Chinese readability analysis using artificial neural networks
Cheng-Chang Jeng, Northern Illinois University, United States
Doctor of Education, Northern Illinois University . Awarded
This study investigated numerous variables that may affect readability and the potential of artificial neural networks (ANN) to estimate Chinese readability while also exploring the creation of more powerful linear models than previous researchers'. To compare the accuracy among four types of methods, the following were compared to each other: a linear regression model proposed in previous research, a linear regression model, ANNs, and human judgments. The results of the comparisons show that the ANN models offered the best estimates among the four models when the comparisons were based on an instrument compiled from 12 official textbooks. Human judgments offered more reliable prediction results and none of the mathematical models achieved good accuracy when the comparisons were based on an instrument compiled from 12 children's books. These results suggest that human judgments are preferable to computational models when unfamiliar readings are involved. Issues discussed include the validity of assigned readability levels and the merits of the various models. Suggestions for future research include adding grammatical variables, using longer passages, and using fuzzy logic in instrument design. Application of the ANN model to English readability is a logical next step. Regardless of language, accurate readability estimates are important to the creation of linguistically suitable instructional materials.
Jeng, C.C. Chinese readability analysis using artificial neural networks. Doctor of Education thesis, Northern Illinois University.
Citation reproduced with permission of ProQuest LLC.
For copies of dissertations and theses: (800) 521-0600/(734) 761-4700 or https://dissexpress.umi.com