Episodic future thinking refers to a human cognitive process which generates successive predictions of events that are likely to occur in a cue-specific context in the future. An emerging view is that semantic memory as well as episodic memory contributes to this process, but the exact mechanism remains unclear. We built a computational model that learned to predict the next event upon a presented event (sequence prediction model). After learning the statistical structure in the training sequence, the model was tested for generating successive self-predictions of events triggered by a cue. The generated sequence of events captured some phenomenological features of patients with semantic dementia when the semantic system of the model was damaged. The role of semantics in episodic future thinking and the usefulness of a sequence prediction model are discussed.