Peer assessment is a promising solution for scaling up the grading of a large number of submissions. The reliability of evaluations is one of the critical issues in peer assessment; several probabilistic models have been proposed for obtaining reliable grades from peers. Peer correction is a similar framework, in which students are instructed to correct the errors in submissions from other students. "Peer correction" is typically performed simultaneously with peer assessment; a reviewer is instructed to correct the errors in a submission and to provide a grade to it. We observe the occasional inconsistency between a grade and the correction; for example, a reviewer provides a high grade for a submission but she corrects many errors in it. Such inconsistencies can point to unreliable reviewers. In this paper, we propose probabilistic models for peer correction, and the combination of the peer correction models and the existing peer assessment models for capturing the inconsistency to accurately estimate the reviewer reliability and the student ability. We conduct experiments using the dataset of an actual peer correction platform for language translation, and the results demonstrate that the combination of peer correction models and peer assessment models improves the accuracy of the student ability estimation. [For the full proceedings, see ED599096.]