Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)
Computer Speech and Language
Towards automatic scoring of a test of spoken language with heterogeneous task types
EANL '08 Proceedings of the Third Workshop on Innovative Use of NLP for Building Educational Applications
| Hi-index | 0.00 |
Automated testing of spoken language is the subject of much current research. Elicited Imitation (EI), or sentence repetition, is well suited for automated scoring, but does not directly test a broad range of speech communication skills. An Oral Proficiency Interview (OPI) tests a broad range of skills, but is not as well suited for automated scoring. Some have suggested that EI can be used as a predictor of more general speech communication abilities. We examine EI for this purpose. A fully automated EI test is used to predict OPI scores. Experiments show strong correlation between predicted and actual OPI scores. Effectiveness of OPI score prediction depends upon at least two important design decisions. One of these decisions is to base prediction primarily on acoustic measures, rather than on transcription. The other of these decisions is the choice of sentences, or EI test items, to be repeated. It is shown that both of these design decisions can greatly impact performance. It is also shown that the effectiveness of individual test items can be predicted.