An enduring challenge in automated text generation lies in sustaining semantic coherence across extended sequences, where conventional memory mechanisms often fail to maintain contextual consistency over time. Introducing a syntax-aware retention approach, this study proposes the Layered Syntax-Aware Memory Retention Mechanism (LSMRM) as a novel framework that dynamically prioritizes memory retention based on syntactic cues, thereby enabling enhanced thematic continuity and structural coherence in long-form text. Through integrating LSMRM into a state-of-the-art open-source model, experiments revealed marked improvements in semantic coherence, syntactic alignment, and memory efficiency, particularly across sequences that demand stable narrative progression. Coherence metrics, such as context overlap ratio and syntactic alignment score, demonstrated that syntax-sensitive retention yielded sustained context and lexical diversity, effectively reducing thematic drift and enhancing narrative consistency in longer outputs. Computational analysis further indicated that LSMRM maintains processing efficiency with modest increases in latency, making it suitable for real-world applications requiring both long-form consistency and computational economy. Overall, the syntaxlayered approach of LSMRM signifies an advance in retention strategies for language models, with broad implications for domains that demand structured, coherent, and contextually adaptive text generation.