On Jul 13, 8:06 am, tejjyid <
andrew.x.w...@gmail.com> wrote:
> You can find my attempt at color coding in the "demo*" tiddlers.
> Definitely the basic code is too dense for any typical presentation
> purpose; I think it's more intended to illustrate the POS tagger results
> (which are quite flawed in some ways).
I did read a bit about POST algorithms at wikipedia [1-2]. Also some
other stuff (can't find the links), where they listed the possible
"accuracy". It seems to be a reasonable algorithm + dictionary if they
are above 90%. Depending on the dictionary (size and "training") it
goes up to 95%+. But for the good ones, you have to "insert coins" :)
At several places, the "wordnet"[3] is mentioned. The natural project,
I mentioned in my last post, has an interface to wordnet. All this
stuff needs big files/dictionaries to work well. ..... So a browser
only approach may have limited success :/
[1]
http://en.wikipedia.org/wiki/Part-of-speech_tagging
[2]
http://en.wikipedia.org/wiki/Word-sense_disambiguation
[3]
http://wordnet.princeton.edu/
> Anyway, for teaching purposes I will pretty much always be contrasting 2 or
> 3 features, because more is confusing for students. Also, I'll be talking
> to my classes about ways they think they can use the tool. But for 2 or 3
> features, I think color works fine. You can see this at
> andrewsimon.tiddlyspace.
The explanations, are extremely interesting. I did find the "demo*"
stuff again :) IMO there should be a possibility to get better
formatting. ...
> The 2-line approach seems the best way for maximum detail/clarity. I might
> combine that with inline table editing as a way of manually fixing the
> errors, where 100% accuracy is important, but that has other implications
> (as in, a loss of dynamism). so might remins the operative word.
You are right. 100% accruacy will not be possible with software only.
Which is a good thing too, because it means, our brain can't be
replaced that easy :))
have fun!
mario