Voice mismatches may arise in VP ellipsis and Pseudogapping between source and target. No such mismatch has been reported for Right Node Raising (RNR).
However, French corpora provide many examples, where the shared participle must be interpreted as active (reflexive) and passive. Since RNR requires semantic contrast between the conjuncts, we conducted two experiments to test semantic (agent) contrast and syntactic (voice) match:
+contrast/-match: Ce pharmacien doit des explications à ceux qui se sont *ou qui ont été* mobilisés pour lui. / This pharmacist owes explanations to those who themselves have or who have been mobilized for him.
-contrast/-match: qu'on a *ou qui ont été* / who one has or who have been
+contrast/+match: qui se sont *ou qu'on a* / who themselves have or who one has
-contrast/+match: qui étaient *ou qui ont été* / who were or who have been
Acceptability judgements only showed a significant effect of contrast (p<.01), not of match. In an eyetracking reading experiment, Gaze and Regression Path Durations show significant effects of semantic contrast in both voice match and mismatch sentences (critical zone between *) and no effect of match. Hence, given the appropriate semantics, voice mismatch is acceptable in RNR.