The Language of Collaboration: Investigating Human-AI Synergy Through Longitudinal Linguistic Analysis
The team
- Dr Yuan YuanLink opens in a new window (Project Lead), Warwick Manufacturing Group, University of Warwick
- Dr Seungcheol (Austin) Lee, College of Media and Communication, Texas Tech University
- Dr Zizhou Peng, Birmingham Business School, University of Birmingham
The project
This project investigated how sustained and iterative interaction with generative AI changes human writing and creative work. Rather than treating AI assistance as a one-off intervention, the study captured the process through which people prompted, evaluated and revised AI-generated suggestions. An online experiment compared human-only editing, AI-only editing and interactive human-AI editing of social-media advertising content.
Intended outcomes: To create a process-level dataset linking original texts, editing histories, interaction logs and final outputs; identify when AI complements rather than replaces human contribution; examine the balance between linguistic improvement, source fidelity and output homogenisation; and develop an interdisciplinary paper and future research proposals on responsible human-AI collaboration.
Project outcomes
Activities completed: The research team designed and implemented the online experiment, completed participant recruitment and data collection, and cleaned and linked the resulting datasets. The study involved 776 participants across three stages: 155 participants produced 310 source posts; 312 participants completed human-only editing; and 309 participants completed interactive human-AI editing with their full interaction logs retained. AI-enhanced versions of the source posts were also generated for matched comparison.
Key outcomes: A matched dataset has been constructed for analysing how human and AI editors transform the same source material. Preliminary analysis shows that AI editing introduces more extensive transformation and stronger advertising conventions, including calls to action, hashtags and emojis. Human editing retains more of the source language, produces more diverse outputs and adheres more closely to the Twitter length constraint. These findings have led to a new theoretical distinction between human editing as selective stewardship and AI editing as generative normalisation.
Next steps: The team is developing a new paper on human versus AI editing and will use the dataset to support further work on sustained human-AI collaboration, linguistic diversity and the preservation of human voice. The project is also informing teaching on AI literacy, digital collaboration, critical thinking and responsible use of generative AI at WMG. Analysis and manuscript development are now under way.
Enhancing interdisciplinarity
The award strengthened collaboration across strategy and digital innovation at Warwick Manufacturing Group, linguistics and communication at Texas Tech University, and marketing and experimental design at Birmingham Business School. This combination has enabled the team to connect process-level interaction data with theories of writing, creativity, human-AI complementarity and responsible AI.