A Reliable Approach to Automatic Assessment of Short Answer Free Responses

This paper discusses an innovative approach to the computer assisted scoring of student responses in WebLAS (web-based language assessment system)- a language assessment system delivered entirely over the web. Expected student responses are limited production free response questions.The portions of WebLAS with which we are concerned are the task creation and scoring modules. Within the task creation module, instructors and language experts do not only provide the task input and prompt. More importantly, they interactively inform the system how and how much to score student responses. This interaction consists of WebLAS' natural language processing (NLP) modules searching for alternatives of the provided "gold standard" (Hirschman et al, 2000) answer and asking for confirmation of score assignment. WebLAS processes and stores all this information within its database, to be used in the task delivery and scoring phases.