Textoshop: Interactions Inspired by Drawing Software to Facilitate Text Editing

Research Background and Issues

• Identified Problems and Challenges:

  1. Traditional text editors (e.g., Microsoft Word) have limited capabilities for organizing text and do not naturally support nonlinear writing processes.
  2. Limited options for text rewriting make it difficult to explore variations in tone, sentence structure, or organization.
  3. The process of reorganizing expressed ideas is cumbersome, lacking simple methods for merging or splitting existing text.
  4. Insufficient tools for managing different text versions often require reliance on manual backups or inconvenient "undo" functions.

• Significance of the Research: Text editing tasks are central to creative writing, academic writing, and professional communication. Improving text editing tools can enhance user efficiency and creativity. In recent years, generative AI (e.g., large language models) has shown great promise in text generation. However, user interaction with these systems is often constrained by complex prompt-based inputs, which lack efficiency and intuitiveness.

• Research Motivation and Related Work:

  1. Interactive and flexible features in graphic design software (e.g., Adobe Photoshop) can inspire innovations in text editing, such as drag-and-drop, zooming, and intelligent layer management.
  2. Existing research has attempted to use generative AI for text editing but often lacks sufficient user control.
  3. Previous studies on multimodal interaction interfaces suggest that adopting familiar tool features can help create intuitive and user-friendly software.

Proposed Solution

• Proposed Approach: The authors propose a text editing tool called "Textoshop," inspired by the interaction methods of graphic design software. It offers a set of innovative features to make text editing faster, more intuitive, and more flexible.

• Innovative Features:

  1. Intuitive Interaction Inspired by Graphic Design Software: Treats text as pixels, sentences as regions, and tone as colors, drawing on the "direct manipulation" methods used in graphic and image editing.
  2. Tool and Function Mapping: Introduces innovative features such as tone selection, layered editing, Boolean operations, drag-and-drop, and rotation.
  3. Tone Sliders and Tone Picker: Visualizes tone adjustments, enabling exploration of tone dimensions such as formal/informal, positive/negative, and complex/simple using a "color space" metaphor.

• Specific Features and Technologies:

  1. Direct Manipulation:
     • Drag-and-drop: Reposition text.
     • Zoom: Adjust text length while maintaining semantic integrity.
     • Rotate: Change the word order of sentences (e.g., switching between active/passive voice).
  2. Toolset:
     • Tone Brush: Apply a specified tone.
     • Tone Eyedropper: Extract tone from selected text.
     • Rhetorical tools: For example, a "smoothing" tool to connect sentences and a tense-switching tool to change verb tenses.
  3. Boolean Operations:
     • Union: Combine two text segments.
     • Intersection: Extract common information.
     • Subtraction: Remove content from one text segment based on another.
  4. Text Layering:
     • Multi-layer structures separate editing tasks, with each layer independently toggleable for display, aiding version management and text organization.

• Implementation Process:

  • The front-end interface is built using TypeScript, supporting dynamic interaction.
  • The OpenAI GPT-4o model is employed for handling complex text modification tasks, supplemented by natural language processing techniques.

Research Outcomes

• Key Results:

  1. Participants were more successful and preferred using Textoshop over traditional tools for text editing tasks.
  2. The tool significantly improved user efficiency and satisfaction in nonlinear writing, tone exploration, version management, and text creation.

• Comparative Advantages:

  1. Compared to traditional editors, Textoshop is more intuitive and requires less time to complete complex text operations.
  2. Compared to text-prompt-based large language model interfaces, Textoshop offers greater direct control and operational flexibility.

• Experimental and Evaluation Results:

  1. Significant improvements in user success rates and usability scores: task completion ratings increased by 14%, and the System Usability Scale (SUS) score rose to 90 ("excellent").
  2. Average task completion time was reduced by approximately 55 seconds.
  3. Most participants found the design intuitive, with tone tools and layering features particularly helpful for complex writing tasks.

• Limitations and Future Directions:

  1. For users with no experience in graphic design software, the learning curve may be steep.
  2. Current text generation still relies on large language models, which face potential challenges such as bias, privacy concerns, and semantic inconsistencies.
  3. As with existing text editing tools, tracking subtle changes in modified text requires further optimization.
  4. Future work could incorporate more graphic design software features (e.g., "adjustment layers" or blending modes) into text editing to further enhance system flexibility and creative support.
  5. The research aims to extend the study to long-term scenarios to better understand the tool's impact on writing workflows and creative output.