posted on 2021-01-21, 14:08authored byLeah Henrickson
Natural language generation (NLG) is the process wherein computers produce output in readable human languages. Such output takes myriad forms, including news articles, sports reports, prose fiction, and poetry. These computer-generated texts are often indistinguishable from human-written texts, and they are increasingly prevalent. Yet NLG has not been subject to any comprehensive study within the humanities.
This thesis represents the first such study. Through an interdisciplinary approach rooted in the humanities, but also drawing from the social sciences and computer science, this thesis explores the social and hermeneutic implications of NLG. More specifically, it examines how NLG challenges traditional understandings of authorship and what it means to be a reader. Any act of reading engages interpretive faculties, and readers tend to assume that a text is an effort to communicate a particular message. Readers ascribe authorial intention, hence developing a perceived contract between the author and the reader: ‘the hermeneutic contract’.
Computer-generated texts in their current state bring the hermeneutic contract into question. The hermeneutic contract’s communication principle rests on two assumptions: that readers believe that authors want them to be interested in their texts, and that authors want readers to understand their texts. The contract depends on an understanding of authorship as characterised by intention-driven agency. This thesis presents the results of a series of focus groups and an online questionnaire conducted to discern readers’ emotional responses to NLG and how readers attribute authorship to computer-generated texts. The results of these studies indicate that readers assign a sense of agency to an NLG system, attempting to establish the hermeneutic contract even when faced with a computer-generated text. Consequently, I argue that a continuum from authorship to generatorship is perhaps a more suitable schema for considering computer-generated texts than a simple author-not author dichotomy.
This thesis begins with a historical overview of salient narrative NLG systems. The discussion then shifts to a theoretical consideration of authorship and the role of the reader as pertaining to NLG systems. Then, the results of the abovementioned empirical studies are reviewed and analysed. The final chapter argues for a semantic shift from describing NLG systems as tools to social agents in themselves; the NLG system truly becomes an algorithmic author.
NLG is here, and it is everywhere. This thesis explores what that means from a humanities perspective.