Over the years I have conducted a number of studies of task-based teaching. One of my earliest studies (Nobuyoshi & Ellis, 1993) used a narrative task to investigate what eff ect prompting learners to self-correct their past tense errors had on their accuracy in the use of this tense in a subsequent narrative task. The study was motivated by Swain’s (1985) claims about the importance of ‘pushed output’ for language learning. It pre-dated Lyster’s later research (Lyster, 2004) on prompts. Subsequently, I turned my attention to input-based tasks – in part because I wanted to conduct a classroom-based study with students who had very limited speaking abilities and in part because input-based tasks were convenient for investigating Long’s (1983) Interaction Hypothesis. In Ellis et al. (1994) we used a listen-and-do task where students had to listen to the teacher’s description of where diff erent objects were located and note their location on a picture of a kitchen. We found that when learners had the opportunity to request clarifi cation if they could not understand the teacher’s description, they both performed the task better (i.e. they were successful in locating the objects in the kitchen picture) and they also learned more of the diffi cult words we had embedded in the description than if they had no opportunity to negotiate their understanding. Later, in Ellis and He (1999), we used another listen-and-do task (the students listened to descriptions of where to locate diff erent objects in a diagram of an apartment) to investigate the relative eff ects of simplifi ed input, interactionally modifi ed input and pushed output, reporting that the output condition led to more new words being acquired. These (and other studies) are described and discussed in Chapter 3.