Semantic based neural model approach for text simplification

Abstract

The machine translation systems affect by various difficulties like long-distance dependency and long sentences having complex syntax. Text Summarization (TeSu) and Text Simplification (TeSi) are the important ways of simplifying the text for users who are having the poor reading capability, including non-native speakers, functionally illiterate and children. TeSu produce a brief summary of the main ideas of the text, while TeSi aims to reduce the linguistic complexity of the text and retain the original meaning. In many text generation tasks, sequence-to-sequence model depends upon approaches of TeSu and TeSi achieves more success, recently. Text data have low Semantic Relevance (SR), but the Simplified Text (SiTs) which generate from the Source Text (SoTs) are more similar. The goal of the paper is to work for TeSu and TeSi, for improving the SR between the original texts and the modified texts. The proposed method encouraging high semantic similarity between texts and summaries by implementing SR based Neural model (SRN). The encoder represents the SoT, whereas, the decoder produced the summary representation. During training, the representation provides maximum Similarity Score (SS) and the experi-ments conducted on the approach using two benchmark datasets. The experimental results showed that SNR approach provided better per-formance compared to the existing method in terms of metrics such as readability metrics, human-sentence level evaluation, and Post Editing (PE) evaluation.