Prince of Biscari Network

Context

In 2021, the “Archives and Big Data” research fellow project focused on digitizing the “Correspondence” section of the Biscari Archive (Paternò Castello family) located at the State Archive of Catania (Italy). I created a database and compiled the digital edition of the missives in order to support the PNRR (National Recovery and Resilience Plan) philosophy, and contribute to the dissemination of the Italian archival heritage, through a website.

Inside the extensive archive are 2,000 folders composed of hundreds of thousands of documents (legal disputes, political decisions, trade records, and personal letters). The “Correspondence” section consists of more than 42,493 sheets grouped into 84 archival units, representing a wide range of dates from the second half of the seventeenth century through to the first half of the twentieth century. Within this section, there are different types of epistles, ordered by the sender and chronologically, followed by cards arranged according to the original alphabetical order and mainly relevant to administrative affairs. The section closes with other correspondence, organized by criteria established by the “Regolamento di Servizio pella Segreteria dell’Amministrazione”, introduced in 1845 by Roberto (eighth prince of Biscari) and his sister Marianna (Calabrese, 2003).

In folder 1642, which was chosen for creating the Biscari Epistolography digital edition and the website, there are 366 epistles and a manuscript by Emile Rousseau (a total of 591 papers), covering a period from 1680 to 1844.

The documents have been documented in several papers published in the scientific journals «Umanistica Digitale» (Spina, 2023a) and «Aidainformazioni» (Spina, 2023b). The essays highlight contemporary methodologies available to historians, such as artificial intelligence tools like Transkribus and ChatGPT, as well as some tools in Computational Linguistics, such as Keyphrase Digger.

Steps – The aim

Digitization necessitates more complex attention and effort when it comes to encoding archival documentation. This is particularly challenging when the archival material primarily comprises manuscripts, which pose a genuine obstacle to the application of computational analysis tools. When discussing “digitization”, it is important to consider encoding processes that create a machine-readable text, not simply a photographic acquisition (Spina 2022a). It is also important to consider the difficulties historians face when manually transcribe thousands and perhaps more archival documents. Fortunately, thanks to the development of two artificial intelligence tools, Transkribus and ChatGPT, scholars can move beyond mere photographic acquisition of a document, a serious limitation in the field (Adamek, O’Connor, and Smeaton, 2007; Cheriet et al., 2009; Archives, 2018; Deng and Lin, 2022; Kasneci et al., 2023; Fostikov, 2023).

Steps – Digitization workflow

I captured the document photographs with a Nikon D610, equipped with a AF-S Nikkor 24–120 mm f/4G ED VR lens. To address the challenges associated with this photographic equipment, as highlighted by the Federal Agencies Digital Guidelines Initiative (FADGI), the photographs were captured utilizing the following parameters: (1) shutter speed of 1/6s, facilitating an extended exposure to natural light; (2) aperture set at F/22, accompanied by a dynamic focal point area of 39 points for optimal clarity. The ambient lighting was counterbalanced with warm light at 4200 K to achieve a neutral white balance and permit the camera’s CPU to regulate the white balance.

Subsequently, the various shots were duplicated for further processing using Adobe Lightroom software to enhance the contrast of black tones and to prepare the files for flipbook processing.

Steps – The database and the automatic transcription

All photos were collected in a Filemaker 19 database, which afforded entry of metadata and indexing of all entities (sender, consignee, places, date) into a relational structure.

After meta-dating, each letter was merged and exported in PDF format and then uploaded to the Transkribus READ servers for automatic transcription (Kahle et al., 2017; Muehlberger et al., 2019; Milioni, 2020). This application compiled the digital edition and created PDF files that let me create the “flipbook” format for website browsing, and the XML-TEI files for computational and linguistic analysis.

Steps – ChatGPT

To correct the transcriptions, and to compile the records of the digital epistolography, I elected to test the LLM (GPT 3.5), whose internal structure and the training support entity recognition (González-Gallardo et al., 2023; Spina, 2023a). GPT-3.5 facilitated analysis of all the letters, extraction of various entities (names, dates, places, and events), and compilation of the searchable digital database on the website.

Steps – The website

The https://biscariepistolography.altervista.org website was designed to meet the expectations of the historical community, who demand increased internet accessibility to archival material for their research, allowing them to navigate through digital documents. For this reason, each epistle has been exported in PDF format and converted into a “flipbook” format to enable a close-reading virtual experience. A flipbook is an interactive digital text based on PDF files that facilitates internal searching and enhances the reading experience with multimedia elements, thus expanding the reading experience into hypertextual design dimensions.

On the other hand, XML-TEI files have been made available (open access) to all scholars who would like to go beyond simple reading and wish to analyze the documents using algorithms and tools in Computational Linguistics. For instance, Keyphrase Digger (Moretti, Sprugnoli, and Tonelli, 2015) has been utilized to extract key concepts from all the epistles, aiming to identify valuable information for reconstructing events experienced by members of the Paternò Castello family.