John Swygert

November 13, 2026

DOI: To be assigned

Abstract

This paper proposes a reproducible method for using iterative language-based reasoning as an early-stage instrument of scientific discovery. The method converts intuitions, observations, and personal testimony into explicit claims, definitions, competing models, and potential experiments while preserving the distinction between linguistic coherence and empirical truth. Within TSTOEAO, articulation is treated as a pathway through constrained conceptual alternatives: each revision can expose assumptions, contradictions, or testable consequences. The proposed protocol includes versioned claim records, independent criticism, preregistered translation criteria, blinded evaluation, and failure reporting. Its value must be established by comparing measurable research outputs against conventional ideation workflows, not by assuming that articulate theories are correct.

1. Purpose and boundaries

Language can organize inquiry before measurements exist, but it cannot substitute for measurement. “Lingual science,” as used here, names a disciplined workflow for conceptual refinement, not an autonomous empirical science or a separate ontological theory. It is one methodological application of TSTOEAO’s broader interest in boundaries, pathways, and expression. The proposal is agnostic about whether any underlying universal relational substrate exists.

2. The translation problem

Novel ideas often begin as analogies, unusual observations, lived experiences, or loosely related concepts. These sources may motivate hypotheses without establishing the interpretation attached to them. An effective method should preserve the original observation, distinguish it from inference, articulate alternatives, specify measurable variables, and identify what would disconfirm the idea. The key output is not agreement but a traceable progression from ambiguous language toward independently testable statements.

3. Core protocol

Stage 1 — Capture: record the source observation or intuition in its original terms, date it, and note uncertainty. Stage 2 — Decompose: separate observations, assumptions, metaphors, causal claims, and predictions. Stage 3 — Define: specify terms operationally and identify ambiguities. Stage 4 — Challenge: construct strong rival explanations and counterexamples. Stage 5 — Formalize: translate relationships into diagrams, logic, algorithms, or equations without inventing measured quantities. Stage 6 — Operationalize: define instruments, controls, thresholds, and disconfirming outcomes. Stage 7 — Test: preregister and execute an appropriately powered investigation. Stage 8 — Revise: preserve failures and document what changed and why. Iteration is allowed at every stage, but revisions must be versioned.

4. Claim provenance and evidence grades

Every substantive claim should carry a source tag: direct observation, reported experience, established external finding, deduction from explicit assumptions, analogy, or speculation. Evidence grades should not be upgraded because the prose becomes persuasive or multiple AI systems agree. A useful claim record contains a stable identifier, original wording, revised wording, supporting evidence, strongest objection, proposed test, and current status. Contradictions should be retained as research objects until resolved rather than silently harmonized.

5. A relational representation

Let C = {c_1,…,c_n} be a set of versioned claims. Define a directed graph G = (C,E), where labeled edges represent support, contradiction, implication, analogy, or operational dependence. Let B be the declared constraints on acceptable revisions: logical consistency, provenance preservation, testability, and nonfabrication of evidence. A revision operator R(C,B,q) produces candidate changes in response to critique q. A quality function Q may assess clarity, operational specificity, falsifiability, and provenance integrity; it must not be interpreted as a probability of empirical truth. The graph is a documentation and reasoning aid, not proof that reality has the same graph structure.

6. Worked example: from metaphor to test

Initial intuition: “A boundary changes which pathways are available.” Clarification: a boundary may mean a measurable constraint, not a physical wall. Candidate model: in a controlled network, altering a gate parameter changes the set or probability of reachable states. Independent variable: prespecified gate intervention. Dependent variable: accessibility measured by a defined reachability metric over a fixed time horizon. Rival account: observed changes arise from uncontrolled noise or generic network density rather than the proposed boundary variable. Disconfirmation: the predicted direction or cross-system relationship fails under preregistered controls. The initial metaphor has now become a candidate experiment; it has not become a verified universal principle.

7. AI as a critical collaborator

Language models can generate alternative formulations, detect some inconsistencies, propose counterexamples, and accelerate documentation. They can also hallucinate citations, produce spurious formalism, reward agreeable phrasing, and repeat shared training biases. The protocol therefore requires independent source checking, explicit uncertainty, adversarial critique, and human accountability for claims and research decisions. Agreement among several models is not independent experimental replication. A full transcript is useful as provenance but not automatically a validated scientific dataset.

8. Evaluation against conventional practice

The central methodological question is whether structured iterative articulation improves downstream research quality. Randomly assign comparable research teams or participants to (A) the staged language protocol and (B) a conventional brainstorming-and-drafting workflow, with equivalent time and resources. Blind independent evaluators would score preregistered outcomes: proportion of claims with operational definitions, number and quality of genuinely discriminating predictions, independent reproducibility of proposed procedures, detection of planted contradictions, and the rate of unsupported factual assertions. Secondary outcomes could include time-to-testable-protocol and evaluator agreement. Analyze effect sizes and confidence intervals, not just favorable examples.

9. Safeguards and failure criteria

Preregister scoring rubrics, evaluators, exclusions, and primary endpoints before comparing groups. Include negative controls, deliberately ambiguous prompts, and tasks with known solutions. Preserve original and revised wording to prevent hindsight reconstruction. If the protocol yields no reliable improvement, increases false confidence, or performs worse on factual accuracy, its scope should be reduced or abandoned. Any benefits should be compared with simpler alternatives such as structured checklists, peer review, and formal requirements engineering.

10. Relationship to TSTOEAO and other research

The protocol uses boundaries as explicit constraints on acceptable reasoning moves and pathways as traceable transitions between conceptual states. “Expression” corresponds operationally to committing a claim to a versioned, inspectable statement; “non-expression” refers to unselected candidate formulations, not a demonstrated physical state. This is a methodological analogy unless independent evidence supports a deeper equivalence. The same protocol can help refine questions about cognition, engineering, or physical systems without treating those domains as identical.

11. Conclusion

Disciplined language-based refinement can be investigated as a tool for generating clearer, more falsifiable scientific proposals. Its success should be measured by reproducible downstream improvements, not rhetorical elegance. The method makes the passage from intuition to experiment visible, preserves uncertainty, and allows ideas to fail in informative ways. Whether it adds value beyond established scientific practices remains an empirical question.

References and research context

Bunge, M. (1967). Scientific Research I: The Search for System. Springer.

Kahneman, D., & Klein, G. (2009). Conditions for intuitive expertise: A failure to disagree. American Psychologist, 64(6), 515–526.

Lakatos, I. (1978). The Methodology of Scientific Research Programmes. Cambridge University Press.

Nosek, B. A., et al. (2015). Promoting an open research culture. Science, 348(6242), 1422–1425.

Popper, K. (1959). The Logic of Scientific Discovery. Hutchinson.

These works supply methodological background; no cited work establishes the proposed protocol’s effectiveness.