Towards the Automatically Semantic Scoring in Language Proficiency Evaluation

Many features have been proposed to evaluate examineespsila language proficiency. However, few of them are semantic based. In this paper, a novel feature for semantic scoring is presented. It is designed for a typical question type in language tests, namely reading-answering-problem. The proposed feature extraction process involves several operations: transcribing the speech data, automatically tagging the transcribed text and scoring the tagged text. The pattern based tagging is performed on the pre-designed Finite State Machines (FSMs) and the scoring fusion is based on the semantic calculations in a knowledge database. Experiment on Mandarin data validates the effectiveness of the semantic feature in the language proficiency evaluation.