Comparing Human and GPT Model in Chinese Tourist Text Simplification
Yue Xu, Shaojie Geng, Liangjie Yuan, Wei Wei, Zhimin Wang · 2023
This study evaluates the ability of tourist texts to be simplified by humans and machines (ChatGPT-3.5), and examines the differences and characteristics between human-generated and machine-generated texts at the lexical, syntax, and discourse levels, focusing on text readability. The results indicate significant differences in vocabulary and discourse between Human-simplified texts and Model-simplified texts, while syntactic simplification remains relatively consistent. Human simplification considers various aspects of vocabulary factors, such as the impact of word length and lexical level on text difficulty, whereas model simplification slightly lacks in this regard. The model-simplified texts seem to convey richer discourse coherence information through an increased density of conjunctions. This study provides empirical research references for subsequent text generation based on Large Language Models and offers guidance for machine-assisted human simplification.