Measuring English Readability for Vietnamese Speakers

Measuring English Readability for Vietnamese Speakers Thuan Nguyen and Alexandra L. UITDENBOGERD RMIT University - School of Science 124 La Trobe St Melbourne VIC 3000 john.nguyen09@outlook.com,sandra.uitdenbogerd@rmit.edu.au Abstract Reading is important for language learners, but text difficulty needs to match a reader’s skill level for efficient vocabulary acquisition. Tra- ditional readability measures may not be ef- fective for those who speak English as a sec- ond or additional language. This study ex- amines English readability for Vietnamese na- tive speakers (VL1). A collection of text dif- ficulty judgements of nearly 100 English text passages was obtained from 12 VL1 partic- ipants, using a 5-point Likert scale. Using features from traditional readability measures, support vector machines and Dale-Chall fea- tures gave more accurate predictions than lin- ear models using either Flesch or Dale-Chall features. VL1 participants’ text judgements were strongly correlated with their past En- glish test scores. This study introduces a first approximation to readability of English text for VL1, with suggestions for further improve- ments. 1 Introduction Extensive reading, that is, reading a large amount of text at a comfortable level of difficulty, is an efficient way to improve language skills, as learn- ers acquire vocabulary as they read, retain it for longer than if they use rote memorisation (Her- mann, 2003), and greatly improve their receptive skills (Elley and Mangubhai, 1983). However, the level of difficulty of the text needs to match the learner, for example, to guess the meaning of new words, readers typically need to know at least 95% of the words (Laufer, 1989). Further, text needs to be well below the learner’s frustration level (Klare, 1988). As most text written for native (L1) speak- ers is beyond the beginner and intermediate lan- guage learner, students need to start with simple or simplified text and work their way up to more advanced texts as they learn. Thus a method of measuring the readability of text is crucial for lan- guage learners, and can be incorporated in to read- ing recommender systems. There have been many proposed English text readability measurement techniques, most of which were modelled on native English-speaking (EL1) children, or texts written for them (for example (Flesch, 1948; Dale and Chall, 1948; Schwarm and Ostendorf, 2005)). English com- prehension is different for people with differ- ent backgrounds and language skills, causing readability measurement techniques developed for EL1 to perform poorly for other L1-L2 combi- nations (Oller et al., 1972; Uitdenbogerd, 2005). The focus of this study was English readability for Vietnamese speakers (VL1). To our knowledge, no techniques have been built and tested specifi- cally for this cohort. While there is existing work on readability for non-native (L2) speakers, the majority is trained on data sets that assume texts written for different language levels match the comprehension experi- enced by L2 speakers. Examples include Franc¸ois and Miltsakaki (2012) for French L2 and Xia et al. (2016) for English. Xia et al. (2016) extended readability measurement techniques by adapting a model trained on texts for English …

Content truncated to 500 words