人工智能生成隐喻的局限性——“咬”的个案分析

Limitations of AI-Generated Metaphors—Case Analysis of “Bite”

摘要: 隐喻生成能力作为人类认知的重要特征，在人工智能领域的模拟依旧是一项复杂的挑战。本研究基于概念隐喻理论，对比分析100条人工智能生成的“咬”类隐喻与人类样本的认知特征。研究发现：AI虽能通过字面义生成表层隐喻，但其跨域映射深度显著受限。具体表现为目标域受限，显著依赖明喻以及跨域映射失准。这些差别印证了概念隐喻理论的核心观点——人类隐喻思维根植于身体经验与概念系统的动态互动，而当前人工智能隐喻生成仅能模拟语言表层的语义关联，突显了当前人工智能系统在模仿人类隐喻创造力的深度和丰富性方面所存在的固有局限，并为人工智能隐喻能力发展提供了改进方向。

Abstract: Metaphorical generation capability represents a critical feature of human cognition, yet its simulation in artificial intelligence remains a complex challenge. This study, grounded in Conceptual Metaphor Theory, conducts a comparative analysis between 100 AI-generated “bite” metaphors and human-created counterparts. The findings reveal that while AI systems can generate superficial metaphors through literal associations, their cross-domain mapping is significantly constrained. Specific limitations include restricted target domain mappings, overreliance on simile structures, and inappropriate cross-domain mappings. These discrepancies corroborate the core tenet of Conceptual Metaphor Theory: human metaphorical thinking is rooted in the dynamic interplay between bodily experiences and conceptual systems. In contrast, current AI metaphor generation merely simulates surface-level semantic associations, highlighting the inherent limitations of existing AI systems in replicating the depth and richness of human metaphorical creativity, and providing an improved direction for the development of the metaphorical ability of AI.