Swedish MuClaGED: A new dataset for Grammatical Error Detection in Swedish

Abstract

This paper introduces the Swedish Mu-ClaGED dataset, a new dataset specifically built for the task of Multi-Class Grammatical Error Detection (GED). The dataset has been produced as a part of the multilingual Computational SLA shared task initiative. In this paper we elaborate on the generation process and the design choices made to obtain Swedish MuClaGED. We also show initial baseline results for the performance on the dataset in a task of Grammatical Error Detection and Classification on the sentence level, which have been obtained through (Bi)LSTM ((Bidirectional) Long-Short Term Memory) methods.