To investigate the use, or otherwise, of conversational adjustments (CAs), in a normal instructional setting, the quality of speech generated during meaning negotiation, and learner perception of the task under study, a quantitative and a qualitative analysis was carried out of language produced in a dyadic set-up in a one-way information task, a two-way information task and a decision-making task. The study revealed that the quantitative analysis supported the results usually found in Second Language Acquisition (SLA) studies: that the use of CAs generated by the 10 dyads in the two-way communication task was indeed significantly higher than in the one-way task and the decision-making task. However, while the overall group's quantitative behaviour confirms the use of CAs, the qualitative analysis shows that the individuals' performance in their use differs widely within and across task types. Learner idiosyncrasy is therefore perhaps a far more important phenomenon than commonality, and learners' perception of the task may be much more relevant than its logical construction. Both aspects, learner idiosyncrasy and learner perception of tasks, suggest that predictability will be a perennial problem, not just a temporary technical one that some imaginative task designers will soon resolve. This calls into question the desirability of output predictability as an aim for task-based materials design. The report ends with further emphasis on the value of including learners in the development of understanding of classroom life.