[tug-summer-of-code] [Fwd: Re: A couple of project proposals]

Jonathan Fine jfine at pytex.org
Tue Feb 17 12:35:26 CET 2009

Karl Berry wrote:
>     True, no one has asked specifically for a symbol-search tool based on
>     text recognition, but I still believe that such a tool would be great
>     for the "my professor drew this symbol on the blackboard, and I want
>     to typeset it in LaTeX" crowd.
> FWIW, when talking with mathematicians, I have been asked several times
> "what's the LaTeX for this symbol" (and they draw it on a piece of paper).
> All I can do is point them to the symbol list.
> So I think having an interactive tool would be very nice, and it should
> go on the list.

The key to creating an interactive tool for finding a symbol from the
list is to create and suitable metadata about the symbols.  The user can
then search on the basis of that metadata.

This is precisely how the KanjiPad tool works.  The metadata is the
standard strokes for drawing the character, and the drawing tool is a
convenient (and fun) way for the user to provide the metadata.

For the LaTeX symbols there are many other types of metadata that could
be used.  For example:
*  What subject area (analysis, group theory, logic, ...)
*  Unary operator or Binary Operator or Constant
*  Is it a letter
*  Does it look like ...
*  Have I used it before

I've seen handwriting recognition for mathematics, as part of Microsoft
Word 2008 as I recall.  It sometimes guesses right.

I'm all in favour of an interactive tool.  And there's no reason not to
use both methods at the same time.  This is like training the
handwriting recognition program.

I'm fairly confident the project will be able to produce something
useful, based on the 'conceptual' metadata.  I have no such confidence
for the handwriting recognition approach.

Here's a link to what may be the state of the art.

You might also want to look at


Here's the search I used.

I suggest we consult some of these people, and ask for their opinion.

BTW, KanjiPad has not got back to us, and the email to JavaDict bounced.


More information about the summer-of-code mailing list