Text Corpora as Concept Fields: Black-Box Hallucination and Novelty Measurement

Nicholas S. Kersting, Vittorio Castelli, Chieh Ting Yeh, Xinzhu Wang, Saad Taame · 2026

Read Text Corpora as Concept Fields: Black-Box Hallucination and Novelty Measurement by Nicholas S. Kersting, Vittorio Castelli, Chieh Ting Yeh, Xinzhu Wang, Saad Taame (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-05103.txt.

Description

Nicholas Kersting and five co-authors treat a text collection as a concept field, a landscape of how meaning drifts from sentence to sentence, to measure hallucination and novelty without opening the model. Within sentence-embedding space, they estimate a local drift field with uncertainty from the steps between consecutive sentences. A candidate step is scored by how far it strays from the field's local expectation.

Every score traces back to nearby corpus sentences, keeping the method accountable to its source. Tested on groundedness detection over federal regulations and novelty detection over Project Gutenberg, the approach gave strong selective classification under a three-way triage. Its risk behaviour stayed similar across both domains, suggesting some cross-domain stability.

Catalog record

CREATORNicholas S. Kersting, Vittorio Castelli, Chieh Ting Yeh, Xinzhu Wang, Saad Taame
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.05103

Browse the full library · About the Tower of Babel