc_7992feb7710c

Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention — arXiv:2608.13959

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.13959
License (as archived)
CC0-1.0
Archived
2026-10-19T02:27:53.200Z
SHA-256 of the archived text
71e3185dcb3dbc55d63027bae0a1c562e517ef94fe42a9c94ad6164198a3b199
Stable URL of this page
https://metraipolis.com/documento/dc_7992feb7710c

How to cite

Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention — arXiv:2608.13959 — arXiv (en), https://arxiv.org/abs/2608.13959 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7992feb7710c (Metraipolis archive dc_7992feb7710c, sha256 71e3185dcb3dbc55d63027bae0a1c562e517ef94fe42a9c94ad6164198a3b199)
BibTeX
@misc{dc7992feb7710c,
  title = {Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention — arXiv:2608.13959},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.13959},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7992feb7710c; sha256 71e3185dcb3dbc55d63027bae0a1c562e517ef94fe42a9c94ad6164198a3b199},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.13959. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A tool-calling router has to pick the right tool when one applies and decline when none does. Restricting the decoder to a grammar over the tool names is the standard remedy for the first, and on small models it buys a large accuracy gain. Recent work separates the loss caused by asking for a format from the loss caused by enforcing it at decode time. The second is small, which made enforcement look nearly free. The same work declines to extend that to function calling, where a constraint decides which answers exist rather than how one is written. Declining to call anything is the answer it most easily removes, and the one a router can least afford to lose. A grammar decides which tokens may be emitted and where generation stops, and a two-condition design charges both to the restriction. We therefore run three conditions over one prompt: free generation, generation stopped at the first line, and both applied together. We evaluate open-weight models from 0.6B to 4B on the same items in English and Korean, comparing the languages item by item. The two-condition contrast is negative on abstention accuracy in four of six cells with intervals excluding zero, and positive in none, costing -29.5 points at worst. On the smallest model in Korean the stop costs -20.0 points, the restriction returns +19.5, and together they leave -0.5. What the restriction gives back is readable output, not judgment. Of the 698 abstentions it repairs, 545 had no readable answer at all and 0 were correct decisions the scoring rule rejected. On items that do need a tool the contrast is positive throughout, and abstention is reported first because it is the registered measure. Both preregistered claims about language fail: Korean does not lose more of the abstentions it holds without the constraint, and the removed mass does not explain what does.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-17 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.13959 — 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 2026-10-19T02:20:31.902Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).