DFG Project - Computational Writing Tools
This Human-Computer Interaction (HCI) project aims to empower people in working with text through new interactive tools developed in a user-centred approach.
Currently, established digital writing software does not exploit its medium's computational capabilities. At the same time, we see a recent explosion of ideas around using Large Language Models / AI, often with a vision of replacing human writing. In order to productively and reliably make use of AI here, and crucially to empower people instead of replacing them, we need new fundamental interaction concepts and user interfaces.
Thus, this project's overarching research question asks:
"How can computational methods for processing text and user behaviour be integrated into user interfaces to augment what people can do with text?"
Concretely, the project creates and evaluates new user interfaces, interaction concepts, and data-driven models of user behaviour and content in digital writing.
It combines these elements to create applicable tools for planning, creating and revising text.
The key criteria for success are the tools' contributions to more effective, efficient and creative accomplishment of people's goals when working with text.
Technical criteria are feasibility, task-related AI performance measures, and computational costs. These criteria are assessed empirically by conducting user studies with functional prototypes of interactive AI systems and mixed methods (e.g. combining analysis of interaction logs with subjective feedback).
Across such prototypes and studies, the project explores three key challenges:
- tasks - how to support the many subtasks of writing,
- users - how to respect individual work practices, and
- interaction - how to integrate AI in space/time in writing user interfaces.
In a broad view, the project thus contributes to the scientific understanding of human-AI interaction for text, including aspects such as personalisation and adaptation, and the handling of AI biases through interaction.
This project is funded by the German Research Foundation under project number 525037874.