Hi Rene,
I guess we talk about a one-in-a-million problem, and this will always be complicated to solve. And hiring a programmer for those special kinds of tasks may not relativise the benefit of ML for all other tasks which cover the nomal 95% of stuff we normally do, as soco-psychologists.

Have you ever tried to use a IM and and an (old-school) confederate, I think this would be the easiest approach.

All the best,
Thomas