לרט - loyt

[mirjam-amsterdam]

I think, all לרט could be exchanged for לויט

by the way, some of the results are rather badly rendered.

[urieli]

There are many other such pairs. For example, the vast majority of "רי" should be "די".

One solution would be adding more information while performing the OCR itself, for example encouraging more common words. However, for this to work, the common word would have to be found in Jochre's "beam" of possible OCR analyses for a given group of shapes.

Another would be to apply certain post-OCR corrections, either systematically, or using a more sophisticated correction method via machine learning or textual analysis.