Just thought I'd mention an idea seen today in the Tool Box Journal by Jost Zetzsche.
Results differ between neural and statistical Google's MT engine, and apparently, in some cases/language pairs, choosing stastistical over neural engine could be beneficial.
Maybe a checkbox can be added in the preferences, defaulting to neural MT?
You can specify which model to use for translation by using the model query parameter. Specify base to use the PBMT model, and nmt to use the NMT model.
Is a use of the statistical model really still useful? What is the advantage of using the not so good MT engine?
Anyway, I already wrote Jost that CT uses the neural model for about half a year now (trendsetter galore!).