Proceedings of the First Workshop on Intelligent and Interactive Writing Assistants (In2Writing 2022)
Ting-Hao Huang, Vipul Raheja, Dongyeop Grammarly, Daniel Gissin, Mina Lee, Katy Ilonka Gero, Jordan Huffaker, Hwaran Lee, Atr Lab, Risako Owan, Shirley Anugrah Hayati, Chieh-Yang Gatech, Alex Tamkin, Gabriel Poesia, Dae Joon Kim, Joon Sik Park, Alex Calderwood, Kenneth J. Arnold, Melanie Subbiah, Chris Kedzie, Rasa Du, Dhruv Kumar, Adela Schneider, Andreas Madsack, Johanna Heininger, Ching‐Yi Chen, Robert Weißgraeber, Charlotte Li, Lydia B. Chilton, Daniel Adiwardana, Minh-Thang Luong, David R. So, Jamie Hall, Noah Fiedel, Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ B. Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, Tom Brown, Benjamin Mann, Nick Ryder, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric J. Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Matti Wiegmann, Michael Völske, B. Stein, Martin Potthast, Jingjing Li, Zichao Li, Tao Ge, Irwin King, Michael R. Lyu, Nikos Voskarides, Edgar Meij, Sabrina Sauer, Maarten de Rijke, Max Kreminski, Chris Martens, Vivian Liu, Nayu Liu, Wenjing Han, Guangcan Liu, Peng Zhou, Ran Zhang, Xiaorui Wang, Huabin Ruan, Guillermo Bernal, C ̧akir, Dîlan Canan, Elena L. Glassman, Vivian Emily Gunser, Tae Wan Kim, Zae Myung Kim, Li CharlotteOpenalex Unknown2022 · DOI · OpenAlex
The rise of Natural Language Generation (NLG) has transformed text production from a manual craft into an automated industrial process. In the proceedings of the First Workshop on Intelligent and Interactive Writing Assistants (In2Writing 2022), Schneider et al. present a foundational perspective in their lead paper, "Data-to-text systems as writing environment." They argue that we must shift our view of NLG systems from autonomous "black boxes" to specialized environments that support "meta-writing"—the act of designing a system that generates text.
The Challenge of Meta-Writing
Commercial data-to-text systems are increasingly used to generate vast quantities of content, such as product descriptions or financial reports. Traditionally, the success of these systems has been measured by the quality of their output. However, Schneider et al. contend that this ignores the human experience. When a human "authors" a system, they are creating a blueprint rather than a draft. This "meta-writing" process is often hindered by current NLG tools that enforce rigid, linear workflows, clashing with the recursive nature of human cognition.
Modern writing assistants like Grammarly and Wordtune represent the growing field of interactive writing environments discussed at the In2Writing workshop.
To address this, the authors synthesize two frameworks: the Flower-Hayes cognitive model of human writing and the Reiter & Dale NLG pipeline.
The Flower-Hayes Model describes writing as a non-linear cycle of Planning (generating ideas), Translating (turning ideas into words), and Reviewing (evaluating and editing). In contrast, the Reiter & Dale Pipeline is a standard machine architecture involving Document Planning (content selection), Microplanning (lexical choice), and Surface Realization (grammatical formatting).
The comparison reveals significant "friction points." For instance, humans rely on a "phonological loop"—the ability to "hear" the text in their minds to judge its flow and impact. Machines, however, operate on abstract logical formulas. This gap makes it difficult for human authors to intuit the quality of the generated text without seeing concrete examples. When a system forces a user to work purely in the abstract, it increases cognitive load and stifles the natural writing process.
Design Principles for Human-Machine Collaboration
Based on their analysis and experience with commercial systems like AX Semantics, the authors propose seven design principles to create more effective writing environments:
Modular Alignment: Software interfaces should mirror cognitive writing phases, allowing users to focus on specific tasks without being overwhelmed by the entire pipeline.
Task Flexibility: Human writing is recursive. Users must be able to jump between different stages of the process—such as editing a template while still planning content—without losing progress.
External Representations: Systems should provide space for non-verbal notes, outlines, and "scaffolding" that help the author manage complex information before it is turned into rules.
Visual Planning: Logical structures and data mappings should be represented graphically rather than just through code, making the "blueprint" easier to navigate.
