language exercises, hybridization of "fill the blanks" and "drag the words".

On my German language webseite ondaz.de, I have created an exercise where people can practice several things at once:

1. the use of temporal propositions vor and seit (ago / since)
2. the use of the perfect tense in connection with the former 
3. the correct formation of the perfect tense
4. permissable variations in the order of sentence elements.

The model example is like this:
mein Sohn / schwimmen / vor drei Jahren / lernen
a) Mein Sohn | hat | vor drei Jahren | schwimmen | gelernt.
b) Vor drei Jahren | hat | mein Sohn | schwimmen | gelernt. 
c) Schwimmen | hat | mein Sohn | vor drei Jahren | gelernt. 

If only variations of the sentence order were to be practiced, this would be a "drag the words" exercise. However, in order to practice the other aspects, it becomes a "fill the blanks" exercise coded like this:
*Mein Sohn/Vor drei Jahren/Schwimmen* *hat* *vor drei Jahren/mein Sohn* *schwimmen/vor drei Jahren* *gelernt*.

However, with "fill the blank" I can only define which elements are correct in which place but not the interdependence of their positions. Most strikingly, using the same element twice would be recognized as correct by H5P. Excample:

[Mein Sohn] [hat] [mein Sohn] [vor drei Jahren] [gelernt].

This is not a correct solution and makes no sense.

Of course it would be possible to code like this:

*Mein Sohn hat vor drei Jahren schwimmen gelernt/Vor drei Jahren hat mein Sohn schwimmen gelernt/Schwimmen hat mein Sohn vor drei Jahren gelernt*.
However, this requires users to compse the whole sentence without mistake at once. I does not give them immediate feedback on single elements inserted correctly.

So what would be needed is a version of "fill the blanks" with predefined elements and predefined order of them like in "drag the words* which however must be composed by users on the keyboard.