Generation of Chinese classical poetry based on pre-trained model
Project Overview
The document explores the integration of generative AI in education, particularly through the use of a pre-trained model called BART, to generate classical Chinese poetry. It presents innovative methods for producing poetry that adheres to specific meters and styles while tackling challenges such as aligning AI outputs with user intentions and maintaining poetic structure. The authors conducted tests resembling Turing assessments with poetry experts, revealing that AI-generated poetry was often indistinguishable from that created by humans. This research underscores the potential of AI to support contemporary poets who may not possess traditional poetic skills, suggesting that inspiration can take precedence over technical mastery. The findings indicate that generative AI can be a valuable tool in creative education, fostering artistic expression and broadening access to poetic creation.
Key Applications
BART-based poetry generation models (FS2TEXT and RR2TEXT)
Context: Education and creative writing for modern Chinese poets
Implementation: The authors trained the BART model on a dataset of classical poetry, developed algorithms for generating poetry based on user input, and conducted tests to evaluate the quality of AI-generated works.
Outcomes: The AI-generated poems were indistinguishable from those written by skilled poets, indicating that AI can assist in creative processes and inspire poets.
Challenges: Users may still struggle with language and creativity, and there are ethical concerns about AI-generated works in literary contexts.
Implementation Barriers
Technical
The generated poetry may not always align perfectly with user intentions due to the complexity of poetic language.
Proposed Solutions: Implementing algorithms that enhance the relevance between generated text and user input, as well as providing tools for users to guide the generation process.
Ethical
Concerns about the integrity of poetry contests and the potential for AI-generated works to be misattributed to human authors.
Proposed Solutions: Encouraging transparency in the use of AI tools and possibly establishing guidelines for their use in competitions.
Project Team
Ziyao Wang
Researcher
Lujin Guan
Researcher
Guanyu Liu
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Ziyao Wang, Lujin Guan, Guanyu Liu
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai