only keep image descriptions with > 3 words

[humenda]

Enhance the JSON AST parser so that image descriptions are ignored with less
than 3 words. Background is that often, images are not properly described, but
only a replacement word is inserted. E.g. "infobox" does not really describe an
image and therefore does not provide any context to the AI algorithm.

Descriptions with more than 3 words are likely to be a proper description.

[humenda]

closed. Wrong repository; see humenda/CRAFT