BACKGROUND
Prior research has uncovered the significance of eye-tracking experiment with text reading in neuroscience research for assessing cognitive function. Further studies have highlighted the need for more reading materials with higher levels of difficulty to ensure optimal experimental outcomes. Unfortunately, the current approach predominantly relies on laborious and inefficient manual production of such texts.
OBJECTIVE
This study aims to use ChatGPT to generate tailored Chinese reading materials for eye-tracking experiments with Chinese-sentence-reading-tests, taking into consideration the exceptional language analysis abilities of ChatGPT and its user-friendly accessibility worldwide.
METHODS
After selecting appropriate sentence from Chinese literary works, the high-frequency word in the sentences are replaced by ChatGPT with low-frequency word, resulting in sentence with increased reading difficulty.
RESULTS
The word frequency of the low-frequency words generated by ChatGPT notably decreases compared to the original high-frequency words, and the decrease ratio in word frequency can reach 82.3%. Employing the Wilcoxon Rank-Sum Test and T-test to assess the disparities between the low-frequency and high-frequency groups, the majority of groups exhibit P-values below 0.04.
CONCLUSIONS
The findings of this study highlight the varied effectiveness of ChatGPT in word replacement across different parts of speech. Nouns and verbs show superior substitution outcomes compared to adjectives. The performance also varies depending on the specific Chinese literary works employed, with superior performance in generating reading materials from contemporary Chinese literary works as opposed to modern ones.
CLINICALTRIAL
ChatGPT; Eye-Tracking; Reading Test; Chinese; Word frequency