RE: Humanize your bot! - Markov chains
🤔 Doing it on the word level doesn't give great results when strictly using a Markov chain because of course it's memoryless.
An easy way to see that is by tapping one of the random three suggestions on top of the Android keyboard, I'd bet they use that. Here's one I made earlier (edited very lightly to remove personal info, of course it's "learns" from what you type):
I am she's very good quite European as it appears there only thing with the same as it is it's very small to be sure it'll be a few years ago the last two days and nights in a few years ago the same as last year and I will be quality to it because it has a lot too playing here has to stay on liquids is a cryptocurrency is so have you ever went they were rehearsing.
I think the real Laura uses full sentences, parts there of phrases or at least some knowledge of syntax, somehow. 😜
You kind of nailed it, she does, her sentencing is backed up by a neural path "scoring" that get a (+1) or (-1) usage modification every time she "reads" a sentence structure or pattern. This allows her to use "common" language without disregarding more elaborate phrasings.
I thought I clarified, this is a function, part of her tools... Not her whole repertoire of weaponry :p
Ah I didn't pay attention to that qualification 😳 Oops! So thanks, maybe next post on neural path scoring then! Interesting topic.
It can be applied also to music and the exact same issues arise. "Authenticity" of the piece, can you tell the difference, what does it mean if you can / can't, is it really music, etc. etc. 🤖 🎶