Abstract
Natural language generation (NLG) is an interesting application of artificial intelligence which focuses exclusively on the automatic production of understandable written or spoken language. To date, numerous studies have been conducted to explore the design and implementation of NLG systems (Singh et al., 2016), the system evaluation (Van der Lee et al., 2021), and the personalization of NLG systems (Dudy et al., 2021). Reading Computer-generated Text takes a step further by examining NLG from a humanities perspective. It aims to provide a fresh angle to probe into the relationship between authors and readers, and the concept of authorship in the context of NLG.
The book consists of three substantial chapters plus a brief introduction and a conclusion. The author first introduces the concept of NLG and points out that current data-driven reports (e.g. sports, weather, and business) are largely computer-generated. He then elaborates on the concept of hermeneutic contract between authors and readers existing in the conventional author-reader relationship, which is based on two premises: (1) readers believe that authors want readers to be interested in their texts and (2) authors want readers to understand their texts (p. 4). In this regard, the emergence of NLG has engendered a more complicated contract, because of the ambiguous author figure and intention. Hence, the purpose of this book is to tap on the nontraditional change from multi-angles, especially from a humanities one.
Chapter 2, “Discovering Natural Language Generation,” provides definition of technical terms which are essential to understand the workflow of NLG. By offering a short review on several theoretical frameworks on which NLG systems are based, the author explicates that NLG system operates through either preprogrammed rules and templates or machine learning techniques (Tang et al., 2016). However, due to the lack of moral bounds and inability to distinguish between subtle differences underlying natural languages, machine learning algorithms may offend readers with the language it generates. The author also looks back on the general development of NLG systems, from primitive interactive systems, such as chatbots, to interactive fictions; from prose writing to novel writing. As communicative and aesthetic intentions are preprogrammed in the rules and templates of NLG systems, the outputs gradually resemble those of human writers and even exhibit a certain degree of problem-solving ability and creativity. It is noteworthy that since system developers play an indispensable role in the creating, maintaining, and operating stage of the text generating process, NLG systems are nowhere near a human writer in terms of autonomy.
Given the fact that computer-generated texts are joint accomplishments of NLG systems and developers, Chapter 3, “The Development of Authorship,” intends to discuss the connotation of authorship. First, the author clarifies that NLG systems are often endorsed as a method to relieve financial and labor pressure (e.g. costs and management of authors and editors, human errors, and greater economic profitability) (p. 18), and to produce massive works to cater for readers’ distinctive needs. Nevertheless, computer-generated books such as The Official Patient’s Sourcebook on Acne Rosacea have received harsh criticism because they are deemed as adaptions of digital resources, hence may fail to provide unique insight. Then, the change in the concept of “author” and “authorship” is discussed. Authors have shifted from being systematic plagiarist to being cults of personality, and according to Communications Circuit (Darnton, 1982), authors are the originator of a text (p. 23). Accordingly, the copyright of books has altered from being in the hands of printers to being belonged to authors, thus the concept of authorship has coincided with the concept of ownership. Finally, given the data-driven nature, computer-generated texts mainly provide information without personal interpretation from human writers, and readers need to apply existing knowledge in order to analyze information to get new knowledge. Therefore, hermeneutic contract remains feasible in the context of NLG systems, because computer-generated texts (1) reflect communicative intentions and (2) enable readers to convert information into knowledge based on their personal worldview, regardless of the fact that authors of NLG texts are substantially imagined by readers.
In Chapter 4, “Algorithmic Authorship and Agency,” the author incorporates abundant examples to demonstrate that NLG systems function and interact with readers in a way resembles to that of human writers and thus can be considered as a social agent. Because NLG systems may supplement and facilitate the literature production of human writers, they are often viewed as tools. But the author argues that NLG systems are social agents because of their social participation. From an Actor-Network Theory (ANT) (Latour, 2005) perspective, NLG systems, although lack of free-will intentionality, yield unpredictable and original output despite the fact that syntactic and semantic rules and patterns are predetermined. Furthermore, if trained with a corpus vast enough, they may even display the unique style of writers from whom they are learning. However, readers respond to NLG systems as they would to human writers by applying their own social experiences to deduce intentions of imagined writers, through which the hermeneutic contract is perpetuated.
In the conclusion part, the social agency of the NLG system and its identity as an algorithmic author are recapitulated. Acknowledging the NLG system as a social agent and confirming its authorship will encourage readers to appreciate computer-generated texts and promote positive reception among the public, as well as identify its contribution to literature production. Unfortunately, as the author noted, it is just a starting point. Studies concerning system developers, ethical problems of algorithmic authors, and computer-generated output in other forms of arts, are yet to be implemented.
To conclude, this book has painted a fuller picture of the social significance of NLG and paved the road for subsequent studies in related fields. In this connection, it should be on the reading list for those who are concerned with the development or/and application of NLG, and more importantly, for those who have a keen interest in the study of interactions between authors and readers.
