Unfortunately, this is always a challenge. It gets into the "affinity" a character has for LTR or RTL text. For example, Latin characters are LTR; Arabic characters are RTL. However, numbers will attach in either way based on surroundings and markup. It's seriously complex. We could potentially add lrm/rlm markers, but I'm not sure how we'd get away with that as we have no way to know/guarantee what direction is coming. So I'm going to call this a won't fix - in a full RTL translation with RTL titles, it should be ok.
In Google+, they did really clever job. Since, they have a huge language model, hence, they recognize the language of the posted phrase and if it falls under RTL languages then they apply CSS properties accordingly.
Only if you can build your own AI Language Classifier using your own infrastructure or with the help of Google Prediction API. Also, you can use Language Detect feature of Google Translate. Mind that Google Translate API will be shut down by the end of this year. But, you still have first two options.