Text Remix has been incredibly valuable for the development of my project. Holding code and creativity together in any context is a delight for me, but specifically dealing with textual creative work both as a suitable input for code and as its intended output was especially exciting. My project does not directly generate creative work through algorithmic processes, but it does use natural language processing in a creative exploration of words and how code can help us to step outside our assumptions about what they’re capable of signifying.
I’ve also been able to work some practical skills development into my group project for Text Remix. My team is attempting to procedurally generate poetry in the style of E.E. Cummings to understand what sort of patterns within his writing can and can’t be effectively implemented in code. We’re following a few different possibilities, from line-by-line and word-by-word randomization to part-of-speech tagging and substitution; however, the part of the project with which I am most involved is fine-tuning a causal language model based on some 20,000 lines of his work. I have not yet finished training the model, but the level of success I have with this project should inform my likelihood of success for my intended knowledge integration project, give me a deeper understanding of the processes and skillsets needed, and help me understand how much data is really necessary for this kind of training.
I may come back to this post and update it when my Text Remix project is complete, because it is certainly the closest project in terms of ethos, output, and requisite skills to the work I hope to be doing throughout the rest of the year.
