It's useful to be able to constrain inference to by a set of user-provided articulations, as a way to resolve conflicts and provide a third opinion when neither source gives the correct articulation.
I suspect this can be done as a sequence of two two-way merges.