Automatic text simplification: research and directions

Abstract

The article provides a comprehensive survey on automatic text simplification (ATS) as an independent research area in the field of natural language processing. The study aims to systematically and critically present the development, current state, and key challenges of ATS. Based on an extensive literature review in Scopus, Google Scholar and the ACL Anthology, research from 1998 to 2025 is analyzed. The article examines the evolution of ATS development from rule-based approaches to the use of large language models using statistical and neural models. It is shown that this process goes hand in hand with the gradual expansion of high-quality parallel corpora in several languages.

A particular focus is placed on the analysis of lexical simplification, including the (1) identification of complex words, as well as (2) selection, (3) generation and (4) ranking of substitutions. The study shows that isolated rule-based, frequency-based, or purely data-driven approaches often reach their limits and that hybrid, linguistically grounded solutions deliver the best results. Key challenges remain the preservation of meaning and coherence, the strong dominance of English in research, and the lack of resources for typologically complex languages ​​like Kazakh.

The article notes that purely neural approaches to such languages are not enough. Instead, a step-by-step approach is proposed, based on linguistically sound reliable modeling, as well as complemented by automated and neural methods. The survey highlights the importance of improving evaluation procedures for the further development of automatic text simplification, as well as typologically oriented linguistically sound research.

Keywords: text, simplified text, automatic text simplification, the Kazakh language