EXPERIMENT LOG — Chinese poetry machine learning model

Aim

Experiment with machine learning and artificial intelligence processes in the generation of poetry using Runway.

Precedents / context

Following Andrew's feedback to look into these methods as a way to build on Tracery content.  

Process / methods

• Full compilation of inputs and outputs, from a machine learning model trained specifically on Chinese poetry. The outputs were a mixture of Chinese (mostly) and English, which I've put through google translate and copied into here. 
• https://docs.google.com/document/d/1o1vlC7w48b75XqtXz799Ql9g0PNyH8ROt_gSSRNKspk/edit?usp=sharing

Reflection on action

• I was so surprised to have come across this model. It was really exciting to have a readymade process that I could generate content with, that fit really well with the aims of my project. It never occurred to me to look into machine learning, but this was shockingly easy and the outputs are really compelling. 

• I think the google translated versions definitely read as Chinese poems, which is pretty amazing. The content, structure, and tone are all very evocative of the poems and stories I remember learning as a kid. I genuinely can't get over it. The many references to nature and days/time, colours, animals, etc. are such a feature of traditional Chinese poems. The kind of philosophical tone, circular logic/phrases and questions. This is all immediately familiar, but I would have never thought to even research actual Chinese poetry — when I was trying to bring in cultural references through specific words and objects, they were, and read as, surface-level add-ons, whereas this is completely different structurally. It's definitely not something I could've written, either by myself or even something like Tracery. 

• However, I wasn't sure what to do with it. It's not anchored to anything in the project (i.e. it's not like the Tracery stuff where I had more control over the output with specific words), and I wasn't sure if it was where I wanted to go. I think now though, after a week, and revisiting the experiment after Sharon's feedback, if I want to work with words and poetry specifically, this is SO interesting. It'll need to go through a lot of synthesising, editing and rewriting, particularly as I work out what is happening on this 'stage'. I think the key, especially here in semester two, is to work to a purpose. 

Reflection for action

• START mapping out this experience !! What purpose does the poetry serve? Will it be performed, on the screen, part of the code? Start thinking about how you might document this process too — perhaps a zine or some kind of publication? 

• Figure the bounds, and layout of the project, and then dig into, cut together these + the tracery experiments towards A PURPOSE. 

References