Modern computational approaches to AID employ such linguistic “features” in order to train a machine-learned classification algorithm to recognize the writing style of the author(s) of interest. AID is thus employed in any task in which the paternity of a document is unknown, disputed, or possibly forged, and thus must be uncovered. It was indeed linguistic analysis that allowed to uncover the identity of the “Unabomber” (see https://en.wikipedia.org/wiki/Ted_Kaczynski), one of the most famous serial killers of recent history. Moreover, countless manuscripts in our cultural heritage had their authorship lost in the centuries, or purposefully hidden or mystified. We focus on the application of AID in these ancient settings (and, specifically, on the Latin language).

Why can syllabic quantity be relevant for authorship?

As mentioned above, the basis of an AID method are the style markers, a.k.a. “stylistic features.” Since the very inception of stylometry in the late 19th century, researchers have proposed many stylistic features, some of which have become standard nowadays. However, the quest for new, potentially better features is still ongoing. 

The Latin language is based on syllables, that is, oscillations of sound in the pronunciation of a word, that are characterized by their “quantity” (“long” or “short”, depending on the amount of time required to actually pronounce it). In fact, well-chosen patterns of these syllables were the basis of Latin prosody: Latin authors gave a “rhythm” to the discourse by combining long and short syllables in rhythmic schemes, not only in poetry, but in prose as well. Orators such as Cicero were particularly aware of the effects of such metric devices, and many scholars have pointed out how certain Latin authors consistently show a certain (more or less conscious) preference for specific rhythmic patterns. The “musicality” of an author’s production, i.e., the rhythmic patterns s/he used, might thus play an important role in the identification of that author’s style; this is the intuition where our work starts.

Our contribution

In our research, we extract the syllabic quantity patterns from Latin texts and use them as stylistic features for a machine learning algorithm (Support Vector Machine). This algorithm trains a classifier that guesses the author of a Latin document from among a set of possible candidates. Tests that we have run on different datasets of Latin texts show that these rhythmic features improve the accuracy of the authorship identification, thereby confirming the idea that rhythm is indeed a stylistic marker for Latin prose. 

This opens up possible directions for future work, such as investigating the importance of rhythm on AID tasks for other languages, starting from ones linguistically close to Latin such as Italian or Spanish. In particular, it would be fascinating to study whether also modern-day prose writers unconsciously opt for specific rhythmic patterns, to the point of being somehow recognizable thanks to them.

If you are interested in this subject and want more details on the research hypothesis and experimentation setting of this work, the full article by Corbara, Moreo, and Sebastiani is available at http://doi.org/10.1002/asi.24660 .