Instruction-Guided Poetry Generation in Arabic and Its Dialects
A new research paper introduces a large-scale instruction-based dataset for controllable Arabic poetry generation using LLMs. The dataset covers Modern Standard Arabic and various dialects, enabling tasks like writing, revising, and continuing poems based on style and rhyme criteria. Fine-tuning LLMs on this dataset shows effective poetry generation, addressing a gap in existing research that focused on analysis rather than creation.
Key facts
- The paper is arXiv:2604.27766v1.
- The dataset is instruction-based and curated for Arabic poetry.
- It covers Modern Standard Arabic and various Arabic dialects.
- Tasks include writing, revising, and continuing poems.
- Generation is controllable based on style and rhyme.
- Existing research focused on analysis, not creation.
- Fine-tuning LLMs on the dataset yields effective models.
- The work addresses poetry as a cultural art form.
Entities
Institutions
- arXiv