Formula recognition - what should be stored, how and how to deal with partial knowledge and fusion of answers

[MartinThoma]

Symbol recognition is simple. Either you recognized the correct symbol or not. But for formula recognition things are different.

• A classifier can get a subset correct
• A classifier can get the correct segmentation, but the wrong symbols
• A classifier can get the geometry (partially) correct

I would like to keep a simple page where users can see what they wanted to know. For example, they wrote

Then a segmentizer recognizes

As you can see, the segmentizer got it almost correct. Only the = was segmented in two symbols instead of one.

Now, if a user writes

\sum_{n=1}^\infty \frac{n}{2} + \frac{n}{4} \alpha

and another user writes

\sum_{n=1}^\infty \frac{n^2}{2} + \frac{n}{4} \cdot \alpha

There should be a way to mark this as partially correct. Users should also have the possbility to only classify parts of a formula. For example, they might not know \infty but the rest.