Published Paper Details:

Text simplification is one of the main factors making complex information available to diversified audiences, from non-native speakers to students and people with cognitive disabilities. This paper provides in-depth comparative analysis of five state-of-the-art text simplification models: ACCESS, MUSS, T5, GraSP, and EditNTS. We evaluate these models using various metrics: simplification ratio, MOS, precision, recall, and F1-score to examine how effective the models are in making texts readable while not losing their meaning. Our experiments reveal that, on average, ACCESS works better (MOS 4.3, F1-score 0.86) compared with other methods that strike a proper balance between preserving the content of the original and making the text simple enough for readability while MUSS generally is apt for most uses with an acceptable simplification ratio 0.87. T5 provides a precision-sensitive performance with the precision value of 0.86, GraSP provides excellent balance for preserving semantics and EditNTS has the property of aggressive simplification. All these, as our statistical tests, are also confirmed to be statistically significant at p < 0.05. In the concluding section we will reflect on the avenues of possible future work involving these novel models, that is hybrid models and rich capabilities of supporting multiple languages, aside from a fitting and much more accurate assessment framework for such systems. This work contributes to advancements in the technology of text simplification and its applications in making information more accessible across diverse user groups.