First Thoughts

The project will be centred around text:

• inventing textual constraints and using computation to produce texts;
• using computation and machine learning to explore the resulting space of possibilities;
• expand my knowledge of machine learning and other tools in the process. The texts are mostly going to be ‘poems’, or textual fragments, but the idea of producing prose, or simply longer textual objects, is one of my goals.

The movement pursued is twofold:

• building (a database of texts under a constraint);
• mining (the identification and selection of texts of aesthetic relevance).

This project would expand and coalesce three previous projects:

• my ongoing exploration of wordsquares;
• a similar project called wordlaces;
• an earlier concept that could be upgraded with my current advances, subwords.

Courses: Machine Learning:

• Rebecca Fiebrink’s course on Kadenze;
• Parag Mital’s course on TensorFlow on Kadenze, and the associated GitHub repo;
• The NLTK library, in Python, and an introduction book.
• The two courses by Standford on NLP with NLTK and Neural Networks.

Other:

• Develop some knowledge of cloud computing (through Amazon or Google);
• Try the google collaboratory for iPython. Many videos and online tutorials are on the table, but I should also find books and papers, on arXiv for instance.

Questions:

• There are many sorts of machine learning algorithms, and many varieties of neural networks. I read that recurrent neural nets (RNN) are useful for text generation, I should dig deeper into that.
• Another framework for text generation is to be found in Markov chains, and more generally the idea of considering the probability that one word follows another (each word considered as a ‘step’ or ‘node’ of the chain). Another important topic for research.