Linguistic Freedom: Authors should be able to start with concrete sentences and then "de-concretize" or abstract them into rules later, rather than being forced to think in formulas from the start.
Real-Text Previews: Continuous, real-time previews of the final output are essential. This allows the author's internal "phonological loop" to provide immediate feedback on the rules they are creating.
Linguistics-Based Editing: The system should leverage its underlying grammar knowledge to assist the user. For example, if a user changes a noun from singular to plural, the system should automatically adjust the surrounding verbs and articles.
Conclusion and Future Outlook
The work of Schneider et al. highlights that as AI becomes more integrated into the writing process, the design of the interface becomes as critical as the underlying algorithm. By moving toward process-oriented evaluation—measuring how much cognitive effort is required to produce a text—we can build systems that truly empower the "meta-writer."
While the current focus is on rule-based systems for their precision and "semantic fidelity," these principles provide a roadmap for the future. As deep-learning-based writing assistants become more "steerable," the need for environments that respect the messy, recursive, and intuitive nature of human writing will only grow.
Summary updated
Abstract
Today, data-to-text systems are used as commercial solutions for automated text production of large quantities of text. Therefore, they already represent a new technology of writing. This new technology requires the author, as an act of writing, both to configure a system that then takes over the transformation into a real text, but also to maintain strategies of traditional writing. What should an environment look like, where a human guides a machine to write texts? Based on a comparison of the NLG pipeline architecture with the results of the research on the human writing process, this paper attempts to take an overview of which tasks need to be solved and which strategies are necessary to produce good texts in this environment. From this synopsis, principles for the design of data-to-text systems as a functioning writing environment are then derived.
BibTeX
@inproceedings{huang2022proceedings,
author = {Ting-Hao Huang and Vipul Raheja and Dongyeop Grammarly and Daniel Gissin and Mina Lee and Katy Ilonka Gero and Jordan Huffaker and Hwaran Lee and Atr Lab and Risako Owan and Shirley Anugrah Hayati and Chieh-Yang Gatech and Alex Tamkin and Gabriel Poesia and Dae Joon Kim and Joon Sik Park and Alex Calderwood and Kenneth J. Arnold and Melanie Subbiah and Chris Kedzie and Rasa Du and Dhruv Kumar and Adela Schneider and Andreas Madsack and Johanna Heininger and Ching‐Yi Chen and Robert Weißgraeber and Charlotte Li and Lydia B. Chilton and Daniel Adiwardana and Minh-Thang Luong and David R. So and Jamie Hall and Noah Fiedel and Rishi Bommasani and Drew A. Hudson and Ehsan Adeli and Russ B. Altman and Simran Arora and Sydney von Arx and Michael S. Bernstein and Tom Brown and Benjamin Mann and Nick Ryder and Jared Kaplan and Prafulla Dhariwal and Arvind Neelakantan and Pranav Shyam and Girish Sastry and Amanda Askell and Sandhini Agarwal and Ariel Herbert-Voss and Gretchen Krueger and Tom Henighan and Rewon Child and Aditya Ramesh and Daniel Ziegler and Jeffrey Wu and Clemens Winter and Christopher Hesse and Mark Chen and Eric J. Sigler and Mateusz Litwin and Scott Gray and Benjamin Chess and Jack Clark and Christopher Berner and Sam McCandlish and Matti Wiegmann and Michael Völske and B. Stein and Martin Potthast and Jingjing Li and Zichao Li and Tao Ge and Irwin King and Michael R. Lyu and Nikos Voskarides and Edgar Meij and Sabrina Sauer and Maarten de Rijke and Max Kreminski and Chris Martens and Vivian Liu and Nayu Liu and Wenjing Han and Guangcan Liu and Peng Zhou and Ran Zhang and Xiaorui Wang and Huabin Ruan and Guillermo Bernal and C ̧akir and Dîlan Canan and Elena L. Glassman and Vivian Emily Gunser and Tae Wan Kim and Zae Myung Kim and Li Charlotte and C akir and Michael Lyu},
title = {{Proceedings of the First Workshop on Intelligent and Interactive Writing Assistants (In2Writing 2022)}},
booktitle = {Openalex Unknown},
year = {2022},
doi = {10.18653/v1/2022.in2writing-1},
}