- Read time: 1 min[
Machine learning amazes me - and so does my brother in law Aaron Moe, professor of English, literature genius, and ecopoeticist. Whenever I hang out with Aaron, I learn something. Among many things, Aaron is an expert on E.E. Cummings and the works of Walt Whitman. One of our conversations led to an idea: what if we were to adapt some of the advances of leading machine learning engineers to mimic the prose of Whitman? You could feed the bot a “seed” of a few words, and then see how Walt might finish your sentence.
There has been quite a bit of work in this area, most notably the char-RNN work of Andrej Karpathy which has become a very popular algorithm. Char-RNN was originally trained on a corpus of Shakespeare’s works, learning character-by-character how to create snippets of text mimicking the Bard’s great works. Karpathy also trained the network on other sources of text of dramatically different form and function, including Latex code (which nearly compiled). The power of this approach is truly mind-boggling - with a relatively simple architecture, the machine can learn to generate unstructured text in any desired style provided a sufficiently large training set.
The idea to use this approach on Whitman’s work is in the early stages of development. Check back in here for updates, or shoot me a message - I’d love to chat about it, or other cool new ideas or potential collaborations!