Source-linked AI summary
A Design Space for Intelligent and Interactive Writing Assistants
Mina Lee, Katy Ilonka Gero, John Joon Young Chung, Simon Buckingham Shum, Vipul Raheja, Hua Shen, Subhashini Venugopalan, Thiemo Wambsganss, David Zhou, Emad A. Alghamdi, Tal August, Avinash Bhat, Madiha Zahrah Choksi, Senjuti Dutta, Jin L. C. Guo, Md Naimul Hoque, Yewon Kim, Simon Knight, Seyed Parsa Neshaei, Agnia Sergeyuk, Antonette Shibani, Disha Shrivastava, Lila Shroff, Jessi Stark, Sarah Sterman, Sitong Wang, Antoine Bosselut, Daniel Buschek, Joseph Chee Chang, Sherol Chen, Max Kreminski, Joonsuk Park, Roy Pea, Eugenia H. Rho, Shannon Zejiang Shen, Pao Siangliulue
TL;DR
Writing-assistant research is fragmented across communities, while the rapid spread of assistants raises concerns about comprehensive, ethical, and multidimensional understanding. The paper addresses this gap with a collaborative design space built from a systematic review of 115 papers, yielding a framework for examining existing and possible assistants. It identifies underrepresented areas and presents the design space as a practical tool for navigating and comparing designs, while acknowledging limits in its paper coverage and exclusion of commercial assistants.
Problem
Writing-assistant research is fragmented across communities, and rapid adoption raises concerns about desirable use, unforeseen consequences, ethical issues, and model-centric design.
Method
The authors collaboratively and systematically reviewed 115 papers to define five aspects, their dimensions, and potential codes through iterative coding.
Results
The resulting design space contains 35 dimensions and 143 codes and reveals underrepresentation of audience, scalability, and ecosystem dimensions in the literature.
Takeaways & Limitations
The design space offers researchers and designers a practical tool to navigate, comprehend, compare, and envision writing assistants across multiple dimensions.
Takeaways & Limitations
The review’s coverage was restricted by its search criteria and did not explicitly include commercial writing assistants without research papers.
Abstract
from arXiv · showhide
In our era of rapid technological advancement, the research landscape for writing assistants has become increasingly fragmented across various research communities. We seek to address this challenge by proposing a design space as a structured way to examine and explore the multidimensional space of intelligent and interactive writing assistants. Through a large community collaboration, we explore five aspects of writing assistants: task, user, technology, interaction, and ecosystem. Within each aspect, we define dimensions (i.e., fundamental components of an aspect) and codes (i.e., potential options for each dimension) by systematically reviewing 115 papers. Our design space aims to offer researchers and designers a practical tool to navigate, comprehend, and compare the various possibilities of writing assistants, and aid in the envisioning and design of new writing assistants.
1 INTRODUCTION
The paper addresses fragmentation across writing-assistant research communities by proposing a shared design space. Built through community collaboration and systematic review, the framework connects five aspects and supports exploration of diverse designs.
- Writing-assistant research spans NLP, HCI, CSS, and specialized areas with different emphases, fragmenting understanding of these sociotechnical systems.
- The design space provides a structured way to explore the multidimensional space of intelligent and interactive writing assistants.
- The framework organizes writing assistants around task, user, technology, interaction, and ecosystem aspects.
- A systematic review of 115 papers and iterative coding produced dimensions and codes spanning writing contexts, user relationships, learning problems, and ecosystem considerations.The design space includes 35 dimensions and 143 codes.
- The authors position the design space as a practical means to explore, understand, and compare writing assistants while supporting future design work.They release annotated papers as a living artifact for community refinement and extension.
2 BACKGROUND
Writing technologies have evolved from basic text-entry and editing tools toward systems that support writers’ cognitive processes. This expansion creates open questions about appropriate support, interaction, technical enablement, and ethical use across writing contexts.
- Writing technology has progressed from cuneiform and typewriters to word processors with flexible editing functions and cognitive-process support.Examples of cognitive support include text analysis and text-planning support.
- Consistently and persistently supporting the human writing process remains challenging, including deciding what support and interaction to provide and when to fade scaffolding.
