Ordinate与SpeechRater口语自动评分系统评述与启示  

A Critical Review of Two Automated Speech Scoring Systems: Ordinate & SpeechRater

作　　者：罗凯洲[1] 韩宝成[2] 

机构地区：[1]北京外国语大学中国外语教育研究中心,北京100089 [2]北京外国语大学外国语言研究所,北京100089

出　　处：《外语电化教学》2014年第4期27-32,共6页Technology Enhanced Foreign Language Education

基　　金：教育部人文社科重点研究基地北京外国语大学中国外语教育研究中心2012年重大课题"高考英语考试改革研究:考试方案设计及效度论证"(编号DHA070148)

摘　　要：口语自动评分系统是全自动口语考试的核心技术,为大规模口语考试的实施带来新契机。在口语自动评分领域,Ordinate与SpeechRater评分系统被公认为典型代表,但普通考生、教师乃至一些测试研发者对这一新生事物并不十分熟悉,甚至抱有误解。本文试从系统架构(语音识别、特征提取、分数计算)及评分质量视角对比评述两大系统,以期揭开口语自动评分过程的神秘面纱;并从系统与考试构念及任务的关系层面切入,评论两大系统带来的启示。As the core technology of automatic spoken language tests, automated speech scoring system offers new opportunities for large-scale speaking assessments. Ordinate and SpeechRater have been widely accepted as the typical automated systems for scoring spoken responses. Despite their potential uses, relatively work has been done in the detailed introduction and critical review of the systems, and it is therefore not well understood by some test users and even developers and researchers. This paper outlines and evaluates the two systems in terms of their architecture （including speech recognition, feature extraction and score aggregation） and scoring quality. It also examines the implications of two systems for the design of spoken language tests and development of automated speech scoring systems in China.

关 键 词：口语考试 自动评分系统 构念定义 任务设计 

分 类 号：H319.3[语言文字—英语]