Theories of predictive perception posit that perception is the process of inferring the causes of our sensory input by combining that input with predictions derived from our learned and evolved model of the world. It has been suggested that such predictions play a key role in language processing. However, robust neurophysiological evidence of such processes remains elusive – particularly in the context of natural, continuous speech perception. Previous research has suggested that the auditory encoding of words in natural speech is influenced by how semantically related those words are to their preceding context. However, it is unclear whether this effect is driven by predictions per se rather than dynamic modulations of attention. The present study aims to adjudicate between these two alternatives by recording EEG from healthy, neurotypical adults (N=4, so far) as they listen to naturalistic audiobook stimuli. Crucially, specific words within the audiobook were modified to alter their surprisal, while preserving the level of constraint provided by their preceding context. This allows us to disambiguate between attention and prediction error (as indexed by surprisal) as the driving influence on the lower-level sensory encoding. Our preliminary results suggest that dynamic word-to-word fluctuations in top-down attention influence the sensory encoding of natural speech. While these findings do not rule out a parallel role for prediction in natural language comprehension, they do suggest a need to be careful in distinguishing between the influence of top-down attention and top-down prediction in the prelexical perception of speech.