- Researchers have built assistants for varied tasks and domains, including professional help requests, affectionate instant messages, and scientific tweetorials.
- Rapid introduction of writing assistants raises concerns about desirable use, unforeseen consequences, ethical issues, and model-centric design.The paper notes academic concerns about students’ use of AI in writing and homework.
3 APPROACH
The approach combines a defined scope and five core aspects with a collaborative systematic literature review and iterative coding process. The resulting framework treats dimensions as fundamental components and codes as their possible options.
- The design space uses five interconnected aspects: task, user, technology, interaction, and ecosystem.
- The collaborative team included researchers from HCI, NLP, Information Systems, and Education.
- The authors define the scope of intelligent and interactive writing assistants as working definitions aligned with the study’s objectives, not universal definitions.
Technology
The design space maps writing assistants as interconnected sociotechnical systems spanning user, interface, technology, interaction, task, and ecosystem considerations. It was created through staged sampling, collaborative code development, and iterative review of selected papers.
- Interaction: The user aspect includes how people steer systems and integrate system output, while the interface aspect includes metaphors, layouts, paradigms, visual differentiation, and initiation.
- Design space: Interaction dimensions are grouped by user, user interface, and system, while technology dimensions are grouped by data, model, learning, and evaluation.
- Sociotechnical perspective: Writing assistants are framed as sociotechnical systems involving technical components, user behaviors, societal norms, and connections among people and other systems.
- Sociotechnical perspective: The framework adapts the sociotechnical perspective by splitting technology into technology and interaction and renaming structure as ecosystem.
- Review process: The review retrieved 419 candidate papers from HCI and NLP sources and selected 115 papers after scope-based filtering.
- Scope: The review targeted existing assistants containing a specific technology, human user, interface-mediated interaction, and writing context rather than all possible future options.
- Review process: Five aspect-based teams developed codes through iterative review, with at least two authors reading each paper and updating codes when the existing structure was inadequate.
4 DESIGN SPACE
The design space organizes intelligent writing assistants across interconnected aspects, dimensions, and codes, offering a structured vocabulary for examining design choices. It spans task, user, technology, interaction, and ecosystem considerations identified through systematic review and iterative coding.
- Framework: The design space models writing assistants through five interconnected aspects: task, user, technology, interaction, and ecosystem.Each aspect captures a different part of the sociotechnical system surrounding writing assistance.
- Framework: Dimensions represent fundamental components of an aspect, while codes specify potential options for each dimension.The paper presents dimensions as questions and codes as possible answers, supported by examples from reviewed papers.
- Task: The task aspect characterizes rhetorical purpose, writing stage, context, audience, and the specificity of requirements.Examples include planning, drafting, revision, academic writing, narrative writing, and tasks ranging from nonspecific to detailed objectives.
- User: The user aspect includes capabilities such as writing expertise, efficiency, and technical proficiency, as well as user control and ownership of generated text.These capabilities can be targeted for improvement, while system-generated text may influence a writer’s sense of ownership.
- Technology: Technology dimensions frame writing assistance through learning problems and evaluation focuses, including classification, regression, structured prediction, and output quality.Technology design also considers how data, models, and products are accessed and licensed.
5 DISCUSSION
The design space is used to examine writing-assistant research, its gaps, ethical implications, and design trade-offs across diverse stakeholders. The discussion also identifies limitations in the literature and in constructing the framework.
- 5.1 Use Case Scenarios for the Design Space: Two scenarios show how stakeholders can use the design space to examine design choices, experimental contexts, interdependencies, and trade-offs.Researchers use it to plan an assistant for non-native English writers, while policymakers use it to analyze potential opinion-swaying effects.
- 5.2 Trends and Gaps in the Literature: The corpus shows a sharp increase in writing-assistant papers beginning in the mid-2010s, with roughly equal growth from HCI and NLP venues.The authors connect this growth to their motivation for creating a design space for researchers and designers.
- 5.2 Trends and Gaps in the Literature: Audience, scalability, and many ecosystem dimensions are under-represented, leaving gaps concerning readership, model costs, adoption contexts, and long-term effects.The paper calls for more attention to economic and computational costs, ecosystem conditions, and longitudinal changes affecting writers, readers, and the information environment.
