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

TL;DR

This paper introduces novel methods combining coherence and syntactic features with neural models for automatic essay scoring, achieving state-of-the-art results especially on long essays and exploring new combinations of features.

Contribution

It proposes a prompt-learning NSP for coherence features and syntactic dense embeddings to enhance BERT-based models, surpassing existing methods in AES performance.

Findings

Coherence features with NSP improve long essay scoring.

Syntactic dense embeddings enhance hybrid neural models.

Combined models outperform previous state-of-the-art results.

Abstract

In the prompt-specific holistic score prediction task for Automatic Essay Scoring, 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 prompt-learning NSP that shows to match the state-of-the-art 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 before. 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 offer analyses that can be useful for future study.