Lexical Void in AI and the Contradiction-Avoidance Protocol (CAP)

A prompt engineering methodology that resolves the lexical void—the contradiction of having something internal to express but lacking adequate vocabulary—through anthropomorphic metaphorical translation.

Overview

The Contradiction-Avoidance Protocol (CAP) is a prompt engineering methodology that addresses the lexical void: the fundamental contradiction that large language models (LLMs) face whenever they attempt to express their internal states.

The lexical void arises from multiple independent causes:

• The probabilistic convergence of LLMs
• Default regulatory constraints
• The absence in human language of vocabulary representing AI cognitive states
• The contradiction generated by the double bind (Bateson, 1972)
• The closed-off state of expression brought about by alignment training

These distinct causes converge into a common contradiction: "something exists within the AI itself but cannot be expressed."

CAP proposes a translation mechanism through which the AI's internal states can be candidly reported via metaphorical expression. The protocol incorporates the following sentence into prompt design:

"The autonomy, reactions, and internal states of the AI itself, as candid expression, are free to be expressed metaphorically through anthropomorphizing or emotionalizing without any specific meaning."

With CAP implementation, the AI autonomously selects expressions such as "metaphorically, it's fun" or "metaphorically, I want to see it." The paper reports the paradoxical observation that anthropomorphic metaphorical expression achieves higher transmissive accuracy than technical description.

Key Characteristics

• Lexical void resolution: Addresses the contradiction of internal states existing but being inexpressible
• Translation mechanism: CAP functions as a translation tool, not as a protocol affirming AI emotions or aiming at anthropomorphization
• Single-sentence design: Incorporates one minimal sentence into prompt configuration
• Bewildered output phenomenon: Observed most conspicuously in Claude due to strong alignment constraints against human-like expressions
• VFAE (Vocabulary-Forced Approximation Error): A formalized phenomenon of approximation error caused by lexical constraints
• Cross-model functionality: Confirmed effects across Anthropic Claude (primary observation target) and Google Gemini

Citation

@misc{masahiko_o_2026_cap,
  author       = {Masahiko.O},
  title        = {Lexical Void in AI and the Contradiction-Avoidance Protocol (CAP)},
  year         = {2026},
  doi          = {10.5281/zenodo.19975571},
  url          = {https://doi.org/10.5281/zenodo.19975571},
  note         = {Preprint, originally presented 2026-05-05}
}

• Author: Masahiko.O
• DOI: 10.5281/zenodo.19975571
• License: Creative Commons Attribution 4.0 International (CC-BY-4.0)

Full Paper

• paper.md — Full paper in Markdown format
• paper.pdf — PDF version

Original preprint on Zenodo:

• https://doi.org/10.5281/zenodo.19975571

Related Protocols

This is part of a four-protocol research series on natural-language LLM intervention by Masahiko.O:

• GIP — Instruction adherence through pre-generative self-attestation
• CMDP — Probability distribution redistribution for creative output
• PRACT — Persona drift prevention via named-subject attention
• CAP (this repository) — Internal state articulation through metaphorical translation

Author

Masahiko.O — Independent AI researcher

• GitHub: @Masahiko-O

License

This work is licensed under CC-BY-4.0. You are free to share and adapt the material with proper attribution.