When we hear conversation in unfamiliar languages, it can seem fast, with sounds run together and few distinct words. In fact, sounds flow rapidly and overlap in all spoken languages, but knowledge of our own language allows us to use various unconscious strategies to extract words and to interact in fluently-timed conversations. One key strategy for effective spoken interaction is listeners' generation of timing predictions. Sounds are often lengthened at important points in speech, such as the start of words and the end of conversational turns. To detect lengthened sounds, it seems that listeners use the rate of the speech they have already heard to generate expectations about how long upcoming sounds will be. When sounds are longer than expected, listeners can interpret that as an important point in the speech stream, such as the start of a new word. Although there is good evidence that listeners make timing predictions to interpret speech, theoretical understanding of how this is achieved is very limited. In particular, it is unclear what features of speech support listeners' use of timing prediction. For example, one theory about language processing in the brain implies that very regularly-timed speech is more useful for listeners in making timing prediction, but another theory implies that the irregularly-timed flow of natural speech supports timing predictions. These theories also have different implications for understanding how timing predictions are affected by acquired disorders, such as in people with aphasia (PWA) due to stroke. Aphasia typically arises due to interruption to blood-flow in the brain which causes damage in language-processing areas. PWA usually have difficulties producing spoken language, such as specific words or grammatical phrases. PWA can also have problems in understanding speech, but these may be less immediately apparent, despite having a profound impact. The project will test both people with typical language skills and people with aphasia in order to improve understanding of how they produce timing predictions. As these are so important for natural interactions, our findings will have implications for the design of devices using artificially-generated speech, from satellite navigation interfaces to augmentative communication systems for people with speech difficulties. We will test how timing predictions are generated using a new listening task ("nonword segmentation") in which listeners hear short meaningless sequences of syllables, "nonword targets" (e.g., "libeku") followed by longer sequence of syllables, "carrier utterances" (e.g., "mimasikonebubilibekududi"). When listeners hear a nonword target in a carrier, they have to respond by pressing a computer key as quickly as possible. Over multiple trials, with careful variation in targets and carriers, this task will allow us to build up a picture of timing prediction. Importantly, our pilot studies have shown that longer initial consonants make target detection easier, but only when targets are quite late in carriers. This supports the theory that listeners build up timing predictions based on speech already heard, but to fully test this, we need to explore how timing predictions are affected by a range of factors: - Regular or irregular carrier utterance timing. - Hearing speech in noise and/or with meaningful linguistic content. - Hearing familiar voices and accents. - Seeing as well as hearing speakers. Because PWA have variable speech production and perception difficulties, we will test their ability to make timing predictions compared to age-match listeners without aphasia. One theory implies PWA should have relatively good timing predictions compared to their overall language, but another theory implies poor timing prediction in PWA. Ultimately such work will boost understanding of speech perception and comprehension can be affected in aphasia and how any difficulties may be remediated.