Predicting English word concreteness through its multidimensional perceptual and action strength norms

Abstract

Many datasets resulting from participant ratings for word norms and also concreteness ratios are available. However, the concreteness information of infrequent words and non-words is rare. This work aims to propose a model for estimating the concreteness of infrequent and new lexicons. Here, we used Lancaster sensory-motor word norms to predict the word concreteness ratios of an English word dataset. After removing the missing values, we employed a stepwise multiple linear regression (SW-MLR) procedure for choosing an optimum number of norms to develop a predictive multiple regression model. Finally, we validate our model using 10-fold cross-validation. The final model could predict concreteness by Residual Mean Standard Error equal to 0.723 and R-Square of 0.515. Also, our results showed that all 11 variables of this dataset except the Head-mouth parameter are useful predictors. In conclusion, as a growing demand to know the concreteness values of non-words and also infrequent words is evident, our statistical method can pave the way for controlled experiments when choosing non-words as a stimulus is critical.