poetry generation

Amid the Great Plague of London (1665–1666), a man named John Peter developed a peculiar system allowing for the procedural generation of Latin poetry. A decade later, in 1677, Peter's system was published in a landmark booklet, titled "Artificial Versifying," whose subtitle proclaims that anyone "that only knows the A.B.C. and can count 9" may use it to produce "true Latin, true verse, and good sense" [1].The system itself centers on six tables in which letters are distributed across grids of cells. To generate a line of poetry, the user first produces a string of six digits (e.g., "952129"). Next, each digit is used to retrieve a sequence of letters from the table corresponding to that digit's position in the string. The letters obtained from a given table form one of nine words contained in that table, and the concatenation of the six chosen words constitutes a line of Latin verse in dactylic hexameter. The system is capable of generating 9^6, or 531,441, lines of verse.As a bizarre forerunner of electronic literature, "Artificial Versifying" was wildly successful: the booklet appeared in three editions, and its procedure was reprinted in books and periodicals for the next 200 years [2-5]. Sadly, Peter's innovative system has received scant treatment by scholars working in this area today [6-9]. This limited coverage is incommensurate to its importance as a groundbreaking work produced centuries ahead of its time. Indeed, its combinatorial method is similar to those employed in early computer poetry, such as Theo Lutz's "Stochastische Texte" [10-11].We have carried out the first translation into English of the "Artificial Versifying" system. While it would be easy to translate any one of the 531,441 hexameter verses that the system can produce, we sought instead to translate the system itself into English. This only entailed translating the 54 words in the six tables, but the process raised a number of interesting challenges nonetheless. The major difficulty is in preserving both meaning and meter, and in total we identified twelve features of the original system that we sought to maintain. In wrangling with interrelations between these features at the level of combinatorics, our design space was not unlike Peter's. While a core aim of this process has been to make "Artificial Versifying" accessible to non-Latin speakers today, this act of translation has helped us to better appreciate the triumph of the system's design.While our project seeks to celebrate an unheralded pioneering effort in the area that became electronic literature, we situate this work amid emerging scholarship on the challenges and opportunities of translating computational textual artifacts. This subarea of translation studies [12] is perhaps best characterized by the Renderings project carried out by Nick Montfort, Piotr Marecki, and other collaborators in the last decade [14-17], though others have taken it up [18-21]. In this paper, we will show that the peculiar considerations inherent in the translation of computational textual artifacts are already present in protocomputational works that are sufficiently procedural, such as "Artificial Versifying."

Computer programming is a general-purpose way of using computation. It can be instrumental (oriented toward a predefined end, as with the development of well-specified apps and Web services) or exploratory (used for artistic work and intellectual inquiry). Professor Nick Monfort’s emphasis in this talk, as in his own work, is on exploratory programming, that type of programming which can be used as part of a creative or scholarly methodology. He says a bit about his own work but uses much of the discussion to survey how many other poet/programmers, artist/programmers, and scholar/programmers are creating radical new work and uncovering new insights.

09:08 p5.js12:38 The Deletionist14:26 Permutated Poems of Poems of Brion Gysin18:18 Curveship21:00 A Noise Such As a Man Might Make24:03 Oral Poetics29:35 Q&A

We present a simple yet effective method for generating high qual- ity classical Chinese poetry with Generative Pre-trained Language Model (GPT)[5]. The method adopts a simple GPT model, without using any human crafted rules or features, or designing any additional neural compo- nents. While the proposed model learns to generate various forms of clas- sical Chinese poems, including Jueju(绝句), Lu ̈shi(律诗), various Cipai(词牌) and Couples(对联), the generated poems are of very high quality. We also propose and implement a method to fine-tune the model to generate acrostic poetry. To the best of our knowledge, this is the first to em- ploy GPT in developing a poetry generation system. We have released an online mini demonstration program on Wechat1 to show the generation capability of the proposed method for classical Chinese poetry.

Randomly generated content poses problems for theories of digital art: such content is resistant to structural theories, which can only provide templates, and one cannot assume a shared text for close analysis. Instead of reaching fixed endings, such works also tend to be of indefinite length or at least suggest indefinite possible combinations. I argue that the impact of such works can instead be found in how one attempts to work through their underlying grammar, based on limits in the algorithms that generate the content — not those limits themselves, but how their outlines come to be known. Repetitively iterating through these works simultaneously upholds the chance nature of the epiphenomenal occurrences while also illustrating the sameness of the underlying algorithm over time, creating a future-oriented interpretive arc. I examine two works that play off of this technique in different ways: Nick Montfort’s Taroko Gorge, a poetry generator which uses random generation to distill the essence of its object’s possibility, and the action role-playing game Torchlight, which attempts to elevate chance beyond a mere gameplay mechanic and toward an ethic.

(Author's abstract at DHQ)

Chris Funkhouser's notes on Gnoetry and the Gnoetry Daily site, included as a full text pdf attachment.