We propose a neural network-based system for automatically grading essays written in Swedish. Previous system either relies on laboriously crafted features extracted by human experts or are limited to essays written in English. By using different variations of Long Short-Term Memory (LSTM) networks, our system automatically learns the relation between Swedish high-school essays and their assigned score. Using all of the intermediate states from the LSTM network proved to be crucial in order to understand the essays. Furthermore, we evaluate different ways of representing words as dense vectors which ultimately have a substantial effect on the overall performance. We compare our results to the ones achieved by the first and previously only automatic essay scoring system designed for the Swedish language. Although no state-of-the-art performance is reached, indication of the potential from a neural based grading system is found.

This master thesis work presents a novel method of automatic essay scoring for Swedish national tests written by upper secondary high school students by deploying neural network architectures and linguistic feature extraction in the framework of Swegram. There are four sorts of linguistic aspects involved in our feature extraction: count-based,lexical morphological and syntactic. One of the three variants of recurrent network, vanilla RNN, GRU and LSTM, together with the specific model parameter setting, is implemented in the Automatic Essay Scoring (AES) modelling with extracted features measuring the linguistic complexity as text representation. The AES model is evaluated through interrater agreement with human assigned grade as target label in terms of quadratic weighted kappa (QWK) and exact percent agreement. Our best observed averaged QWK and averaged exact percent agreement is 0.50 and 52% over 10 folds among our all experimented models.