Arai et al. (2007) showed that structural priming in the comprehension of English dative sentences only occurred when the verb was repeated between prime and target, suggesting a lexically-dependent mechanism of structure prediction. However, a recent study in Mandarin comprehension found abstract (verb-independent) structural priming and such priming was stronger when the structure was unexpected (e.g., when a verb biased toward the double object [DO] structure is followed by an unexpected prepositional object [PO] structure; Chen et al., 2022). The latter finding of inverse preference priming is consistent with error-based implicit learning accounts, which suggest structural priming is based on learning from prediction errors (Chang et al., 2006). Here we tested the mechanism of structure prediction (lexically-dependent vs. abstract) in four visual-world comprehension experiments in Dutch. Dutch is a Germanic language like English; it is biased toward the PO structure like Mandarin. We not only found structural priming when the verb was repeated, but also when the verb was different: During target sentence processing, comprehenders looked more often at the recipient (predicting a DO structure) than at the theme (predicting a PO structure) after a DO prime and vice versa after a PO prime. Importantly, abstract structural priming only occurred when the target structure was relatively unpredictable. We interpret the inconsistent findings across languages in terms of an effortful process of structure prediction in comprehension (Pickering & Gambi, 2018): it occurs when it is needed to disambiguate the postverbal arguments, but not when it is optional.