CRII: HCC: Interactive Natural Language Technology for Supporting Writers in Structuring and Revising Documents

CRII: HCC: Interactive Natural Language Technology for Supporting Writers in Structuring and Revising Documents Written communication and documentation empower collaboration, decision-making, and execution. It is challenging to write documents that are well structured and serve the needs of readers. Tools like Grammarly help fix errors in grammar within single sentences. But they are not designed to help structure paragraphs or documents to serve an audience. Although AI systems (such as ChatGPT) can help writers, they generate text that is not in the author’s own words. The content that comes out can be uninformative, not useful, or effectively plagiarized. These problems mean that many organizations now limit the use of generative technology in content that they publish. This project explores how computer systems can help writers structure their documents. The system would not generate words for writers. Instead, it helps writers craft well-structured documents in their own words. The project will explore how to add features to word processing applications. These features will help writers organize their thoughts and structure their writing. The features may also help predict what questions and confusions that readers might have. The research team is a diverse group of undergraduate students in various majors. They will improve their skills in building tools that use artificial intelligence (AI). These tools will help people improve their writing and communication. The project seeks to study how human writers can interact with intelligent systems during the revision stage. This stage occurs between the initial composition and the final editing for mechanics. The researchers have already prototyped several possible interactions. The team will begin by completing the implementation of these interactions. The team will then refine and test those implementations. Testing will include simulated and authentic revision contexts. Participants for these evaluations will be mainly university students and faculty. Participants will perform typical writing tasks expected for classes and publications. The perspectives of both the writer and of the reader will determine success. Evaluation of writer experience will result from post-task interviews and analysis of interaction logs. Evaluation by readers may include measures of accuracy, structure, and originality. The research will use pre-trained deep neural networks for initial prototypes. This is the same approach that powers ChatGPT. Data collected from experiments with writers will help the researchers improve language technology. They may use prompt engineering, fine-tuning, or other approaches. This project should produce fundamental knowledge about how generative AI systems can help writers. The project also should produce practical open-source tools to improve communication. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.