Author: Fumio Miyata https://orcid.org/0009-0008-8797-5578
Affiliation: Noetics Institute / noetics.institute
Published at: Zenodo Preprint
DOI: 10.5281/zenodo.20060710
Original record: https://zenodo.org/records/20060710
Abstract
Reliable compositional reasoning remains a central challenge in artificial intelligence, particularly in settings requiring deep structural manipulation. While modern large language models demonstrate impressive linguistic capabilities, they often exhibit instability when reasoning over nested or structurally complex tasks. This paper introduces Morphic Inner World (MIW), a cognitive architecture designed to explore the potential for Language- and Platform-Invariant Deterministic Intelligence. Specifically, MIW models reasoning as a structure-preserving geometric projection from natural language into a symbolic manifold defined by a free term algebra. By employing a strategic longest-match tokenizer, a fixed dictionary of 44 Morphic Primitives, and a decoupled Wisdom Base, MIW maintains high logical consistency across multiple languages and execution kernels. Evaluation across 60 tasks within the defined benchmark scope demonstrates deterministic reasoning with 100.0% accuracy and bit-identical parity across execution kernels, demonstrating that logical substance can remain invariant across linguistic and computational substrates.
Download
- PDF (recommended for Google Scholar indexing)
Citation
English citation:
Fumio Miyata, “Morphic Inner World”, Zenodo, 2026.
DOI: 10.5281/zenodo.20060710.
Also available on ResearchGate
https://www.researchgate.net/publication/404799308_Morphic_Inner_World_Achieving_Language-and_Platform-Invariant_Deterministic_Intelligence_through_Geometric_Term-Algebraic_Projection
This paper is also listed on ResearchGate for academic visibility and researcher discovery.
BibTeX
@article{miyata_miw_2026,
title = {Morphic Inner World},
author = {Fumio Miyata},
year = {2026},
doi = {10.5281/zenodo.20060710},
url = {https://doi.org/10.5281/zenodo.20060710},
journal = {Zenodo Preprint}
}Code & Data
All experimental resources and implementations are openly available on GitHub: