Tuesday, October 25, 2016

RIC 9 published to OCRE

RIC volume 9 has been published to Online Coinage of the Roman Empire. This represents about 1,700 types and 3,200 subtypes. In total, there are now more than 43,000 Roman Imperial coin types in OCRE, spread over half a millennium from Augustus to Zeno. This was a huge undertaking with many collaborators from the ANS and DAI, as well as contributors of data from more than a dozen American and European cultural heritage institutions. Without generous support from the National Endowment for the Humanities, we may never have completed this project, which will officially come to a close in December. Since publishing the types to OCRE yesterday, I have begun the process of harvesting relevant coins from partner institutions. The British Museum alone has contributed an additional 11,600 RIC 9 coins to OCRE, and the total number of physical specimens linked into the project stands around 93,000. We hope to surpass 100,000 when the ANS and Fitzwilliam Museum coins are added soon.

While some more work remains in tying up loose ends regarding meeting every specification of the NEH grant (with respect to correcting reference numbers in our curatorial database and completing photographic coverage of our Imperial coinage), we are nearing the final phase of the project, which will draw to a close by the end of the year. Nevertheless, the project will continue to evolve in a variety of ways. We anticipate aggregating content from more partners, especially from the archaeological community. There are more than 200,000 Roman Imperial coins in the Portable Antiquities Scheme, but so far barely over 300 have been linked to OCRE URIs. I am continuing to build more sophisticated analysis and visualization interfaces. These advancements have been implemented directly in Nomisma.org, but I anticipate porting these code updates into OCRE and various other Numishare-based coin type projects. We also plan to unveil two new features by the end of this year: an intuitive coin type identification interface that non-specialists (collectors or archaeologists working in the field) might use to identity coins, and a faceted search function for architecture depicted on Roman coinage (which extends into Republican coins in CRRO).

While the NEH funding was instrumental in the development of OCRE specifically, the open source code and the workflows we developed for this project have had an impact on our ability to publish similar online type corpora. In 2015, we saw the release of Coinage of the Roman Republic Online and PELLA. Since the multilingual and visualization functionality are inherent to Numishare, our other projects benefit from the funding the NEH invested directly into OCRE. One of these, obviously, is the Egyptian National Library collection of Islamic coinage, which is available in both English and Arabic.

Thursday, October 20, 2016

3,500 Fitzwilliam Museum (Cambridge) coins harvested into Nomisma

Using a combination of the University of Cambridge Fitzwilliam Museum's API to query for Roman Republican and Alexander the Great coins and PHP-based screen-scraping of reference numbers and measurement data, I was able to harvest more than 3,500 coins from the Fitzwilliam the other day. The script, which I have published to Github, took a few hours to write and another hour or so to fully execute.

The script queries the API and iterates through each page of the JSON response. Since coin type reference numbers are not stored or indexed into their ElasticSearch application, the script must request each HTML page from their public-facing database. With some XPath to parse the HTML and regex to look for RRC or Price numbers, matching numbers are checked against CRRO and PELLA to confirm the validity of the reference.

The metadata for each record are stored in an array, which is serialized into Nomisma-compliant RDF and a CSV concordance list when the process completes.

I hope to tweak the script and re-apply to Cambridge's Roman imperial coins. This will be a tremendous enhancement to OCRE, as we look to crossing over the 100,000 coin threshold in that project alone.

Tuesday, October 18, 2016

Distribution Visualization Tools in Nomisma.org

I have spent the last month working on and off on a new set of research tools in Nomisma.org's interface. As users of individual ANS type corpus projects (like OCRE) will know, we have exposed simple weight and diameter analyses and the distribution of typological categories (like denomination, mint, authority, etc.) as graphs through a visualization interface. These interfaces query only those coin types or physical specimens connected with the type corpus as a whole. Plainly, this means that the interfaces in OCRE only query coin types from RIC or display the average weights of coins that have explicitly been connected to URIs published by OCRE. Partners with find or excavation databases that contain uncertain coins are excluded from these queries, even if the portrait or mint can be positively identified, but the wear of the legend is too great to allow for a certain RIC attribution.

