嵌入语言深度感知的汉语作文评分算法

Chinese Composition Scoring Algorithm Embedded with Language Deep Perception

从语言深度感知设计了一种汉语作文自动阅卷评分算法,抽象出语感特征的计算模型以模拟人类对自然语言的评价标准,弥补了早期作文自动阅卷中的机械统计式自然语言处理技术的不足。采用AC自动机对语感支撑要素,即作文作者的个人语言素材,进行快速分析。利用文本分词和主干提取等技术实现了对诸如句子流畅度等语句级评价,并将待评测作文的上下文结构与标准作文框架进行相似性比对,从而在作文自动评分系统中建立对作者语言运用能力的综合评价。实验结果表明,该算法增强了自动评分的语言能力评定的合理性,也更加贴切与专家校准后的人工评分样本。

The paper designs an automatic scoring algorithm for Chinese composition from the perspective of language deep perception. The computational modeling on the abstract features of language intuition mimics the evaluation criteria of human for natural language. It makes up for the insufficiency of mechanical statistical natural language processing technology in early automatic scoring of compositions. Aho-Corasiek automata is built in to quickly analyze the elements supporting language intuition, i.e. the indexes of personal language corpus of the writer. In evaluating the language ability on a sentence level, such as sentence fluency, technologies like text segmentation and structure recognition are embedded in the system, together with the similarity comparison between the composition to be treated and the standard modes of discourse, and therefore a comprehensive evaluating algorithm is created to author’s language competency in the composition automatic scoring system. Experimental results show that the algorithm improves rationality of automatic language ability scoring and gets closer to fit the expertized scoring samples.