Exploring the Authorship of the Historia Augusta Using Computational Stylometry

The Historia Augusta is an anonymous collection of biographies of 30 Roman emperors who ruled from 117 to 284 CE and serves as the principal Latin source for this period of Roman history. Although the work is attributed to six named authors, these identities are widely considered pseudonymous, leaving its true authorship unresolved. This project applies computational stylometry to evaluate whether contemporary writers, including Aurelius Victor, Symmachus, Eutropius, Festus, and Julius Obsequens, exhibit stylistic patterns consistent with the Historia Augusta. While no single author demonstrates overwhelming similarity, the results capture meaningful stylistic signals and align with prior observations of overlap with Eutropius. These findings suggest that, despite limitations in available data, computational methods can reflect established patterns and provide a scalable framework for future authorship analysis.

Data

This project uses texts from the Historia Augusta alongside works from known Latin authors, including Aurelius Victor, Eutropius, Festus, Julius Obsequens, and Quintus Aurelius Symmachus. All texts were obtained from public domain sources and combined into a unified corpus for analysis.

Preprocessing

All text was normalized by converting characters to lowercase and removing non-content metadata such as section headers. Orthographic variation was reduced by standardizing Latin character usage (e.g., treating j as i).

Chunking Strategy

Texts were divided into fixed-length chunks to enable consistent comparison across authors. Remainder chunks smaller than 80% of the target size were discarded to avoid unstable samples. A chunk size of 500 words was selected to maintain stylistic signal while maximizing the number of usable samples under limited Latin data availability.

Feature Extraction

Character-level 4-grams were extracted from each text chunk. A global frequency analysis was performed across the corpus, and the top 400 most frequent n-grams were selected as features for comparison. Each chunk was represented as a normalized frequency vector over these features.

Similarity Measure

Burrows' Delta was used to measure the stylistic similarity between text chunks. Feature vectors were standardized using z-scores, and distances were computing using Manhattan (L1) distance.

Visualization

Principal Component Analysis (PCA) was applied to the standardized feature space to visualize clustering patterns between chunks for assessment of stylistic similarity between authors.

This study found no single author with overwhelming stylistic similarity to the Historia Augusta, suggesting that authorship cannot be definitively attributed using the available data. However, the results are consistent with prior scholarship indicating a stylistic overlap between the Historia Augusta and Eutropius, demonstrating that the methodology is capable of capturing meaningful signals.

These findings also highlight key limitations, particularly the scarcity and uneven distribution of available source texts, as well as variation in writing domains, both of which introduce instability into the analysis.

Future work will focus on expanding the corpus as additional texts become digitized, allowing for a broader set of candidate authors and more robust comparisons. The methodology presented here provides a scalable framework that can be applied to larger datasets and refined with additional features to improve the reliability of computational authorship studies in classical literature.

Martins, Armando, Clara Grácio, Cláudia Teixeira, Irene Pimenta Rodrigues, Juan Luiz Garcia Zapata, and Lúgia Ferreira. “Historia Augusta authorship: an approach based on measurements of complex networks.” Applied Network Science 6 (2021): 50. https://doi.org/10.1007/s41109-021-00390-7.