Reproduce “Texterial: A Text-as-Material Interaction Paradigm for LLM-Mediated Writing” on OpenResearch

About this paper

Jocelyn Shen, Nicolai Marquardt, Hugo Romat, Ken Hinckley, Nathalie Riche, Fanny Chevalier

The research introduces the "Texterial" interaction paradigm, which treats text as a manipulable material to make Large Language Model (LLM) capabilities more intuitive and controllable for writing tasks. Through conceptual frameworks and prototype evaluation, the work demonstrates how this approach shifts users' mental models, fosters direct manipulation, and enables more creative, non-linear, and exploratory writing processes than traditional chat interfaces.

Key insights

• Material metaphors effectively bridge the "gulfs of envisioning, execution, and evaluation" in LLM-mediated writing by providing intuitive mental models and making LLM functionalities more tangible.
• Different material metaphors resonate with distinct writing tasks and stages, such as "Text as Clay" for narrative shaping and "Text as Plants" for early-stage ideation and growth.
• Direct manipulation through gestures and spatial organization fosters non-linear, exploratory, and ambient ideation workflows, moving beyond the sequential nature of chat-based interactions.

Summary by alphaXiv.