I have begun to implement distribution visualization interfaces directly in Nomisma that query its SPARQL endpoint. At present, the distribution analyses are based on coin types. For example, you can generate a chart that shows the distribution of denominations issued by Augustus. You can place further filters on this query, for example, to show the distribution of denominations issued by Augustus at the mint of Rome. The interface allows you to gather multiple datasets for comparison, e.g., to include the distribution of denominations for Nero and Hadrian. These queries factor in all coin types, regardless of the scope of the type series, enabling you to query across all coins of the Roman Republic and Empire simultaneously.

Distribution of denominations of Augustus, Nero, and Hadrian from the mint of Rome

Some work remains in enhancing the complexity and scope of these queries. It is important to introduce start and end date filters so that it would be easier to evaluate changes in distribution over time. Furthermore, the categories for visualization (authority, issuer, mint, region, denomination, material, object type) need to be expanded to include portraits and depicted deities. This adds a layer of complexity in the query--these objects are referenced in the obverse and reverse of the coin type, not directly in the coin type data object itself. Presently, deities are an underutilized query parameter across all our projects. We use British Museum URIs, but have not integrated SKOS PrefLabels or any other properties about these URIs into the Nomisma SPARQL endpoint that would allow us to build a human-readable interface. Adding supplementary, non-numismatic datasets is on the agenda for Nomisma development.

The distribution analysis interface is available on the pages for individual Nomisma IDs (if these IDs have associated coin types) and on a separate interface. The distribution interface renders the chart by passing URL parameters to itself, so the page URL can be copied and pasted, and the charts can be shared with others. The charts on the Nomisma ID page are generated directly with ajax calls (so it is easier to view changes in the graph when modifying underlying queries), but you can click a button load the chart in the distribution page. There is an option to download the data as CSV in both interfaces.

Next: Metrical Analysis

Once I implement portraits, deities, and date filters, I will get to work on interfaces that will analyze weights and diameters. This interface will open the door to integrating physical coins that have been connected to typologies implicitly through OCRE, CRRO, and PELLA URIs or explicitly, enabling us to compare all coins of an emperor even when the specific coin type attribution cannot be ascertained.

These interfaces will not only show the average weights of typological categories, but changes in weight over time (e.g., to view the change in weight of the denarius over 400 years). I'll also implement standard deviation visualizations to cluster coins together of nearly identical weights, making it easier to spot outliers that might be fragmentary coins or even forgeries.

Thursday, September 1, 2016

Remainder of RIC 5 Published to OCRE

Nearly 4,000 more types from the latter portion of RIC 5 have been published to Online Coins of the Roman Empire, predominately from the Gallic Empire and usurper sections. With RIC 5 completed, this brings to total number of Roman Imperial coin types to nearly 42,000. David Wigg-Wolf will finish checking the RIC 9 spreadsheet very soon. We expect to complete the project by the end of September (months ahead of schedule).

In many ways we have already exceeded the specifications of the NEH grant-funded project. The coins from the University of Virginia and Berlin have recently or are currently being processed to link to these new types. The ANS coins will become available in OCRE in the near future (as early as tomorrow). By the time RIC 9 is published and our partner collections have been reprocessed to add in physical representations of these coin types, OCRE will be very close to exceeding 100,000 coins from museum and archaeological databases. The number of partners and coins has increased in the last year, and we expect this level of growth to continue for the foreseeable future, especially with some of the major numismatic collections coming online (e.g., Vienna and the Bibliotheque nationale de France).

Since we are poised to finish the original parameters of the project three months early, we are going to work on adding value to our data by linking architecture displayed on coins to either broad architectural typologies (e.g., temple, fountain) or specific, known monuments (Temple of Mars Ultor). I think that this will open the door to more sophisticated query and analysis of Roman architecture and its relation with official imperial political messages to society (both in Rome and the provinces).

We are also going to work on putting together a user interface that will make it easier to identify coins, a feature that will be useful to casual collectors, museum curators, and archaeologists working in the field.

