So it's a full-sentence auto-completer? Interesting; would be nice if it would wait for a whole word (or at least a second without typing) before completing though
It seems to be buggy? There were 9 JavaScript errors in the process of typing, "How do I compare thee to a summer's day?" (I know that's not the original text of the sonnet.)
Yeah, only listing the top 10 chars to write next results in a few misses. Its also bit of a pain to test a recursive function that uses random numbers.
I can understand that, but in a web app like yours which is meant to be very responsive, I think you would be better served by handling misses rather than letting errors throw away the stack (so to speak) every time a miss happens.
Gibberish I guess. Doing a lookahead for letters was how I started, and its kinda interesting to see how it turned out as a quick project. Storing the probabilities for each letter given a string prefix is the limiting factor, and took a bit of effort to get it to reasonable size for a browser. I think using words instead of letters may result in better sentences with the same amount of space though...
I think this is a really neat side project. It's almost like a chatbot where you drive. If you can get sentence structure working by doing intelligent stuff with the tokens it could be even cooler. What do you do fulltime?
I build web trading apps for banks. Big, big javascript apps that stream prices and handle trades. Not the kind of apps that require shakespearean spam though...
What is it supposed to do? I try to type a sentence but it is immediately garbled by random words. If there were two boxes, one for input and output, things wouldn't get mixed up, or perhaps a "Shakespearize" button. Right now the edit box is not really usable because the script interferes with normal editing.
EDIT: Another note, while the references to Shannon & Markov chain are correct, in NLP this would more commonly be referred to as a character n-gram model.
It's meant to interfere with your writing, like a backseat driver interferes with your driving. Your idea of separation is interesting, do you see it as a kind of translator?
I imagine a backseat driver shouts suggestions, but does not take over the wheel; like the way autocomplete is there without preventing you from typing something else. It doesn't have to be a translator, but there should be some distinction between user input and what comes from the model, and the user should have the final say.
16 comments
[ 1.9 ms ] story [ 41.0 ms ] threadEDIT: Another note, while the references to Shannon & Markov chain are correct, in NLP this would more commonly be referred to as a character n-gram model.