Context is vitally important for how we understand the world around us, and this is particularly so with language. We know that humans use a range of linguistic cues when understanding language to help tailor their expectations about upcoming linguistic material, contributing to the seamless nature of comprehension. Linguistic cues fall into two different levels of context – the local, sentence context and the wider, global context. Some linguistic cues that are considered to be global are knowledge about the world or the speaker. These have been found to facilitate semantic processing when they align with linguistic expectations through faster reading times of predictable content. One other global contextual cue that may be of use to how we represent meaning and understand the world around us is the coherence of a discourse. The current, pre-registered study investigates the interaction between discourse coherence and word predictability during language comprehension. To do so, we used an online, self-paced reading paradigm and had participants read three-sentence discourses that differed on the coherence of the final target sentence with the first two preamble sentences, and also on the predictability of a critical word within the target sentence. From our first experiment, we found that people were sensitive to the coherence of the overall three-sentence narrative, with slower reading times for trials that had less coherent preamble contexts. After exploration of our data, we pre-registered two further studies to investigate whether this effect is still present in more extreme experimental settings. For Experiment 2, we altered the ratio of less and more coherent contexts to see if people still make use of discourse coherence as a linguistic cue for informing their expectations when there was a greater proportion of more coherent trials. We found this to be the case. In Experiment 3, we replaced our less predictable critical words that were semantically relevant to the overall message of the target sentence with completely anomalous words. Here, we found people were faster to read the highly predictable critical words and slower to read the anomalous critical words when they first read more coherent contexts, but not when they first read less coherent contexts. This suggests that people are able to use relevant linguistic cues from both levels of context, and do so flexibly depending on the degree of contextual support at the global discourse level.