Thursday, July 28, 2016

Nomisma.org crosses 100,000 coin threshold

Following the publication of RIC 6, 7, 8, and 10 to Online Coins of the Roman Empire, I extracted over 18,000 coins with references to these volumes from the British Museum SPARQL endpoint (with this query) and successfully matched and imported about 17,000 of these into Nomisma's SPARQL endpoint. There are now about 80,000 physical coins linked to the 38,000 Roman imperial coin types in OCRE. I also reprocessed the Berlin LIDO export and published 1,000+ coins from late Roman coinage into OCRE.

This recent import has brought the total number of physical specimens linked to online type corpora (including Coinage of the Roman Republic Online and PELLA) to 116,964, about 98,000 of which come from the British Museum and American Numismatic Society alone. The number of coins has more than doubled in a year.

Technical Process

The XML response to the SPARQL query linked above was processed through a PHP script I wrote several days ago. The script iterates through each result in the XML document in order to parse the reference text with regular expressions in order to generate a type ID that conforms to the OCRE convention (e.g., ric.7.anch.109) and test to see whether the URI exists. The result of the concordance process is written as a CSV file, which is then processed by another PHP script into RDF conforming to the Nomisma.org ontology. An additional SPARQL query is executed on the British Museum endpoint for each row in the table in order to extract the weight, diameter, image, etc. The RDF is written to disk and then imported into the Nomisma SPARQL endpoint, where the data are immediately available in OCRE.

Friday, July 22, 2016

ANS coins from RIC 6-10 published to OCRE, and other updates

Following the release of volumes 6, 7, 8, and 10 to OCRE, we have republished our coins from these volumes to link them into the newly-published coin type URIs. This represents an addition of more than 17,000 physical specimens of late Roman coinage into OCRE, including photographs for more than 3,000 of these (and photographic gaps from previous volumes of RIC). There are now 36,000 Roman imperial coins from the ANS collection in OCRE, and 60,000 in total from all our partners. Including CRRO and PELLA, there are just under 100,000 physical coins aggregated by Nomisma.org's SPARQL endpoint.

In addition to these coins, the Portable Antiquities Scheme provided access to several hundred imperial coins linked to OCRE URIs. The PAS had previously linked its entire collection of Republican coins (nearly 1,000) into CRRO, but the inclusion of imperial material in OCRE is a watershed moment for the study of Roman numismatics. These are the first few hundred of potentially hundreds of thousands of coins published in their database, each with attested findspots. This will have a dramatic effect on geographic analysis of ancient monetary circulation and trade.

The Harvard Art Museums API was also reprocessed. Harvard's coverage of late Roman coinage is quite good, and their contribution to OCRE has more than doubled to 1,300 coins.

Update (July 25): We have also added 174 coins published by OpenContext.org from the Domuztepe excavations into the Nomisma triplestore to make them available in OCRE. These are late Roman coins, predominately from RIC 7 and 8.

Monday, June 27, 2016

RIC 6, 7, 8, and 10 published to OCRE

After filling in gaps in Nomisma IDs necessary for the publication of RIC 10 (including the extension of Numishare to support the creation and publication of monogram URIs with images), we have pushed RIC 6, 7, 8, and 10 into Online Coins of the Roman Empire (OCRE). The symbol publication extension of OCRE can be seen at http://numismatics.org/ocre/symbols. Currently, each image is a PNG file, but will be replaced with SVG soon. We have developed a workflow to covert True Type Fonts representing Roman imperial numismatic symbols provided by Slovenian partners to SVG with a combination of the open source Linux tools, FontForge and Inkscape.

These four volumes represent a huge number of types, increases OCRE's types in excess of 13,000 up to a total that now exceeds 38,000. David Wigg-Wolf is working on the RIC 9 spreadsheet, and we hope to publish both that and the remainder of RIC 5 by the end of the year.

The next task for this OCRE update is to process several thousand images through our workflow and publish the coins in the ANS curatorial database that link to OCRE IDs into Mantis so that they will become available in OCRE via the Nomisma.org SPARQL endpoint.