Story Generation from Sequence of Independent Short Descriptions
Project Overview
The document examines the application of generative AI in education, particularly focusing on its ability to produce coherent narratives from short descriptions through Natural Language Generation (NLG) techniques. It contrasts two primary methodologies: Statistical Machine Translation (SMT) and Deep Learning, particularly sequence-to-sequence models, which aim to replicate human-like storytelling. While these systems have shown progress in narrative generation, they are still considered weak AI due to their inherent limitations in creativity and originality, indicating a gap between machine-generated content and true human creativity. Furthermore, the study points out the challenges in evaluating such creative outputs, highlighting the inadequacies of current assessment metrics for measuring the quality of AI-generated narratives. Overall, the findings underscore the potential of generative AI to enhance educational tools and learning experiences while also revealing the ongoing need for improved creativity in AI systems and more effective evaluation methods.
Key Applications
Story Generation from Sequence of Independent Short Descriptions
Context: Educational context where students learn about natural language processing and machine learning applications.
Implementation: The system implements Statistical Machine Translation and Deep Learning (sequence-to-sequence) models to generate stories from short descriptions.
Outcomes: The system generates coherent story-like summaries from independent descriptions, demonstrating potential for creative text generation.
Challenges: Current systems are limited in their creative abilities and struggle to produce semantically rich stories, often relying on co-occurrence statistics rather than true narrative understanding.
Implementation Barriers
Technical Limitations
Existing systems are categorized as weak AI, struggling with higher levels of creativity and originality in language generation tasks.
Proposed Solutions: Future work should focus on developing more sophisticated models and creation of better training datasets to improve the quality of generated stories.
Evaluation Challenges
There is a lack of adequate evaluation metrics for assessing story generation quality, which currently relies on n-gram based methods that do not capture creative aspects.
Proposed Solutions: Developing new trainable metrics that evaluate creativity, coherence, and novelty in story generation.
Project Team
Parag Jain
Researcher
Priyanka Agrawal
Researcher
Abhijit Mishra
Researcher
Mohak Sukhwani
Researcher
Anirban Laha
Researcher
Karthik Sankaranarayanan
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Parag Jain, Priyanka Agrawal, Abhijit Mishra, Mohak Sukhwani, Anirban Laha, Karthik Sankaranarayanan
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