Uploaded May 2023 | Updated September 2026, 23 hours ago
Abstract: Writing is a challenging and demanding task that plays a critical role across various domains. Creative writing calls for innovative and attractive ideas, while scientific writing emphasizes clarity and a
comprehensive presentation of concepts. Different fields present unique writing challenges and thus require different forms of support. In this talk, I will concentrate on two writing support projects: (1) story writing
and (2) scientific figure captioning. In the first part of the talk, I will discuss our attempts to support story writing by introducing Heteroglossia,
an editor designed to help writers generate story plot ideas to continue their stories. Heteroglossia allows users to retrieve ideas from four different functions: (1) crowd ideation, (2) plot ideation, (3) GPT-3 plot
ideation, and (4) GPT-3 completion. Transitioning to scientific writing in the second segment, I will focus on our scientific figure captioning project. Drawing from our annotations on arXiv cs.CL papers, we found that
53.88% of captions were unhelpful for readers, highlighting the need for better captions. Our analysis also suggested that the scientific figure captioning task should be tackled as a text summarization task since a
significant portion of the information is presented in the sentences referring to the figure. Lastly, I will briefly discuss potential future directions for these two areas of writing support.
Bio: Chieh-Yang is a Ph.D. candidate at the College of Information Sciences and Technology at Pennsylvania State University. He specializes in Natural Language Processing (NLP) and Human-Computer Interaction (HCI), and his research focuses on developing AI-powered and crowd-powered tools to support various language-related tasks. Specifically, Chieh-Yang has built systems to support (i) creative writing, (ii) scientific writing, (iii) data annotation, (iv) language learning, and more. Chieh-Yang cares deeply about how technologies, particularly Language Models (LLMs), can benefit humans in different tasks.
Abstract: Writing is a challenging and demanding task that plays a critical role across various domains. Creative writing calls for innovative and attractive ideas, while scientific writing emphasizes clarity and a
comprehensive presentation of concepts. Different fields present unique writing challenges and thus require different forms of support. In this talk, I will concentrate on two writing support projects: (1) story writing
and (2) scientific figure captioning. In the first part of the talk, I will discuss our attempts to support story writing by introducing Heteroglossia,
an editor designed to help writers generate story plot ideas to continue their stories. Heteroglossia allows users to retrieve ideas from four different functions: (1) crowd ideation, (2) plot ideation, (3) GPT-3 plot
ideation, and (4) GPT-3 completion. Transitioning to scientific writing in the second segment, I will focus on our scientific figure captioning project. Drawing from our annotations on arXiv cs.CL papers, we found that
53.88% of captions were unhelpful for readers, highlighting the need for better captions. Our analysis also suggested that the scientific figure captioning task should be tackled as a text summarization task since a
significant portion of the information is presented in the sentences referring to the figure. Lastly, I will briefly discuss potential future directions for these two areas of writing support.
Bio: Chieh-Yang is a Ph.D. candidate at the College of Information Sciences and Technology at Pennsylvania State University. He specializes in Natural Language Processing (NLP) and Human-Computer Interaction (HCI), and his research focuses on developing AI-powered and crowd-powered tools to support various language-related tasks. Specifically, Chieh-Yang has built systems to support (i) creative writing, (ii) scientific writing, (iii) data annotation, (iv) language learning, and more. Chieh-Yang cares deeply about how technologies, particularly Language Models (LLMs), can benefit humans in different tasks.
