Automatic evaluation of English retelling proficiency in large machine based oral English tests
Feng Guo–dong · Journal of Tsinghua University(Science and Technology) · 2009
Generalized computerized language evaluation systems need to be text independent.This paper describes an automatic grading system for English retelling proficiency for large machine based oral English tests.The system combines continuous speech recognition(CSR) and natural language understanding(NLU) in an automatic grading system.In their retelling of a story,the examinees don't need to exactly describe the articles as in the original texts and the corpus may be collected in a noisy environment.Therefore the system needs various techniques such as acoustic model adaptation using the reading data,language model(LM) pruning based on given texts,and specific grading features based on word graph of recognition results.System performance on 339 oral English tests of university students agreed 84% with two experts and exceeded the performance of mass on spot grading.Further tests show that this system can help teachers give more objective scores in English retelling examinations.