Abstract
Since the 1990s, the development of automated essay scoring (AES) has progressed through three major phases: feature-based statistical machine learning, neural network-based deep learning, and generative pre-trained language models. Throughout this evolution, scoring rubrics and language intelligence techniques have intertwined and advanced synergistically, driving AES toward greater efficiency and rationality through their competitive pursuit of dominance in the AES. AES research has evolved from focusing on superficial statistical linguistic features to exploring complex semantic and formal features, enabling comprehensive scoring along both formal and content-based dimensions. Research on AES for Chinese as a Second Language (CSL) has primarily focused on the MHK test and the HSK test, which are designed for domestic ethnic minorities and foreign learners of Chinese, respectively. It lacks the verification of the effects on various neural network models or large-scale datasets. Technological advances in Large Language Models suggest potential solutions to current problems concerning AES for CSL. Future AES research needs to move from resembling the outcomes of manual scoring to imitating the thinking of manual scoring, transition from holistic rating scoring to analytic rating scoring, and enhance AES model reliability and interpretability through both chain-of-thought fine-tuning and high-quality corpora.
Key words
automated essay scoring (AES) /
scoring rubrics /
linguistic features /
large language models /
language industry
Cite this article
Download Citations
Shi Jinsheng and Li Xueting.
Scoring Rubrics of Automated Essay Scoring for Chinese as a Second Language Writing: Development and Reflections#br#[J]. Chinese Journal of Language Policy and Planning. 2026, 11(1): 42-52 https://doi.org/10.19689/j.cnki.cn10-1361/h.20260104
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}