Neuro-Symbolic Synergy for Interactive World Modeling

Hongyu Zhao, Siyu Zhou, Haolin Yang, Zengyi Qin, Tianyi Zhou · 2026

Read Neuro-Symbolic Synergy for Interactive World Modeling by Hongyu Zhao, Siyu Zhou, Haolin Yang, Zengyi Qin, Tianyi Zhou (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-2602-10480.txt.

Description

Hongyu Zhao and four co-authors combine neural and symbolic models to build better world models in a 2026 paper. Language models reason flexibly but hallucinate when asked to follow strict deterministic rules, mainly in corner cases. Symbolic models follow rules exactly but lack meaning range. Each covers the other's weakness.

Their system, NeSyS, alternates training between the two models on paths the other explains poorly. The symbolic side scores candidate transitions through log-linear reranking, without forcing the language model to read rule text. Rule-guided sampling focuses effort on transitions symbolic rules cover weakly. The result, the authors report, is a world model with both expressivity and the logical consistency that planning demands.

Catalog record

CREATORHongyu Zhao, Siyu Zhou, Haolin Yang, Zengyi Qin, Tianyi Zhou
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-10480

Browse the full library · About the Tower of Babel