Perseus News: In praise of the open digital classics ecosystem

In praise of the open digital classics ecosystem

Gregory Crane

We have an article that is about to appear in the Journal of Open Humanities Data on “Perseus and the Open-Source Ecosystem of Greco-Roman Studies,” but I thought I would precede that more formal discussion with a very recent example of what you can do in this open ecosystem. The discussion below talks about how quickly we were able to provide a usable first digital version for the Dionysiaca of Nonnus. For now, Nonnus appears in the exploratory Perseus Multitext Viewer and is hosted on Github pages: https://gregorycrane.github.io/persverscomp/.

I was looking at Hypotactic, David Chamberlain’s foundational site with metrical analysis of Greek and Latin poetry — 250,000 lines (probably more) and found this update from July 12, 2026.

David was not kidding about having made a lot of corrections — I went through and checked the GitHub repository.

Because David had put so much work into cleaning up our file and adding metrical analysis, I thought about bumping Nonnus up — way up — the list of those works I am experimenting with in the exploratory Perseus Multitext Viewer.

Adding the Greek was easy enough but most readers also want an English translation. I looked back at the 1940 Loeb edition and was very pleased to see that W. H. D. Rouse, one of my favorite British translators from the 20th century, had produced the translation. After I looked into converting the Rouse translation into digital form, however, I thought to check ToposText and discovered that their Nonnus was Rouse and was based on a digital version that Theoi.com had created. If we were entering the translation, we would try to keep the notes in place and perhaps add some other features, but we can’t do everything and we are very behind in providing translations.

ToposText uses HTML. Not so long ago, I would have written a program to convert their HTML into XML. Now I was able to ask OpenAI’s 5.6 Sol model to do the conversion for the whole work. Thus we had the traditional Greek text with English translation.

Figure 1: the Greek text of Nonnus’ Dionysiaca in the Perseus Multitext Viewer.

But, of course, David Chamberlain had labored over the text of Nonnus so that he could produce a metrical analysis of the poem. Chamberlain publishes his work under a Creative Commons license. Several years ago, I wrote a program to convert the HTML format that David has used into something that I found more convenient. This summer I asked Claude to write a program that would do the conversion — which it did in a few minutes of work. A handful of issues arose but I soon had what seems a plausible first version of the Chamberlain analysis.

Figure 2: Hypotactic metrical analysis of the Dionysiaca in the Perseus Multitext Viewer.

And, of course, we can provide more support than just adding a translation. Giuseppe Celano had published morpho-syntactic analyses (a Treebank) for each of the c. 125,000 words in the Dionysiaca. This provides the dictionary form for each inflected form in the text, as well as its syntactic role and dependency.

Ideally we will add alignments at the word and phrase level between the Greek source text and translation. When readers have both translation alignments and treebanks, they can begin to engage directly with source texts in unfamiliar languages. Automatic translation alignment, however, remains tricky (though we can do a much better job thanks to the Ugarit project). On the other hand, Large Language Models (LLMs) have become proficient at providing not only literal translations of Greek and Latin but also word-by-word contextual glosses. We used DeepSeek to augment Giuseppe Celano’s Opera Graeca Adnotata (OGA) treebank with translations and glosses to provide another layer of support for readers.

Figure 3: OGA Treebank (via Mark Bilby’s reformatting) for Nonnus with glosses and translation added by DeepSeek (deepseek-v4-pro).

Those working with large screens can open up to seven windows side-by-side, allowing them to compare different categories of data (or different translations or editions).

Figure 4: Four columns in the Perseus Multitext Viewer.

The automatic glosses are good but not perfect. Thus, we draw upon Helma Dik’s short definitions of Greek (based originally upon, but a massive improvement over, short definitions published by Perseus many years ago) to provide a short explanation for each word. While most users will exploit the glosses as a reading aid (as with traditional interlinear translations), the digital glosses can be searched and used for semantic analysis.

Figure 5: Words glossed by “anger” in the Homeric Corpus and in Nonnus.

Figure 5 shows those words in Homer and in Nonnus with glosses that contain “anger.” Obviously, we need to consider English synonyms (e.g., “anger,” “wrath”, “fury” etc.) but we can use BERT-like systems to locate synonyms and mitigate the problem. The figure above also includes some basic statistical information so that readers can assess the significance of the results. In the figure above, the ranges for the Iliad and Odyssey do not overlap. Thus, by the metric that we have chosen, we may be 95% confident that “anger” would be more frequent in the Iliad than the Odyssey if both works were infinitely large. Again, statistical analysis must be used critically but we need some to beyond comparing raw numbers when comparing search results.

Finally, one of the charms of the Dionysiaca is that it refers to places that we can locate on a map. Much of the Odyssey takes place in a never-never land that later Greeks eagerly linked to their own world, at least some of us would argue that it is important to imagine the world of gods and monsters through which Odysseus travels as being fundamental separate from the world in which humans live. The authors of later epics wanted to connect myths to their world and often made sure to locate their events in the geographic models of their time. ToposText had acquired the translation so that it could link the place names to their gazetteer and then plot them on a map.

ToposTexts makes their data as well as their texts available under an open license. We were thus able to provide a map that illustrate the places mentioned in any book or passage of the Dionysiaca.

Figure 6: places referenced in book 10 of the Dionysiaca with data published by ToposTexts in the Perseus Multitext Viewer.

A quick look at the fine print above will show yet another layer of open Digital Classics: the basemaps for the Greco-Roman world are from the Consortium of Ancient World Mappers.

While LLMs pose a range of challenges (including bot swarms that have crippled Perseus 4), it is important to emphasize that I would not have been able to create the integrated reader above this summer if Large Language Models had not advanced so quickly. I use Gemini (to some extent) and Claude (primarily) to design the data-structures and to generate the front end. The reader currently runs on Google Pages and can be moved to Tufts for long term support. (The source code is here and the data for the reader is here). LLMs also allowed me to convert the ToposText HTML into Perseus compatible XML and the Hypotactic HTML into a tab-delimited format. LLMs also allowed me to fashion a viewer for the ToposText geospatial data and the CAWM maps.

Conclusion: In the early 21st century, Digital Classicists, with the papyrologists playing a leading role, adopted the practice of publishing their data under open licenses. Decades later we can see the benefits compounding, as data circulates from one project to another, acquiring new features as it moves. We can create effective new reading environments even for less read authors such as Nonnus, allowing readers to engage directly with sources in ways that were not feasible in print culture.

by Gregory Crane Posted on August 31, 2026

This entry was posted in Uncategorized. Bookmark the permalink.