Between Myself, Yi Sang, and Google

My approach to ‘After Yi Sang’ has been to use Google Translate as a compositional partner. In that sense I think of this project as a collaboration between myself, Yi Sang, and the Google Translate algorithm. As I’m sure other poets have discovered, Google Translate can be endlessly inventive, given the right inputs. For my purposes, Google Translate’s most useful feature is the (fairly new) voice recognition button. Voice recognition usefully compounds the error inherent to machine translation with the error inherent to voice recognition software, and error, in this context, is the mother of invention.

To gum things up even further, I only speak to Google in languages other than English, specifically Chinese, French, and Korean. I don’t speak any of these natively, though I’m reasonably fluent in Korean—my mom is Korean, I’ve lived there off and on, and I studied it for years in college. Mandarin Chinese I hardly know at all, but I studied it for a year in college and was told at one point that my pronunciation was very good. My French is self-taught, but I’ve been at it for years.

The point, for the purpose of this project, is that I pronounce all of these languages ‘well-enough’ to be understood, but still with a foreign accent. A person would have no problem understanding me, but Google Translate is not (yet) a person. At times my speech is halting and inconsistent, and Google rewards my clumsiness with pages of useful material.

It might be easier to explain how all this works by way of illustration. Below I’ve included some screenshots illustrating different stages of the ‘translation’ part of the process. In the first set, the English text on the left side is my translation of a widely anthologized poem of Yi Sang’s called ‘Mirror’ (거울). On the right we have Google Translate’s rendering of my translation into Chinese, Korean, and French, respectively:

One aspect of this project that I’ve especially enjoyed has been discovering the various ‘personalities’ of Google Translate’s language algorithms. For one thing, some algorithms are clearly much better than others when it comes to generating ‘accurate’ renderings of complex English texts, poems in particular. Anyone who reads Korean, for instance, will see that Google’s rendering is especially bad—the resulting text doesn’t even read like grammatical Korean, much less as a good translation of the input text. The French, in contrast, is really not bad, relatively speaking. I don’t read Chinese well enough to know how it compares to the other two, but I have reason to believe that it’s at least better than the Korean.

The next set of screenshots shows the next stage of the process, in which I take the (re-)translated texts generated above and read them back to Google using the voice recognition feature. Here, too, the different algorithms have their particular quirks and tend to generate characteristic results. I haven’t yet spent enough time with them to be able to describe these characteristics very well, but I think what’s shown here is enough to give a sense of what I mean.

In the next step of the process, I collect this mass of ‘translated’ material and work to cut it into manageable lines—about as long on average as a line of blank verse. I then feed these lines into an online random list generator. So what I end up with is a randomized list of hundreds of mistranslated fragments representing a sample of Yi Sang’s collective work. I then ‘write-through’ the resulting randomized list, choosing lines and phrases more-or-less at will and then working to shape the results. So the resulting poems are collages built out of the body of mistranslated material generated in the first two steps.

I have a lot of different (perhaps disparate) ideas about what these poems are doing on a theoretical level. On the one hand they perform ‘decay’ in the form of digital lossiness and systematic disintegration; on the other they perform a weird fecundity. Algorithms don’t suffer writer’s block, so Google Translate can generate infinitely varied (but still recognizable) ‘versions’ of the original poems. By the same token, no random transformation of the materials list is exactly the same as any other (or at least is very unlikely to be), and no writing-through of the ‘same material’ ever produces the same results. The ‘genetic code’ of the original texts is thus placed into the service of an endlessly self-replicating (one might say viral) process.

In one sense a flower is just an algorithmic elaboration of its genetic programming. In that sense I like to think of these poems as a series of fractal time-lapse flowers, blooming, decaying, and collapsing in the space of 15-30 pasted-together lines. Various uncanny flowers make also make appearances in this particular series.

A foundational text for this project is Jack Spicer’s After Lorca. In a fictional letter from Lorca to Spicer that begins the book, Lorca memorably describes Spicer’s fraudulent or otherwise ‘experimental’ translations of Lorca’s poems as ‘unwilling centaurs’ (107). I think that ‘After Yi Sang’ departs from After Lorca in this respect, in degree if not in kind: the mutant poems that ‘hatch’ as a result of the process describe above could not be fairly characterized as centaurs, unwilling or otherwise; the layering of the various procedures and the constant intervention of chance, accidental error, and other sources of translational noise render it impossible to determine where I begin and Yi Sang ends. The ‘original’ poems are mutilated and mutated beyond recognition, not least because the process does not respect the boundaries of individual poems. And yet the results, I think, beget recognition. ‘After Yi Sang’ resembles Yi Sang both thematically and, in some cases, formally (Steinian repetition is a big thing for him). Furthermore, as the procedure described above melts these poems down into their respective ‘genetic codes,’ my ‘writing-through’ of the results ensures that my ‘genes’ are also part of the mix. So in that sense these poems can be thought of as the mutant offspring of this messy encounter between myself, Yi Sang, and Google.

This is, of course, a ‘procedural’ project, but ultimately I’m uninterested in a ‘clean’ conceptualism that follows a constraint or generative procedure ‘faithfully’ to the letter. I take whatever I like from the randomly generated list and shape the results according to my needs.  This part of the process is therefore not subject to any particular procedural constraints, though how the poems turn out is naturally somewhat ‘determined’ by the results of the random list generation: I ‘write-through’ from top to bottom, and what is on top and what is on bottom vary randomly with each iteration.

In that sense I see these poems as representing a ‘mixed’ procedural approach that marries procedural surrender with actively applied artistic intention. So in a way this is a ‘cybernetic’ approach to procedural writing, one that isn’t shy about putting a lot of ghost into the machine and in exploring the uncanny quandary to which this enmeshment of ghost and machine leads us. One description of this quandary that I particularly like appears in Donna Harraway’s Cyborg Manifesto:

Pre-cybernetic machines could be haunted; there was always the spectre of the ghost in the machine…Late twentieth century machines have made thoroughly ambiguous the difference between natural and artificial, mind and body, self-developing and externally designed, and many other distinctions that used to apply to organisms and machines. Our machines are disturbingly lively, and we ourselves frighteningly inert (152).

So entangling myself with the procedural ‘machine,’ I come to see myself as both more and less than myself, the machine as both more and less than machine. This simultaneous expansion and hollowing-out is, of course, characteristic of all intimacy; it also characterizes what it feels like to translate another poet. This is especially the case when one translates a poet with whom one feels some particular affinity or in whom one recognizes some special resemblance. Baudelaire, when asked about his interest in translating Poe, said something to the effect of ‘he resembles me.’ That was my experience in first encountering Yi Sang.