基于人工神经网络的韩国汉语学习者口语流利性评价研究

On the Evaluation of Korean Students’ Oral Chinese Fluency Based on Artificial Neural Networks

流利性是衡量二语学习者口语水平高低的重要标准之一。本研究以HSK(高等)口语测试语料为基础,统计了时间、表达和语言三个维度14项口语流利性测量指标,通过人工神经网络考察口语流利性测量指标对口语水平的预测效度。研究结果表明:选取的14项指标中百音节更改次数、T单位正确率、剔除更改音节占比、修正比例、语速这5项指标的标准化贡献率均大于50%,对汉语二语者的口语水平具有较高的预测价值。

Fluency is one of the important criteria to measure the oral proficiency of second language learners. Based on the materials of the HSK(Advanced) Speaking Test, this study counts 14 indicators of oral fluency in three dimensions: temporal, performance and linguistics,and examines the predictive validity of indicators on oral proficiency through artificial neural networks. The results show that among the 14 selected indicators, the standardized contribution rate of the five indictors, namely the number of repairs per 100 syllables, the ratio of error-free T-units to total T-units, the ratio of pruned length to total length, the ratio of amendments to total repairs and the speech rate are higher than 50%, indicating that this five indicators have a high predictive value for the oral proficiency of Chinese as a second language learners.