Monolingual sentence matching for text simplification

TL;DR

This paper presents a neural network-based approach to improve monolingual sentence matching for text simplification, leveraging semi-supervised training and adaptation to enhance alignment accuracy between standard and simple Wikipedia sentences.

Contribution

It introduces a convolutional neural network model trained semi-supervisedly and adapted with small datasets to improve sentence alignment for text simplification.

Findings

Rescoring improves alignment accuracy.

Model adaptation enhances performance.

Neural approach outperforms baseline methods.

Abstract

This work improves monolingual sentence alignment for text simplification, specifically for text in standard and simple Wikipedia. We introduce a convolutional neural network structure to model similarity between two sentences. Due to the limitation of available parallel corpora, the model is trained in a semi-supervised way, by using the output of a knowledge-based high performance aligning system. We apply the resulting similarity score to rescore the knowledge-based output, and adapt the model by a small hand-aligned dataset. Experiments show that both rescoring and adaptation improve the performance of knowledge-based method.