Making the Familiar Unfamiliar : AI-Driven Synectics Metaphor Generation for Computational Creativity
Year: 2025
Type: conference-paper
Abstract: Metaphors are central to human creativity, enabling abstract reasoning and novel conceptual connections. This paper explores how AI can generate and evaluate metaphors using Synectics, a structured creativity method that makes the familiar unfamiliar. We introduce a GPT-4-based pipeline to generate metaphors across four analogy types: Direct, Personal, Symbolic, and Fantasy. A comparison with 1,000 humancreated metaphors is conducted using semantic coherence (BERT), novelty (TF-IDF), and GPT-4-assessed creativity. Results show that AI-generated metaphors tend to be more novel but less coherent than human ones, with Fantasy analogies rated highest in creativity but lowest in structure. Abstract themes like Dreams and Knowledge lead to more imaginative outputs. While the system produces stylistically rich metaphors, it also reflects biases and lacks functional evaluation. This work contributes to computational creativity by demonstrating how structured prompting supports controllable metaphor generation. We suggest future directions including hybrid evaluation frameworks that combine computational and human judgment to better assess metaphor effectiveness in context.
Author: Palamas, Georgios
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7125571198