Unicode's bidirectional algorithm often fails where there are RTL characters, LTR characters and neutrals such as punctuation in the same paragraph. Often this can be fixed by liberal sprinkling of either the RLM character (in base RTL text) or the LTR character (in base LTR text).

Putting these characters directly into the article text makes such changes difficult to review and edit, since they are invisible in the edit box in major browsers. A better solution is to use HTML's ‎ and ‏ character entities.

By happy coincidence, ‎ has roughly the same effect in the edit box as it does in display, because the latin characters "lrm" are of strong left-to-right type, just like the control character they represent. The same is not so for ‏, meaning that in cases where ‏ is used, the text remains broken on edit while being fixed on display. Here's an example:

http://he.wikipedia.org/wiki/ACID

What I propose is that someone should come up with a translation of "rlm" into Hebrew, Arabic or both, and that we should implement this artificial character entity in the MediaWiki parser.

-- Tim Starling