Organizations: Software Engineering Institute, AI Division, Carnegie Mellon University · Department of Statistics & Data Science, Carnegie Mellon University
Large Language Models (LLMs) are now capable of generating highly fluent, human-like text. They enable many applications, but also raise concerns such as large scale spam, phishing, or academic misuse. While much work has focused on detecting LLM-generated text, only limited work has gone into understanding the stylistic differences between human-written and machine-generated text. In this work, we perform a large scale analysis of stylistic variation across human-written text and outputs from 11 LLMs spanning 8 different genres and 4 decoding strategies using Douglas Biber's set of lexicogrammatical and functional features. Our findings reveal insights that can guide intentional LLM usage. First, key linguistic differentiators of LLM-generated text seem robust to generation conditions (e.g., prompt settings to nudge them to generate human-like text, or availability of human-written text to continue the style); second, genre exerts a stronger influence on stylistic features than the source itself; third, chat variants of the models generally appear to be clustered together in stylistic space, and finally, model has a larger effect on the style than decoding strategy, with some exceptions. These results highlight the relative importance of model and genre over prompting and decoding strategies in shaping the stylistic behavior of machine-generated text.
Figures & tables
Figure 1: Top 15 most important Biber features for distinguishing human and LLM-generated text based on Random Forest classifier on down-sampled model.
Figure 2: Feature Distribution for f_43_type_token
Figure 3: Top 15 SHAP interaction pairs based on Random Forest classifier on down-sampled model.
Figure 4: Ratio of Feature Usage LLM Vs. Human for Selected Features.
Figure 5: Heat map of Log10 Biber Feature Ratios (LLM / Human) for Selected Important Features
Figure 6: Scatterplot of first two PCA dimensions of mean Biber features.
Figure 7: Relaxed consensus tree summarizing hierarchical clustering results over all genres.
Figure 8: Visualizations of Biber features across selected RAID genres. Colors indicate model identity, while marker shapes indicate decoding and repetition-penalty settings. Human text is shown with a distinct marker.
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 9: ROC curve comparing base model and downsampled model.
Figure 10: Distribution of Biber features (Part 1).
Figure 11: Distribution of Biber features (Part 2).
Figure 12: Ratio of Feature Usage LLM Vs. Human for All Features.
Figure 13: Heat map of Log10 Biber Feature Ratios (LLM / Human) for All Features
Figure 14: Hierarchical clustering dendrograms using PCA with components capturing 95% variance on normalized biber features.
Figure 15: PCA visualizations of Biber features across eight RAID categories. Colors indicate model identity, while marker shapes indicate decoding and repetition-penalty settings. Human text is shown with a distinct marker.
Prior work on LLM-generated text has demonstrated quantitative and qualitative departures from text produced by humans. LLM-generated texts differ from human writing in style, resulting in a characteristic textual "feel," while the semantic range of LLMs is much restricted compared to that of humans. In this contribution, I note simple but consistent patterns in the statistical distribution of n-grams within LLM-generated text. Via qualitative analysis of these n-grams, I reveal deficiencies in LLM style. Because higher-order n-grams correlate to semantic content, I conclude that questions of style and semantics are not cleanly separable.
Despite the growing use of large language models (LLMs) for writing tasks, users may hesitate to rely on LLMs when personal style is important. Post-editing LLM-generated drafts or translations is a common collaborative writing strategy, but it remains unclear whether users can effectively reshape LLM-generated text to reflect their personal style. We conduct a pre-registered online study (n=81) in which participants post-edit LLM-generated drafts for writing tasks where personal style matters to them. Using embedding-based style similarity metrics, we find that post-editing increases stylistic similarity to participants' unassisted writing and reduces similarity to fully LLM-generated output. However, post-edited text still remains stylistically closer in style to LLM text than to participants' unassisted control text, and it exhibits reduced stylistic diversity compared to unassisted human text. We find a gap between perceived stylistic authenticity and model-measured stylistic similarity, with post-edited text often perceived as representative of participants' personal style despite remaining detectable LLM stylistic traces.
Interpretable linguistic features offer a promising approach for explaining why a given text appears machine-generated, particularly for non-expert users. However, existing findings on which features reliably indicate LLM-generated text remain fragmented across feature sets, models, and text domains. To address this gap, we conduct a large-scale empirical study assessing the robustness of linguistic signals for characterizing AI-generated text. Our analysis covers 284 interpretable linguistic features across outputs from 27 LLMs and ten text domains under cross-model and cross-domain generalization settings. We show that classifiers based solely on linguistic features can reliably distinguish AI-generated from human-written text. However, many previously proposed indicators prove strongly context-dependent, with the exception of measures of lexical richness, which remain robust signals across model families and text domains. These results demonstrate which linguistic signals generalize across contexts and provide a foundation for more reliable, interpretable analyses of AI-generated language.
Yassir El Attar, Esra Dönmez, Maximilian Maurer +1
Institute for Natural Language Processing, University of Stuttgart · Interchange Forum for Reflecting on Intelligent Systems, University of Stuttgart · GESIS Leibniz Institute for the Social Sciences +1