algorithmic poetics

The paper describes the procedure of porting of one of the first known poetry generators in Russian from a description of a program algorithm published as an article in the USSR Academy of Sciences: Automatics and Telemechanics in 1978. Boris Katz, a computer linguist at MIT in the moment, and at that moment mathematical mechanical faculty of Moscow university graduate was working on the generator in 1972 - 1975. The generator is based on Stone, 1916, the collected poems by Osip Mandelstam. This work was inspired by his elder colleague, a professor of Moscow University, E.M.Landis. Katz started his research on machine poetry and was asking colleagues if they knew anyone working on the theme in the Soviet Union, and they failed to point him to similar work.After several years of developing the program on BECM - 4 (Big Electronic Calculating Machine) he noticed Michael Gasparov’s book Contemporary Russian verse. Metrics and rhythmics. 1974, that analysed contemporary and traditional poetic verse and general laws of organization of Russian verse. This made a considerable contribution to the work.In order to understand the context in which On Program Composing Verse was produced we have to note that unlike in other language contexts the first generated poems in Russian appeared later than musical compositions, even though the beginnings of statistical analysis of literary texts dates back to the end of the nineteenth century. Another component that proved necessary for the computational poetics in the Soviet context was the study of structural properties of literary texts such as metrical analysis of Russian verse undertaken by Vladislav Kholshevnikov, Boris Tomashevsky and Michael Gasparov. So it was important to gain both qualitative and quantitative knowledge in regards of the properties of the poetic text in Russian.Porting or recreating this generator involved creation of a database in which every word of the Mandelstam’s Stone has been classified and included into a database. The program was created by a computer scientist Boris Katz in 1978 for BECM. A poet and computer programmer Anna Tolkacheva used java script for porting the original program. The paper will report on the principles and choices made during the process, as well as the mistakes made at the first iteration of the project and methods implemented for correcting them.

Considering the effects of machine learning in aesthetic practices, the aim of this presentation is to discuss strategies for authorial inscription and the autonomy of literary writers in relation to programmable writing tools.

In a first moment I will apply David Nickel's notion of "proxy writer" (2013) to algorithmic writing agents in order to characterize these agents in what concerns their relative autonomy and place within human writing practices, and argue that digital writing environments and tools have been gradually becoming more alienated from the writer's control. Vilém Flusser's notion of "functionary" will be applied to computational writing practices in order to situate these in the broader context of writing media.

In a second moment I will discuss the writing strategies present in Jhave's ReRites (2017-18) in order to assess how such strategies cope with the high level of autonomy of neural-networks in text-generation, and how they function as a necessary precondition for literary inscription on a highly mediated writing space.

I will also discuss the reading modalities of ReadingRites, sessions in which "Poets & audience members read poetry written by artificial intelligence at the rate that the machine writes"1, and compare these with the reading modalities enabled by the print form, referring to the collection of 12 books that compile the poems which resulted from the human editing of AI-generated texts.

In a third moment, I will apply John Cayley's concept of "grammalepsy" to the human readings of non-edited AI generated texts and discuss the ontological status of machine-generated language. Finally, I will argue that, while being a tool for expanding creativity, autonomous systems also yield an algorithmicization of human writing and reading.

Jhave's Rerites provide an example of the possibility of human inscription and of relative levels of control over autonomous writing systems, allowing us to reflect on automatically generated language and on writing tools by posing the question: who's whose extension?