Lexical complexity prediction

Abstract

This study tested the exploration of phoneme data while faced with the challenge of adding value to Lexical Complexity Prediction. Normally, most solutions use data that emanates from psycholinguistics, morphology, and stats, such as frequency of works, in order to achieve accurate modelling of lexical complexity prediction. In this study, the modelling of letter-to-phoneme translation and syllable boundary labelling took place as prepossessing to ensure the perspective on how the word is viewed phonologically. Using this phonological information, it was then passed into a Convolutional Neural Network for feature mapping to be aggregated. After passing into a Convolutional Neural Network, it was concatenated with other properties and values. These were subsequently passed into a Long Short-Term Neural Network and a Feed-Forward Neural Network to ensure that the data that adds value was garnered, while enabling the categorisation of lexical complexity. This was the first time that phonemic data was used during the process of lexical complexity prediction. Although phonemic data was partially used in one study, it seemed to be implemented in an incomplete manner.