Extensive research has shown converging evidence for prediction in the context of sentence comprehension tasks. While there is little debate that predictive processing is happening during language comprehension, precisely how we are able to generate these predictions is relatively unknown. The current project explores the cognitive factors supporting accurate prediction performance, focusing on the role of verbal working memory. The project takes a case study approach, examining the ability to predict upcoming words in individuals with different types of verbal working deficits following brain damage. Evidence from brain damaged individuals has shown that verbal WM can be divided into separable semantic and phonological WM systems which can be separably damaged. Four aphasic individuals were recruited as participants, one with phonological WM deficit (TP), two with semantic WM deficit (WC, DW), and one with both a phonological and semantic deficit (KA). We hypothesized that semantic WM, but not phonological WM, is related to the ability to make predictions during a language processing task. Experiment 1 compared the aphasic participants’ the prediction performance measured through a sentence completion task to a group of control older adults. All 4 participants showed worse prediction performance than controls, suggesting both phonological and semantic WM are important for accurate prediction, contra our prediction. However, it is possible that deficits in the speed with which language can be processed also contributes to the ability to predict particular for those individuals with phonological WM deficits. Experiment 2 explored this possibility by slowing down the audio presentation rate in the sentence completion task with the same four brain-damaged participants. However, no clear effect of presentation rate on prediction performance was observed. What these results mean for the role of different kinds of verbal working memory in prediction will be discussed.