Syntax-driven Sentence Simplification: A Computational linguistics Approach to enhance text Readability for Non-native English Speakers

Abstract

This thesis presents a comprehensive investigation into syntax-driven sentence simplification as a computational linguistics approach to enhance text readability for non-native English speakers. Drawing upon established research in syntactic simplification and text cohesion (Siddharthan, 2006), this study develops a rule-based text simplification system that targets complex syntactic structures, particularly relative clauses and coordinate structures. The research methodology employs dependency parsing using SpaCy v3.5 to identify and transform syntactic patterns that pose comprehension challenges for non-native speakers. The system implements transformation rules based on syntactic simplification principles established by Chandrasekar et al. (1996) and extends the work of Medero and Ostendorf on identifying targets for syntactic simplification. The evaluation framework incorporates SARI (System for Assessing Text Simplification) metrics as proposed by Xu et al. (2016), alongside human evaluation protocols to assess simplification quality, meaning preservation, and grammaticality. Results demonstrate significant improvements in text readability while maintaining semantic coherence through the application of rhetorical structure theory and centering theory principles. The study contributes to the field of computational linguistics by providing a systematic approach to syntactic simplification that addresses the specific needs of non-native English speakers in educational contexts.