Predicting Japanese primary schoolchildren's English vocabulary knowledge from a sound-meaning recognition test

The purposes of this study were to: (1) build an appropriate model for predicting primary schoolchildren’s English vocabulary knowledge; (2) examine whether the developed model applies to new data; and (3) discuss how to apply the model to L2 vocabulary instruction. More specifically, the study asked third- and fourth-grade public primary school students to take a sound-meaning recognition test. All the study’s target words were katakana English and on the first 1,000-level of word frequency according to JACET (Japan Association of College English Teachers) 8000 and SVL (Standard Vocabulary List ) 12000; however, different numbers of word phonemes were included. The collected data were divided into training and testing data. To build the predictive model from the training data, the study used a generalized linear mixed model, which revealed that: (1) third-graders’ scores were as high as those of fourth-graders, and (2) the more phonemes a word has, the lower the test score. From the model, the study developed a regression formula that made it possible to predict how many primary schoolchildren knew the meaning of a certain word when they listened to its pronunciation. It also found that the predicted percentages from the training data correlated moderately with the actual test scores from the testing data. Therefore, the model achieved moderately high quality for predicting new data.