Answers are grounded in this paper and cite their sources.
Corpus analysis reveals distinct length, lexical diversity, and syntactic patterns in student versus AI essays, highlighting detectable stylistic markers between author groups.
This research compares the stylistic contrasts in student and LLM essay content within the Hugging Face dataset “knarasi1/student_and_llm_essays.” The dataset consists of 1733 documents – 1375 student essays and 358 LLM essays. The style comparisons will include document length, type-token ratios (TTR), percentages by part-of-speech (POS), patterns in exploratory function words and TF-IDF; and, use logistic regression to classify. In terms of overall averages, there was a difference in both word count and sentence count. The LLM-labeled essays had lower word counts but a higher TTR than the student-labeled essays. Two-sided Mann-Whitney U tests indicated statistically significant group differences for word count, sentence count, and TTR, with Cliff’s delta magnitudes of medium, large, and large, respectively. Most examined POS distributions also differed; VERB was the only category that was not statistically significant. A logistic-regression model using document-level and POS features achieved a mean five-fold stratified cross-validation accuracy of 0.9833 (population SD = 0.0085; sample SD = 0.0095), with balanced accuracy of 0.9729 and macro F1 of 0.9744. Overall, the findings provide an overview of the style differences between the two labeled groups within this data set. The findings are unique to the corpus being examined. Therefore, they cannot be considered indicative of general differences between LLM generated writings and those written by humans.
No takes yet. Share an insight, caveat, or question.
Vladislav Koltunov (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: