Texterial: A Text-as-Material Interaction Paradigm for LLM-Mediated Writing

Paper Title

Texterial: A Text-as-Material Interaction Paradigm for LLM-Mediated Writing

Publication Info

• Topic area: Human–AI interaction in writing interfaces
• Keywords: Text-as-material, generative AI, large language models, writing tools, interaction design, material metaphors, direct manipulation, creativity, user interfaces, HCI

Background and Problem

• Problem / challenge: Current LLM-powered writing tools rely on rigid, linear interfaces (e.g., chat-based systems) that limit users' ability to fluidly interact with and explore the capabilities of generative models. These interfaces often fail to surface the expressive potential of LLMs, making it difficult for users to conceptualize, execute, and evaluate text manipulations.
• Significance: Addressing these limitations can enhance creativity, foster new workflows, and make AI-mediated writing more intuitive and accessible, especially for ideation, editing, and refinement.
• Motivation and related work: Prior research has explored metaphors in HCI, conceptual frameworks for writing interfaces, and novel interaction techniques for LLMs. However, these approaches often lack a cohesive framework for integrating material metaphors into writing tools. This paper builds on these ideas to propose a new paradigm for interacting with text as a malleable, living medium.

Solution

• Proposed approach: Texterial, a conceptual framework and interaction paradigm that treats text as a material, enabling users to manipulate, grow, and refine text through material-inspired metaphors such as clay and plants.
• Novelty:
  1. A conceptual framework linking material affordances to LLM capabilities, structured around Norman’s stages-of-action model.
  2. Two technical probes (Text as Clay and Text as Plants) demonstrating how material metaphors can support different phases of the writing process.
  3. Empirical insights from a formative study and focus groups on how material metaphors shape mental models and workflows in LLM-mediated writing.
  4. A design vocabulary for creating metaphor-based interactions with generative AI.
• Procedure and key techniques:
  • Conducted a formative study using inspirational cards to explore how participants conceptualize text as material.
  • Developed two prototypes: Text as Clay (gestural sculpting for iterative refinement) and Text as Plants (gardening metaphor for ideation and growth).
  • Evaluated the prototypes through focus groups, analyzing how material metaphors influence mental models, usability, and creativity.

Results

• Concrete findings:
  • Material metaphors (e.g., clay, plants) help users conceptualize and interact with LLM capabilities intuitively.
  • Gestures like pinching, stretching, and smudging in Text as Clay supported localized, iterative refinement, while Text as Plants enabled slow, reflective ideation and organic growth of ideas.
  • Spatial layouts and visual feedback facilitated non-linear exploration and organization of text.
• Advantage over baselines:
  • Reduced cognitive load compared to prompt-based interfaces by enabling direct manipulation and immediate feedback.
  • Supported creative experimentation and non-linear workflows, which are rarely afforded by traditional chat interfaces.
• Experiments / evaluation:
  • Formative study (n=4) explored initial reactions to material metaphors.
  • Focus group study (n=10, 4 groups) evaluated the prototypes, revealing how metaphors shaped mental models and workflows.
  • Participants represented diverse writing contexts, including scientific papers, creative briefs, and presentations.
• Limitations and future work:
  • Small sample size and short-term evaluation limit generalizability.
  • Ambiguities in gesture-operation mappings and metaphor fidelity need refinement.
  • Future work includes longitudinal studies, integration of multi-modal metaphors, and exploration of collaborative writing interfaces.

Summary

Texterial introduces a novel paradigm for LLM-mediated writing by treating text as a material, enabling users to sculpt and grow text through metaphors like clay and plants. The proposed framework bridges cognitive gaps in envisioning, execution, and evaluation by aligning LLM capabilities with material affordances. Two technical probes demonstrated the feasibility and creativity of this approach, with user studies highlighting its potential to foster non-linear, exploratory, and collaborative workflows. Texterial expands the design space of AI writing tools, offering a foundation for future systems that make generative AI more intuitive and expressive.