WordReference Corpus is a large learner corpus of four languages: English, Spanish, French, Italian. It contains ca. 170M tokens of natural written production (forum messages at WR Language Forums) in these four languages by both native and non-native speakers.
The corpus can be downloaded at Språkbanken's website. A set of useful tools can be found on this Github repo.
Please cite this paper when using the corpus: Berdicevskis, Aleksandrs. 2020. Foreigner-directed speech is simpler than native-directed: Evidence from social media. In Proceedings of the Second Workshop on NLP and Computational Social Science @EMNLP, 163–172.