Human vs AI-Generated Academic Abstract Readability Metrics

150 human-written and 150 AI-generated abstracts from high-impact linguistics and computer science journals were analyzed for readability and writing style. The dataset contains quantitative metrics from a Readability Scoring System and expert evaluations, compiled by Yumei Zou and last updated in April 2026. Analysis was performed using SPSS 27 with non-parametric statistical methods.

Use Cases

• Compare readability scores like Flesch-Kincaid Grade Level between human and AI-generated abstracts across two disciplines.
• Analyze discipline-specific differences in eight writing style metrics for computer science abstracts.
• Train classifiers to distinguish AI-generated text using lexical diversity and complexity features from the scoring system.
• Investigate correlations between expert evaluation scores and quantitative readability metrics for AI outputs.

Strengths

• Contains 300 total abstract samples (150 human, 150 AI) for a balanced comparison.
• Provides analysis across eight distinct readability and writing style metrics for each text.
• Source texts are from high-impact journal articles in two defined academic fields.

Limitations

• Limited to 300 total samples, which is a small dataset for training complex machine learning models.
• Scope is restricted to only two disciplines (linguistics and computer science), limiting generalizability.
• The dataset size is 207.7 KB, indicating a limited number of features or rows.

Provenance

Source

Figshare, authored by Yumei Zou.

Collection Method

Abstracts were sourced from high-impact journal articles; AI counterparts were generated from the same article corpus. Metrics were computed using a Readability Scoring System.

Time Range

null

Freshness

Last updated in April 2026, indicating recent compilation.

Geography

null