My experience showed me that the digital assessment field is more focused on item building and test delivering than correction because correction can be automatic. But human correction has still an enormous value and a reduced conception of corrections that avoids it limits the strength of digital assessment. I wanted to explore what could be a sophisticated digital correction tool and how this kind of tool could help students and teachers. Because of the breakthrough of AI, correction will be at the center of deep learning logic. Can we anticipate that ? Some fields must be considered : supports that need human correction could be text, image, sound, video… some concepts start to emerge : the intention of correction and the focused points, sensitive text, the error modelisation, granularity, the reporting and the instantaneous analysis, the effect of correction : preconizations. And also how to integrate these tools in a deep learning process in AI perspectives.
This first version of correction platform was developped for DEPP (French Ministery of Education). It is able to load files (images, text, audio or video) that reflect the student’s performance. It can attribute corrections to correctors and can monitor the correction process. Once a correction task is created, the type of correction define the correction tool interface.
Modules of corrections
The interface of correction is a module. It can be plugged in the platform. Modules must be optimized to reduce the time of correction to some clicks. They must be adapted in term of support (media, text, images), subject, and granularity
High granularity correction
This module example show a high precision correction level to estimate out loud reading. It is using the concept of sensitive text, all elements: words, punctuation, spaces, lines can receive observations (badges) and comments. A categorisation of errors can identify the type of difficulties and a report system gathers all corrector’s infos and build preconizations.