Tapping the Potential of Coherence and Syntactic Features in Neural Models for Automatic Essay Scoring

Abstract

In the prompt-specific holistic score prediction task for Automatic Essay Scoring (AES), the general approaches include pre-trained neural model, coherence model, and hybrid model that incorporate syntactic features with neural model. In this paper, we propose a novel approach to extract and represent essay coherence features with NSP that matches the state-of-the-art (SOTA) AES coherence model, and achieves the best performance for long essays. We apply syntactic feature dense embedding to augment BERT-based model and achieve the best performance for hybrid methodology for AES. In addition, we explore various ideas to combine coherence, syntactic information, and semantic embeddings, which no previous study has done. Our combined model also performs better than the SOTA available for combined model, even though it does not outperform our syntactic-enhanced neural model. We further compare with the pure neural models and analyze the strengths and weaknesses of our methodologies.