Issue 3 of Startwords, a research periodical published by the Center for Digital Humanities at Princeton, is now available.
Since it was published in March 2021, “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” has sparked impassioned conversations on the unintended consequences and potential harms of prominent natural language processing (NLP) projects. While this groundbreaking paper has been influential in computer and data science — prompting reflection on the dangers of relying on poorly conceptualized and curated data — it is only beginning to be discussed by humanities scholars who use NLP methods in their research.
Startwords Issue 3, “Parrots,” features three leading digital humanities researchers — Gimena del Rio Riande (University of Buenos Aires), Lauren Klein (Emory University), and Ted Underwood (University of Illinois, Urbana-Champaign) — discussing the implications of “Stochastic Parrots” for humanities research employing NLP methods. Following an introduction by Toma Tasovac (DARIAH) and Natalia Ermolaev (Princeton University), the authors show how the paper’s attention to process and ethics in AI can be turned to humanists creating data and models for the study of literature, history, and culture.
- Underwood, “Mapping the Latent Spaces of Culture”
- Gimana del Rio Riande, “On Spanish-Speaking Parrots” (also available in Spanish)
- Lauren Klein, “Are Large Language Models Our Limit Case?”
The roundtable that kicked off this issue was part of the New Languages for NLP project, generously funded by the National Endowment for the Humanities, and was co-sponsored by the Center for Statistics and Machine Learning at Princeton and DARIAH-EU.