Listeners are able to use previously acquired knowledge to predict the content of upcoming speech when the latter signal is distorted or degraded. The present study investigated this perceptual “pop-out” phenomenon. Previous research has shown that misperceptions (i.e., prediction errors) of single words can occur when there is near-identical overlap in phonological onset (cohort) and offset (rhyme) between prior knowledge and distorted speech with relative preservation of the temporal envelope. Here we investigated whether misperceptions during pop-out can also occur at the sentence level with spectrotemporally degraded speech that drastically reduces phoneme recognition and removes prosodic features of the acoustic-phonetic signal such as intonation and lexical tone. In a series of experiments using a probe-prime-probe design with congruent, incongruent, and neutral conditions, we presented participants with three different types of spectrotemporally degraded probe sentences: real English, nonsense (containing phonological neighbour words but semantically empty), and pseudo sentences (containing nonwords). In the nonsense and pseudo-sentence experiments, words and nonwords were phonotactically matched to the reference words using homophonic transformation. For real sentences, we observed accuracy rates of 95% in the congruent condition. Crucially, “accuracy” rates for congruent nonsense and pseudo-sentences were also high (70% and 74%, respectively), indicating participants were able to assemble lexical information from the context of the prime sentence producing ‘misperceptions’ of the probe sentence. These findings show that prior knowledge of sublexical information is largely responsible for listeners being able to predict the content of spectrotemporally degraded speech.