(If you aren't already familiar with recurrent neural networks, why not see Andrej Karpathy's excellent blog?)
(If you aren't already familiar with recurrent neural networks, why not see Andrej Karpathy's excellent blog?)
These days, thanks to The Wonders of Science[TM], we can train neural networks to imitate different styles of text by showing them some examples. Often the results are gibberish, but occasionally in this gibberish there is a nugget of... less gibberish. There are many fine Python libraries out there to let one run RNN experiments: I am using textgenrnn, and fine-tuning its stock model on data of my own whimsical fancy. Here is a selection of the most interesting, perplexing, or otherwise notable outputs.
This was my favourite experiment to date: I compiled a list of ~500 English words that amused me, from various web lists and acquaintances' suggestions, like "crapulence", "dingus", and "merkin". I gave a neural network a good long train on this list, and asked it to generate some fresh examples. One striking thing about the raw results is that the network rarely produced anything that wasn't at least semi-plausible: it learned quite well which combinations of letters were likely to follow others, consonants and vowels in the appropriate places, etc. Almost everything was at least pronounceable! And the examples, by and large, had that ineffable quality of "amusingness" on the tongue. There is some property of the training list coherent enough for this technique to capture...
One note: now, multiple different networks all trained up on totally different set of phrases have independently and repeatedly produced the word "sluggle" or its variant "sluggled". I... have no idea what this means.
Sometimes by pure chance, the network generates a real word that really ought to be considered silly, but was not part of its training data:
Others are fake but plausible:
My personal favourites have to be: