Language comprehenders constantly anticipate what is coming next. Yet many psycholinguistic studies and computational models assume that such predictions can only be generated once a word has been fully identified and all of its internal information is available. In this talk, I will present evidence from eye-tracking experiments that challenge this assumption. The findings show that comprehenders can exploit word-internal information, including morphological cues, as soon as they become available, and use them to anticipate upcoming words in real time. I will then demonstrate how these incremental predictions can be computationally formalized within cue-based retrieval models of sentence processing. Together, the findings suggest that prediction is sensitive to the internal structure of words and highlight the importance of incorporating morphological information into computational accounts of real-time sentence comprehension.