- 5.2 Trends and Gaps in the Literature: Foundation-model papers increased from 1 in 2020 to 13 in 2023, without corresponding growth in research on trust, transparency, controllability, or ethics.The authors expect foundation-model use to grow and present the design space as a way to encourage consideration of these issues.
- 5.3 Ethical Implications of Writing Assistants: Writing assistants raise ethical concerns involving plagiarism, deceptive content, opinion manipulation, labor markets, and accommodation of diverse users.The discussion frames these concerns as requiring scrutiny, ethical governance, and attention to varied user needs and preferences.
- 5.4 Challenges in Developing a Design Space: The design space is limited by search criteria, exclusion of commercial assistants, possible annotation errors, difficult code discretization, and overlapping aspect boundaries.The authors prioritize useful, extensible dimensions and codes over exhaustive paper coverage, while acknowledging that some ecosystem dimensions were often inferred rather than explicitly reported.
6 CONCLUSION
The paper presents a design space for examining and exploring intelligent and interactive writing assistants. Built through community collaboration and systematic literature review, it organizes five aspects into 35 dimensions and 143 codes.
- The design space provides a structured framework for navigating, comprehending, comparing, and designing writing assistants.
- It covers task, user, technology, interaction, and ecosystem aspects through 35 dimensions and 143 codes.
- A large collaboration divided authors into five aspect-specific teams that developed dimensions and codes before broader annotation across all papers.
B TERMINOLOGY
The paper distinguishes writing assistants from models, systems, and technology by requiring a user-facing frontend.
- Writing assistants are computational systems that assist users with writing and include a frontend interface, unlike models, systems, and technology.
C.1 Technological Evolution in Writing Assistants
Writing-assistant technology evolved from small, human-labeled datasets supporting rule-based and statistical error detection toward models consuming and capturing patterns from increasing amounts of data.
- Early-2000s writing assistants used human-labeled data for rule-based methods or statistical models that detected errors and suggested corrections.
C.2 User Interaction & Interface Evolution in Writing Assistants
Writing-assistant interaction evolved from next-word prediction for reduced typing toward longer generated text and more varied controls embedded across writing interfaces.
- Augmentative and alternative communication research introduced next-word prediction to reduce manual typing, especially for people with motor impairments.
- Generated text expanded from single words to phrases and whole drafts depending on the use case.
- Interfaces now support implicit steering through preceding draft text, explicit instructions, inline previews, pop-up selections, sketches, and two-dimensional text canvases.
C.3 Ecosystem: Going from Micro-HCI to Macro-HCI
The paper extends writing-assistant analysis from micro-HCI, centered on task, user, technology, and interaction, to macro-HCI ecosystem concerns in broader sociotechnical contexts.
- Micro-HCI focuses on interfaces, users, and software-level task, user, technology, and interaction concerns.
- Macro-HCI addresses wider concerns including affective experience, social participation, trust, empathy, responsibility, and privacy.
- Writing assistants require ecosystem analysis because they become embedded in broader sociotechnical contexts beyond the individual writer and software.
D SYSTEMATIC LITERATURE REVIEW
The systematic literature review narrowed its search to selected HCI and NLP venues, using keyword-based retrieval that produced comparable numbers of papers from each field.
- The review included all paper tracks from selected HCI and NLP venues but excluded workshops.
- Selected HCI venues included CHI, CSCW, UIST, IUI, C&C, DIS, and ToCHI, while NLP venues included ACL, NAACL, EMNLP, EACL, and TACL.
- ACM Digital Library retrieval used “write” as a keyword, matching variations in paper titles or keywords.
- ACL Anthology retrieval used “writ” and “wrote” in titles to manually capture variations of “write.”
- 60 papers came from HCI and 55 from NLP, yielding similar numbers across the two fields.
D.4 Additional References for Technology
The technology team added references through a broader search intended to improve coverage of recent writing-assistant technologies.
- The technology team selected 25 additional papers for broader, deeper, and more relevant coverage of recent technologies.
- The expanded search used keywords such as “text revision” and “text editing” and retrieved papers from additional venues.
- The additional-paper selection process began with 80 papers before deduplication.