Antilocality effects provide strong evidence for expectation-based sentence parsing models. Previous discussion of the antilocality effect, however, largely focused on the argument-verb dependencies in verb-final constructions, for which a memory retrieval-based account has been argued to be equally adequate. To test whether the principles of expectation/memory-based accounts hold for a wider range of dependencies, we report on two self-paced reading experiments that compared two different determiners in German: the morphologically complex determiner "derjenige" 'the-jenig,' which obligatorily requires a relative clause, and the bare determiner "der" 'the,' which does not trigger such expectations. The first experiment did not show the expected antilocality effect, but the reliability of our results was restricted by the experiment's low statistical power. In a large-scale second experiment we addressed confounds in the design of Experiment 1 and found evidence for an antilocality effect with the complex determiner. As the antilocality effect found in our study does not involve argument-verb dependencies, the memory-based account cannot be extended to the current case. Thus, our findings provide novel empirical support for the expectation-based antilocality effect. At the same time, the experiment attests to a processing cost in later sentence regions, hinting that memory- and expectation-based effects can co-occur within the same structure.