Filtering for Creativity: Adaptive Prompting for Multilingual Riddle Generation in LLMs

2025-08-26

AI Summary

Large language models (LLMs) frequently exhibit cultural misalignment and limited creativity in multilingual riddle generation; conventional prompting methods often yield templated repetition or superficial paraphrasing. Method: We propose the Adaptive Originality Filtering (AOF) framework—a tuning-free approach that dynamically suppresses redundancy via a cosine similarity–driven redundancy rejection mechanism and lexical novelty constraints, thereby jointly enhancing cultural adaptability and abstract creativity across languages. AOF integrates zero-shot/few-shot prompting with chain-of-thought reasoning. Evaluation employs Self-BLEU (lower scores indicate reduced redundancy) and Distinct-2 (higher scores reflect greater lexical diversity). Results: Experiments span three state-of-the-art LLMs and four language pairs. GPT-4o achieves 0.177 Self-BLEU and 0.915 Distinct-2 on Japanese—significantly outperforming baselines—demonstrating AOF’s effectiveness and cross-lingual generalizability for creative text generation.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Multilingual riddle generation challenges large language models (LLMs) to balance cultural fluency with creative abstraction. Standard prompting strategies -- zero-shot, few-shot, chain-of-thought -- tend to reuse memorized riddles or perform shallow paraphrasing. We introduce Adaptive Originality Filtering (AOF), a prompting framework that filters redundant generations using cosine-based similarity rejection, while enforcing lexical novelty and cross-lingual fidelity. Evaluated across three LLMs and four language pairs, AOF-enhanced GPT-4o achieves exttt{0.177} Self-BLEU and exttt{0.915} Distinct-2 in Japanese, signaling improved lexical diversity and reduced redundancy compared to other prompting methods and language pairs. Our findings show that semantic rejection can guide culturally grounded, creative generation without task-specific fine-tuning.

Problem

• Generating multilingual riddles with cultural fluency and creativity
• Overcoming memorization and shallow paraphrasing in LLM outputs
• Ensuring lexical novelty and cross-lingual fidelity in generation

Innovation

• Adaptive Originality Filtering framework for multilingual riddles
• Cosine similarity rejection to filter redundant generations
• Enforces lexical novelty and cross-lingual fidelity