Exploring hybrid linguistic feature sets to measure filipino text readability
The proper identification of difficulty levels of reading materials prescribed in an educational setting is key towards effective learning and comprehension. Educators and publishers have relied on readability formulas in predicting text readability. While the English language boasts a rich history...
Saved in:
| Main Author: | |
|---|---|
| Format: | text |
| Language: | English |
| Published: |
Animo Repository
2021
|
| Subjects: | |
| Online Access: | https://animorepository.dlsu.edu.ph/etdm_comsci/5 https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1000&context=etdm_comsci |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | De La Salle University |
| Language: | English |
| Summary: | The proper identification of difficulty levels of reading materials prescribed in an educational setting is key towards effective learning and comprehension. Educators and publishers have relied on readability formulas in predicting text readability. While the English language boasts a rich history of research efforts in readability assessment, limited work has been done for the Filipino language. This study explores the use of an extensive range of linguistic predictors identified by experts spanning traditional, lexical, language model, syllable pattern, and morphological features to train an automatic readability assessment model using Logistic Regression, Support Vector Machines, and Random Forest. Over 265 story books and passages from Adarna House Inc. and DepEd Commons covering Grades 1, 2, and 3 were used for training the models. Results of feature selection process show that the optimal subset of linguistic feature sets achieving the highest performance of 66.1\% accuracy is a hybrid Random Forest model using the combination of traditional (TRAD) and syllable pattern (SYLL) features. Performing global and local model interpretation showed that surface-based features such as word count, average sentence length, and sentence count used in old readability formulas remain relevant in measuring the readability of Filipino texts, but combining them with deeper linguistic features would yield better performance of models. Future directions of the study include the use of various types of written literature, not only story books, to develop a more generalized readability assessment model as well as the use of deep neural networks for automatic feature extraction.
Keywords: Readability Assessment, Filipino, Linguistic Features, Story Books |
|---|