Abstract
This paper presents the analysis on a corpus of learner data involved in a language competence test in three languages: Italian, German and English. The focus is on the paired task in which two candidates discuss together their opinion and defend their position against it, and on the sequences of repair initiated within the deployment of this task. Despite the expectations the only three types of repair found are reformulation, false-starts and word-searches. Results show that repairables in the data are mainly self-repaired and do not lead to the production of sequences of meaning negotiation. The paper discusses the instances of repair management in the data and compares differences and similarities among the three languages taken into consideration.