Success Story
Published: 8 months ago

Uncovering the secrets of the Portuguese Inquisition with Hervé Baudry

“According to the inquisitorial rules in the past, everything had to remain secret.” To punish people accused of straying from the Christian belief, the Portuguese Inquisition conducted trials clouded in secrecy and censorship. However, these trials were recorded in court files that are currently stored in the National Archives of Portugal (ANTT), waiting to be transcribed and analysed. Today, thanks to the hard work of the TraPrInq project, long-kept secrets of the Portuguese Inquisition are gradually being revealed. 

In order to uncover the contents of inquisitorial trial records, the TraPrInq team set out to use an AI-powered text recognition model to transcribe these documents dating from 1536 to 1821. Hervé Baudry, senior researcher at the CHAM-Centre for the Humanities of the New University of Lisbon and coordinator of the TraPrInq project, explained to us how it was possible to transcribe the inquisitorial court trial records with the help of Transkribus.

Official Website of the TraPrInq Project. © TranPrInq

The TraPrInq Project

300 years of archival output and more than 40,000 files from these court trials of the Portuguese Inquisition are held by the National Archives of Portugal, waiting to be transcribed and analysed. Recognising the potential of the cour trial records, ‘Processos’, the TraPrInq project was created. The unique name ‘TraPrInq’, a nonce word, sums up its mission: “Transcrever os Processos da Inquisição Portuguesa, 1536-1821”, which translates as “Transcribing the court trials of the Portuguese Inquisition, 1536-1821”. 

However, there is a significant challenge in trying to unlock the history hidden within thousands of pages of documents – accessibility.  Not everyone can access these historical files or read centuries-old manuscripts. 

Given his previous knowledge of Transkribus, Hervé Baudry recognised this challenge and thought of a solution: “In 2020, a webinar for Brazilian undergraduates on Inquisitorial Studies led me to conceive a project that could help to overcome the difficulties faced by the non-paleographers to read and consult the archives.” This led to the creation of the TraPrInq project, consisting of a team of eleven paleographers from Portugal and Brazil, with the primary goal of creating a versatile text recognition model using Transkribus and making it available to the public. 

Trial of Manuel Gois, Court Trials of the Portuguese Inquisition,1586. © ANTT

The court trials of the Portuguese Inquisition

Making these ‘Processos’ accessible for research and to the public would present a valuable resource, Hervé Baudry explains. “In the Portuguese culture, as in the Spanish and Italian ones, early modern censorship is strongly linked to the Inquisition.” The close link between early modern censorship and the Inquisition in Portuguese, Spanish and Italian cultures can be attributed to the Inquisition’s dual role as a religious and political authority. Identifying and suppressing religious deviations from Christianity within the country, the Portuguese Inquisition put people on trial to imprison them on the accusation of heresy or witchcraft. When censorship became part of inquisitorial activities, everyone who had contact with the Inquisition, even inmates who were released due to overcrowded prisons had to sign an oath of secrecy. 

Apart from the general historical value of discovering information about the past, revealing the contents of documents treated with secrecy can lead to new discoveries and insights. Determined to do so, the TraPrInq team came together to unlock the trials of the Portuguese Inquisition using Transkribus.

Trial of Isaac Almosnino, Court Trials of the Portuguese Inquisition, 1618. © ANTT

Transcribing centuries-old documents with Transkribus

The material used to train the Transkribus text recognition model consisted of handwritten documents of court trials from the Portuguese Inquisition from the years 1536 to 1821. From the existing digital database of the National Archives of Portugal, TraPrInq extracted over 6,000 images and transcribed over a million words for the creation of Ground Truth

To create the Ground Truth and train an efficient customised model, firstly a reasonable number of texts had to be transcribed and secondly, texts from different hands had to be allocated among the project members. Hervé Baudry explains further that “there was an imperative for quantity and quality, renewed at each session following training on the model.” Over the course of nine training sessions, starting in June 2022 and ending in July 2023, the team transcribed the entire corpus of 6000 digitised documents. After the fourth training session of the Portuguese text recognition model, the results started to become more efficient, to the point where “manual transcribing could be gradually substituted by only correcting the errors made by the [text recognition] engine”.

There was still a struggle when transcribing damaged documents or records with complex layouts. “Such images have a CER above 15-20%”, states Baudry. 

While there is no way to undo damage to historical documents, Transkribus has already improved layout recognition with new trainable layout models. With the Field Models (already available on beta.transkribus.eu) and Table Models (coming soon) Transkribus models can be trained to recognise and transcribe tricky layouts such as periodicals, registries, legal records or forms automatically. 

Overcoming the challenges, the final Character Error Rate (CER) after training was 5.2%, making the accuracy of the Transkribus text recognition model as high as 94.8%.

Public Transkribus text recognition model: Portuguese Handwriting 16th-19th century. 

A successful outcome of the TraprInq project

The goal of the 18-month-long TraPrInq project was to create a public text recognition model that is available to anyone. And indeed, in the summer of 2023, the new model for Portuguese Handwriting from the 16th-19th century was made publicly available for every Transkribus user.

Find the Model here: Portuguese Handwriting 16th-19th century

Baudry is satisfied with the outcome of the project: “Members of the team and people from other projects who contacted ours used the model with high efficacy”. Hervé Baudry adds that, more significantly, the texts automatically recognised by this new Portuguese model were not part of the TraPrInq project and came from non-inquisitorial sources. This means that the model is effective with different texts and document types. These results were presented at the final meeting of the project, Humanidades Digitais e Estudos Inquisitoriais (Conference on Digital Humanities and Inquisitorial Studies) in June 2023.

Furthermore, Baudry states that the project served as an initial step toward systematic document transcriptions, to make the digitised and transcribed documents available on Transkribus Sites (former Read&Search).

Reflecting on the project, Hervé Baudry expresses his admiration for all the participants in the TraPrInq project who, despite the uncertainty of the outcome, “took the plunge anyway, just by accepting my invitation to participate!”. And together they have succeeded in achieving their goal. He would also like to express his gratitude to the late Dirk Alvermann

who, at a crucial time for TraPrInq, put all his skills and sympathy at the service of the project.

Unlocking opportunities for future projects

Handwritten Text Recognition (HTR) technology is not new to the archival community, and while there is a great deal of interest, there are not always the means to use it in such a capacity.  “TraPrInq was a first step toward something massive and ambitious”, Hervé Baudry observes. Until now, only a fraction of the entire 3.4 million pages of the Inquisitorial Archives have been studied, and only a small part of the 40,000 files from the National Archives of Portugal have been transcribed and edited.

Baudry explains that this Transkribus model was created as a tool to make inquisitorial archives more accessible, and that hopefully one day all records of the Portuguese Inquisition will be transcribed and made legible for all.

Transkribus offers the potential to extract information and knowledge from historical documents and unlock the written past. With the ability to train custom models using the AI-powered transcription software, projects such as TraPrInq have the potential to transcribe historical documents and make them available for further research. 

When it came to the court trials of the Portuguese Inquisition, Hervé Baudry concluded that Transkribus’ HTR software “was able to break a double secret, one historical, the other, paleographical.”

Hervé Baudry’s Transkribus Tip:

“Transkribus may look a bit complicated to learn, but in a few hours you can use it, and it is worth it.”

Thumbnail: Portrait photo of Hervé Baudry. 

Overview