There were two training and one held-out (in boldface) categories per condition. Each training category was mapped to a different pronominal form (A or B), schematically represented with light and dark gray. Participants had to use the training forms they learned (A or B) to express the held-out meaning (cells with A or B?). The two rightmost columns state which of the compatible paradigms participants were predicted to infer under Harley and Ritter’s (2002, H&R) and Ackema and Neeleman’s (2018, A&N) accounts.