智能口语双机评测模式在外语听说机考评卷中的可行性研究  

Feasibility Study of Intelligent Dual-machine Speaking Assessment Mode in Computer-based Foreign Language Listening and Speaking Test

作　　者：沈晨[1] 罗双虎 Shen Chen;Luo Shuanghu(Shanghai Municipal Educational Examinations Authority,Shanghai,200433)

机构地区：[1]上海市教育考试院网络信息中心,上海200433

出　　处：《考试研究》2023年第3期75-90,共16页Examinations Research

摘　　要：基于现有英语听说考试人机互评的评卷模式,探索双机评测模式可行性,使用上海市初中外语听说测试全真模拟数据试验,对比3种独立计算机智能评分算法的效果。结果显示,机评分与报道分一致性达到96%以上,具备良好的效果,但存在1659份样本双机评后仍误判的效果风险,综合考虑双机评测模式的评卷组织、机评评价机制仍不完备,暂不具备可行性,需要进一步的算法提升和应用方法研究;算力改变对比验证结果表明,评分准确性几乎不下降的情况下,采用GPU算力结构的评分算法的运算速度相当于CPU算力结构的6倍,这可以使得评分时间和硬件投入大幅度减少。Based on the existing evaluation mode of human-computer mutual assessment of English listening and speaking test,the feasibility of dual-computer evaluation mode was tentatively explored,and three independent computer intelligent scoring algorithms were compared by using the full-real simulation data test of Shanghai junior high school foreign language listening and speaking test.The results show that the consistency between the machine score and the report score reaches more than 96%,which has good results,but there is a risk that the effect of 1659 samples is still misjudged after the dual-machine evaluation,and the evaluation organization and evaluation mechanism of the dual-machine evaluation mode are still incomplete,and the dual-machine evaluation mode is not feasible for the time being,and further algorithm improvement and application method research are needed.The comparative verification results show that the scoring speed of the scoring algorithm using the GPU computing power structure is equivalent to 6 times that of the CPU computing power structure without the decrease in scoring accuracy,which can greatly reduce the time and hardware spent on scoring.

关 键 词：中考 外语听说测试 计算机智能评分 

分 类 号：G424.74[文化科学—课程与教学论]