Towards Dialogic and On-Demand Metaphors for Interdisciplinary Reading

Research Background and Issues

• Identified Problems or Challenges: The authors point out that interdisciplinary communication in fields like Human-Computer Interaction (HCI) and Science and Technology Studies (STS) is challenging. HCI researchers often face difficulties when reading STS literature due to unfamiliar terminology, unique writing styles, and abstract content. These difficulties may lead to reduced interest, hindered comprehension, and ultimately impact interdisciplinary collaboration.
• Significance: Interdisciplinary collaboration is key to addressing complex social and technological issues. However, differences in norms within disciplines (e.g., writing styles, terminology definitions) often impede communication across fields. Interdisciplinary thinking is especially crucial in addressing global challenges.
• Research Motivation and Related Work: Previous studies have developed tools (e.g., Paper Plain and ScholarPhi) to address specific challenges in reading academic literature, but these tools are limited when dealing with irregular structures and social science terminology. Thus, the authors aim to explore the use of metaphors to help HCI researchers better understand STS literature and reduce these barriers.

Solution

• Proposed Approach or Solution: Combining abstract metaphors from dialogue models with the real-time generative capabilities of large language models (LLMs), the authors propose using metaphors to facilitate interdisciplinary communication between STS and systemic HCI. The study introduces a metaphor exchange model to generate and analyze how metaphors enhance the readability and appeal of academic literature.
• Innovations:
  1. Utilizing LLMs to generate "on-demand" metaphors, fine-tuned to align with the four core characteristics of effective metaphors: experience-based, independent of the text, not overly refined, and maintaining a degree of ambiguity to foster individual understanding.
  2. Integrating metaphor generation into a dynamic dialogue model, emphasizing recipient interaction and feedback to promote shared understanding.
• Implementation Steps and Techniques:
  1. Conducting an open-ended survey with 48 HCI doctoral students to analyze their challenges in reading STS literature.
  2. Providing LLM-generated metaphors as supplementary explanations and collecting participants’ responses to the metaphors.
  3. Encouraging participants to reinterpret the original text and design interactive interfaces to explore their engagement with the metaphors.

Research Outcomes

• Specific Findings:
  1. Metaphors generated by LLMs significantly increased participants’ willingness to continue reading STS literature. The survey revealed that with metaphor support, the likelihood of continued reading increased from 40.5% to 56.3%.
  2. Participants reported improved understanding of STS texts, particularly in decoding complex terminology and abstract concepts.
  3. Metaphors provided more concrete and vivid scenarios, enabling participants to better connect with abstract concepts, even fostering visual and image-based comprehension.
• Advantages Over Existing Solutions: Compared to existing literature assistance tools, this approach emphasizes personalization and dynamic adjustment. By combining metaphor generation with dialogue models, it promotes active user engagement and understanding.
• Experimental or Evaluation Results: Open-ended surveys and sketching exercises showed that participants preferred flexible metaphor generation mechanisms and features allowing them to customize metaphor contexts. These functionalities not only enhanced the reading experience but also provided guidance for further interdisciplinary collaboration.
• Limitations and Future Directions:
  1. Limitations: The study focused on literature from a single STS author, excluding works by other prominent STS scholars; analysis primarily relied on self-reports, and future research could incorporate objective tests based on actual learning outcomes.
  2. Future Directions: Exploring metaphor exchange across different disciplinary backgrounds, further investigation into content from various subfields; studying the synergy between metaphors and other explanatory methods (e.g., summarization or simplification); developing dynamic dialogue-based agent systems.

Through this research approach, the authors provide a novel technical solution to reduce interdisciplinary reading barriers while offering important insights into the application of metaphors in academic learning environments.