计算隐喻处理:从隐喻识别、隐喻解释到隐喻生成OA

While existing surveys have explored various aspects of metaphorical identification,interpretation,and genera-tion,a systematic review of the paradigm shifts and comprehensive impact brought by large language models(LLMs)on the entire field of computational metaphor processing remains absent.To address this gap,this survey comprehensively reviews recent research advances.It begins by elucidating how theoretical foundations,such as conceptual metaphor theory(CMT),provide frameworks for computational modeling.It then systematically summarizes the evolution of research par-adigms and performance breakthroughs across the three core tasks-metaphor identification,interpretation,and generation-tracing the progression from early knowledge-driven methods to the era of large models.Furthermore,it provides a detailed critique of key datasets supporting the development of this field and their characteristics.The core contribution of this sur-vey lies in being the first to systematically synthesize,from the perspective of how LLMs are reshaping computational metaphor processing,the intrinsic mechanisms-such as prompt engineering and chain-of-thought reasoning-through which they achieve performance leaps.Finally,the survey concludes with a forward-looking discussion on future research directions addressing current challenges like data annotation,theory-model integration,and cross-modal understanding.