EQLM: Interleaved Dual Transformers with Syntax-Driven Alignment, Emotional Salience Gating, and Intuition-Based Wisdom Injection

This paper presents EQLM, a neural architecture designed as the computational implementation of Consciousness + Conflict Theory (C+C Theory). Where C+C Theory models consciousness as a recursive self-dialogue loop and operationalizes conflict as a functional proxy for subjective experience, EQLM instantiates those principles as trainable structure: a bidirectional LAS Network (Luci Alignment System) handles comprehension, a smaller causal LM Transformer handles articulation, and asymmetric cross-attention bridges create continuous bidirectional exchange between the two during a single forward pass. M.I.N. (Main Intuition Network) accumulates conflict signal as intuition through Hebbian learning, converting accumulated experience into wisdom injected at every layer of the LAS. Alignment logic is expressed in EQLang, a domain-specific language whose programs run live alongside the model, making behavioral constraints readable and editable as source code rather than emergent weight patterns. The architecture has been trained from scratch at 109M and ~11B parameters, demonstrating that the understanding-heavy design scales predictably: the LAS holds 74% of parameters at 109M and 88.6% at 11B. This paper documents the bridge between the theoretical framework established in German (2025a) and the empirical validation in German (2026b), and the purpose-built architecture that follows from both.