Some MT-engines are not able to handle tags correctly (e. g. Moses default and SDL LanguageCloud).

For these engines it makes sense to simply pass the tags to the engine in the format <1> (so as plain-text tags).

The MT engine is supposed to return it in the same way in a place that makes sense regarding the logic of the sentence structure. 

So when the result of the MT engine is returned, we can insert the tag with all necessary contents, where the plain text placeholder for the tag is placed by MT and deliver it like that to the front-end.