The problem with that approach is that words might be back-translated to the original.

A better way might be to generate the text in Language1 and use a script to split the sentences into two arrays: Array1 containing sentences 1, 3 , 5.. and Array2 containing the sentences 2, 4, 6... . Array1 could then be machine translated from Language1 to Language2 and then to the target language, while Array2 is translated from Language1 to Language3 and then to the target language. These two arrays can be combined to produce the final output text.