First letters found in new scroll

PHerc. 172 (Scroll 5) is here.

PHerc. 172 (Scroll 5)

Today we are thrilled to share a new scroll dataset that is already full of exciting findings. In collaboration with our partners at the Bodleian Library, the University of Oxford, Diamond Light Source, and EduceLab, we present: PHerc. 172!

This scroll has a remarkable past, present, and future (which is where you come in).

Past

As with the other Herculaneum scrolls, the story of PHerc. 172 begins two thousand years ago, when it was written before being buried by the eruption of Mount Vesuvius in 79 AD. But we’re going to skip ahead to modernity, after the scroll was discovered and unearthed.

In 1802 or 1803, PHerc. 172 and five other scrolls were given by the king of Naples and Sicily, Ferdinand IV, to the future George IV in England - allegedly in exchange for some kangaroos. Yes, really!¹ Stay tuned for the rest of the story, as we couldn’t fit it all today with the other exciting info to share.

Present

The scroll is currently kept at the Bodleian Library in Oxford, England. We imaged the scroll earlier this year with our collaborators at Diamond Light Source, a synchrotron facility conveniently located nearby. Similar to previous scans, this one was acquired with a voxel size of 7.91 μm and an incident energy of 53 keV. The scroll is large enough that it had to be scanned in two halves, but they’ve been merged into a single, complete volume.

Stephen at the synchrotron, a CT slice, Julian debriefing the scan with the Bodleian Library team, and PHerc. 172 in its case.

Ink appears

While preparing for the release, our segmentation team created some initial segments to study the dataset. Immediately, we were struck by the clear appearance of ink, directly in the data!

Greek text visible in texture image generated from CT data, even before machine learning-based ink detection. Taken from segment 20241108120732.

This is the first time the ink of a Herculaneum scroll has presented with such clear contrast in X-ray CT. This suggests a higher density ink, but we don’t yet know its exact composition. Research using Herculaneum scroll fragments has shown² that the inks of different scrolls can have slightly different chemical signatures, even though they are all carbon-based. That could be a factor, as could differences in the papyrus, or in the carbonization process of this scroll. We’re investigating!

There are more questions already forming: why does the ink seem to appear more clearly in the outer wraps of the scroll? Why, in some cases, is it only clear every other line? Did the scribe dip their pen once every two lines?

Greek text

In the meantime, the scan data is already producing great findings. Our team of papyrologists immediately confirmed the text is Greek. Using the raw texture images as well as early ink prediction results³, they are even able to produce preliminary transcriptions in which we are very close to recovering complete words!

Top: unrolled automatic segmentation attempt from ThaumatoAnakalyptor outlining scroll structure. Left: scroll photograph. Center and right: Ink detection images and preliminary transcriptions for segments 445437_495437 and 20241108120732 (roughly to scale and positioned in scroll height).

These letter annotations are conservative, shown only where our team of independent reviewers had the same findings, and could change as ink detection improves. As can be seen, there is yet more text tantalizingly close to legibility. We expect to recover this text as our techniques are refined.

We’re already on the cusp of recovering complete words. At the bottom of segment 445437_495437, there might be the beginning of διατροπή, a word found in other Herculaneum papyri that would mean something like “confusion, agitation, or disgust.” Similarly, in segment 20241108120732, the sequence τυγχαν may be the beginning of the verb τυγχάνω: “to happen,” or perhaps “not to happen.”

Future

One thing we know will happen: we’re going to recover the text from this scroll. But we’ll need your help to do it.

This scroll is eligible for inclusion in a 2024 Grand Prize submission! It is also eligible (and encouraged) for use in the development of a Monthly Progress Prize submission. Of all Herculaneum scrolls scanned to date, this one is of a most favorable prize-winning temperament!

The dataset for PHerc. 172 is now available on our data server under the CC BY-NC license. As this is the fifth intact scroll released via Vesuvius Challenge, it also goes by the name Scroll 5.

To get started, you can quickly view the dataset in your browser:

Viewing Scroll 5 in Neuroglancer.

Next, you can take a look at the files served on the data server. As with the other scrolls, there is a Volume Cartographer-style volume (a folder of .tif image slices)⁴. There is also a chunked Zarr volume available. Or, to play with the data in code without downloading the entire dataset, you can use our Python or C libraries.

There are initial segments produced by the segmentation team, which are high quality segmentations closely following the papyrus surface.

There is also the output of ThaumatoAnakalyptor, an autosegmentation pipeline. The autosegmentation jumps frequently between adjacent sheets, so is not yet precise enough to reveal contiguous texts, but it coarsely follows the entire scroll. The ink prediction images of these segments are therefore great guides to suggest regions of interest for manual segmentation:

Initial autosegmentation output: room for improvement, but much structure and text is evident (large image).

For any of these segmentations, it may be useful to play with the ink visibility by adjusting the number of surface volume layers used in a composite image. There are various ways to do this: general purpose tools like ImageJ or Photoshop, or the specialized Crackle Viewer.

Particularly in the early stages of discovery for a new scroll, small technical improvements have outsized returns on the legibility of the papyrus. A few more segments with clear text, improved segmentation on existing segments, larger and more accurate labels for ink detection - these contributions can provide papyrologists with the clues they need to unlock the next letters.

If you are as excited as we are about noninvasively recovering ancient texts from carbonized scrolls, we think you will love this dataset. A human hand wrote this text 2,000 years ago, and it’s an unforgettable experience to be the first person in that time to see it. Especially with the unusually visible ink, this scroll offers that experience to many of us. We can’t wait to see what you find!

Thank you

It takes a village to scan and read an ancient scroll. Thank you to everyone who has helped make this happen, in particular to Richard, Martin, Tobias, Peter, Nicole, Daniel, Seth, Christy, Brent, Leigh, Robert, Elizabeth, Ben, David, Julian, Giorgio, Youssef, Luke, Francesco, Federica, Alessia, Claudio, Kilian, Marzia, Michael, Rossella, our generous donors, and our incredible community.

1. Carlo Knight. “Canguri e papyri.” Cronache Ercolanesi, 32 (2002): 305-320.

2. Bonnerot, Olivier, et al. "XRF ink analysis of some Herculaneum papyri." Zeitschrift für Papyrologie und Epigraphik (2020): 50-52.

3. Stay tuned for more ink detection information from Youssef!

4. When viewing a Scroll 5 .tif image, the slice number decreases from the screen toward the viewer’s eye. The direction of a given line of writing is counterclockwise around the umbilicus, with the bottom of the letters on lower-numbered images, and the top of the letters on higher-numbered images.