Automated scoring system for short-answer responses

在我们的研究之初，我们计划使用召回率和精度两个指标来衡量输入答案和模型答案之间的相似性。在此之后，事实证明，使用基于合法的自然语言处理技术的文本蕴涵识别是有前途的。我们的研究是先进的，从国家信息学研究所的成员那里获得了知识。主要成果如下：(1)在第10届日语考试学会大会上公布了以我们的实际证明实验为基础的论文评价或评分。(2)关于在作文或作文测试中使用计算机和自动评分，我们在公开研讨会上做了一个讲座，广泛介绍了我们的活动。(3)开发了面向自动评分的统计技术方法，并在著名期刊上发表了多篇论文。

At the beginning of our research, we were planning to use two indices of recall and precision in order to measure a similarity between an input answer and the model answer. After that, it turned out that use of the textual entailment recognition based on the legitimate natural-language-processing technology is promising. Our research was advanced, acquiring knowledge from the member of National Institute of Informatics. Main results are follows:(1) Evaluating or grading essays based on our actual proof experiment was announced at the 10th Japanese Test society convention. (2) About the computer use and automatic grading in an essay or composition test, a lecture was given at open symposium to introduce our activity widely. (3) The method of the statistical technology towards automatic grading was developed, and many papers were published in famous journals.