The Living Machine: Language Models for Exploratory Historical Research

This presentation exemplifies a multi-disciplinary synthesis of computational methods and historical research. It showcases a cutting-edge method that draws on a key new feature of the information landscape – language models – as a means of exploring historical texts. In particular, we examine a long-standing question in cultural history concerning the trope of the 'living machine'. In a self-consciously reflexive manoeuvre, we use techniques of NLP – themselves often portrayed as sentient – to detect figures of speech in 19th-century texts that portrayed machines as self-acting, automatic and, ultimately, alive. We devised a method using masked language models for detecting unusual or surprising turns of phrase, which we collected together into sub-corpora of relevant sentences, which were then *close-read* with detailed context and historical interpretation. The talk concludes that used wisely, language models have the potential to open wholly new avenues of research in cultural and intellectual history.