Verbs show different statistical preferences for syntactic structures. An influential theory for how verb-syntax links are learned suggests that learning is based on, and proportional to, prediction error. However, the evidence is mixed and there is need for evidence from a diverse set of paradigms. We exposed 90 college-aged adults to an artificial language containing novel verbs and sentence structures and tested their production of utterances in the new language. We reversed verb-syntax links from one training block to another and found that participants were able to learn the reversal, as seen in a production test. Pupillometry detected surprise when participants heard sentences in the second training block that had the opposite verb-syntax links than in the first training block. Despite detecting both surprise and successful learning, we did not find an association between the two. Thus, we did not find evidence that learning was based on surprise. We discuss alternative learning mechanisms that can help language users adapt their language based on the input.