Design Principles of PPFE and LMC1000: A Pre-Processing Engine for Meaning Inference and a Minimal Semantic OS

This study proposes a design framework for natural language understanding that explicitly separates language inference into two independent layers:
• L1: the physical layer, concerned with existence, perception, and factual knowledge, and
• L2: the affective layer, concerned with evaluation, meaning, and judgment.

Rather than pursuing completeness or strict formal accuracy, this work argues that inference systems achieving consistency, explainability, and reproducibility are sufficient for practical social use. Foundational inference at the L1 level—such as judgments of existence, perception, and basic knowledge—is regarded as conceptually established
by the late 1990s. Accordingly, this study focuses on the implementation of semantic inference specialized for the L2 affective layer.

Understanding the affective layer requires handling a large number of words derived from human psychological responses, including emotional terms, value-laden terms, and inherently ambiguous expressions. While such vocabulary is effective in everyday communication, it introduces substantial noise in causal reasoning, situational judgment, and affective understanding. Existing large language models (LLMs) absorb this noise through massive parameterization, but this approach faces fundamental limitations in inference consistency, linguistic transparency, computational lightness, and reproducibility.

To address these issues, this study proposes a minimal semantic operating system composed of two modules for lightweight and transparent meaning inference:
 • PPFE (Pre-Processing for Meaning Inference),
       a preprocessing engine that maps emotional, evaluative, and ambiguous words onto a PP14 (14-dimensional evaluation vector) and normalizes sentences into evaluative structures; and
 • LMC1000 (Lexical Minimal Core),
       an inference engine that represents natural language using approximately 1,000 verb-centered lexical units.

PP14 represents evaluative axes indicating the degree of confidence that an action will be socially, causally, and subjectively “completed normally.” From the perspective of L1, these axes can be approximated as success probabilities.

By allowing PPFE to perform evaluative-direction operations and LMC1000 to manage state transitions in coordination with an L1 inference engine, this architecture demonstrates that stable and reproducible causal reasoning, affective inference, and situational judgment—at approximately the level of middle-school language comprehension—can be implemented without relying on large-scale models.