Metaphors for text generators–from “automatic writers” to “assistants,” “support tools,” and “stochastic parrots”–not only shape how text generation technologies are developed and received, but also change how we understand the writers and writing practices against which these technologies are understood. These subtle shifts have real consequences for writers and for writing.

This project explores existing metaphors for text generation through a combination of methods, from using corpus linguistics approaches to map different metaphors across online communities to conducting design-based research in conversation with local writers. These studies explore which metaphors are already being used, by whom, and how the particular aspects of text generators they highlight or obscure might affect human writing in the long run. These studies will provide strong groundwork for designing language technologies that respect writers’ values and practices by finding and promoting metaphors that better reflect model capabilities and support desirable futures for human writing, and writers.